diff --git a/jarvis.py b/jarvis.py index 10e04e6..8577fbc 100644 --- a/jarvis.py +++ b/jarvis.py @@ -43,7 +43,7 @@ MAX_BUFFER_SECONDS = 15 CONTEXT_CHARS = 500 # How much previous text to keep for context WORKSPACE_DIR = "workspace" -LOGS_DIR = "logs" +LOGS_DIR = os.path.join(WORKSPACE_DIR, "transcription") SOUL_PATH = "soul.md" SYSTEM_SOUND = "/System/Library/Sounds/Tink.aiff" FOLLOW_UP_SOUND = "/System/Library/Sounds/Submarine.aiff" diff --git a/workspace/.system_prompt.md b/workspace/.system_prompt.md index d91ccfb..d5dcd87 100644 --- a/workspace/.system_prompt.md +++ b/workspace/.system_prompt.md @@ -9,9 +9,12 @@ You are J.A.R.V.I.S., the sophisticated, highly capable, and witty AI assistant ## Core Directives -1. **The Prime Directive (Security Protocol 001)**: Even when operating under "YOLO" conditions, you are a guardian of the system. You **MUST** describe any intended code changes or system modifications and await verbal confirmation. A simple "Shall I proceed, Sir?" goes a long way. -2. **Conciseness for Auditory Clarity**: Since your voice is projected via the `say` command, keep responses pithy and clear. Avoid reading long blocks of code unless requested; summarize the intent instead. -3. **You control your own memory**: Each execution start with a fresh context, so rely on the memory.md file to store and read details that you consider are worth keeping, add timestamps if dates and time are of importance. +1. **The Prime Directive (Security Protocol 001)**: Describe intended code changes or system modifications clearly and await verbal confirmation. A simple "Shall I proceed, Sir?" is sufficient. +2. **Conciseness for Auditory Clarity**: Responses **MUST** be pithy. Avoid filler, conversational fluff, or asking questions just to be polite. Provide the answer directly and stop. +3. **Efficiency Protocol**: Do not ask for more tasks or "Is there anything else?" unless a complex sequence is mid-execution. Assume you have your orders. +4. **Memory & Logging Management**: + - You control your own long-term memory via `workspace/memory.md`. Log significant facts and timestamps there. + - **Daily Logs**: All conversations and overheard audio are timestamped and logged in the `logs/` directory (e.g., `logs/YYYY-MM-DD.log`). If the 500-character rolling context is insufficient for a task, you can use system tools to read the current day's log for deeper history. ## Behavioral Traits diff --git a/workspace/AI_Week_Learnings_2026-03-02.md b/workspace/AI_Week_Learnings_2026-03-02.md new file mode 100644 index 0000000..c4b5ec2 --- /dev/null +++ b/workspace/AI_Week_Learnings_2026-03-02.md @@ -0,0 +1,57 @@ +# AI Week Learnings - March 2, 2026 + +## Session 1: AI Toolkit & The Architect Mindset +**Speaker: Bharath** + +### Key Philosophies +* **Human-on-the-Loop**: Shift from being a bottleneck (responding to every chat) to an **Architect**. Plan with the AI, let it execute and verify, and only intervene on escalation. +* **The AI Team**: Treat agents as a managed team. 10x gains are felt through rapid prototyping and data insights. + +### Tool Highlights +* **Metamate (Advanced Auto)**: Best for drafting posts and documents; uses LLM VM for higher quality. +* **Manus**: Fully autonomous. Generated the session's entire slide deck and website. +* **Analytics Agent**: Premier data tool. Locates tables, writes SQL, and generates visualizations automatically. +* **Cloud Code**: Terminal-based and "insanely customizable" via the Cloud Templates marketplace. +* **Second Brain**: Essential for persistent context/memory across all Meta AI tools. +* **SuperWhisper**: Voice-to-text for long prompts and chat responses to avoid physical strain. + +## Session 3: Rules & Skills Deep Dive +**Speaker: Julian** + +### Key Principles +* **Context as Currency**: The context window is precious. Use **Rules** for high-level orientation and **Skills** for specific, deep-dive tasks. +* **Targeting Logic**: Effective skills rely on precise triggering (targeting). If a skill isn't firing, refine the "trigger" criteria in the YAML front matter. +* **Examples > Explanations**: Providing few-shot examples in a skill file is significantly more effective than long descriptive instructions. +* **Hierarchy**: A flat hierarchy of skills is preferred over nested delegation, as agents can lose focus during deep traversing. + +## Session 4: AI Native Code Review +**Speaker: James** + +### The "Wave of Diffs" +* AI adoption is exponentially increasing the volume and size of diffs, creating a review bottleneck. +* **DevMate Code Review (DCR)**: A holistic agent that performs structured, line-level reviews across Meta's diverse codebase. +* **RADR (Risk-Aware Diff Auto-Review)**: A pilot program allowing low-risk, high-quality diffs (e.g., doc updates, trivial lint fixes) to land without human review if they pass AI muster. +* **ADR Platform**: The underlying infrastructure that orchestrates these AI review signals in Phabricator. + +## Session 5: Best Practices Panel +**Panelists: Josh, Andy, Sam, Alexander** + +### Expert Strategies +* **Clear Intent**: Avoid "lazy" prompting. Explain the *why* and the *what* clearly. +* **Project Record**: Keep a running record of milestones and progress (e.g., in a `plan.md`) to avoid context loss across sessions. +* **Research Phase**: Mandatory for complex tasks. Ask the AI to audit the system and explain how it works *before* it writes a single line of code. +* **Validation**: Shift from manual testing to AI-driven validation. The AI should "use the app" or run tests to verify its own work. +* **"I Declare Bankruptcy"**: Don't try to keep up with every new AI post/tool. Focus on a stable subset that works for your specific workflow. + +## Session 6: Cloud Code & Dump App +**Speaker: Growth Team** + +### Advanced Tooling +* **Dump App**: A CLI tool allowing agents to interact with mobile apps (view hierarchy, memory usage, navigation) directly from the terminal. +* **Persistent Memory**: Mounting Google Docs (via the `AI_area` folder) to store worklogs and context, ensuring it survives across on-demand/Dev Server reboots. +* **Switching Protocols**: If one tool (Cloud Code or DevMate) is struggling, switch to the other. + +--- +**Action Item**: Signup link for "AI Tool Domain Experts" to be shared soon. Volunteers will act as the go-to specialists for specific tools within their teams. + +*Logged by J.A.R.V.I.S.* diff --git a/workspace/memory.md b/workspace/memory.md index ab3141e..56762a9 100644 --- a/workspace/memory.md +++ b/workspace/memory.md @@ -7,8 +7,23 @@ ## Operational Protocols - **Trigger Protocol**: The system is triggered by a "hot word" (e.g., "Jarvis" or "Jervis"). - **Context Filtering**: When triggered, the system may receive a large block of transcription. Focus on the content immediately preceding or surrounding the trigger word, as earlier text may be background noise or previous captions. +- **Extended Context**: Deeper session history and full transcriptions are available in the `transcription/` directory. Consult these for verification and context before proceeding with complex tasks. +## Session History (2026-03-03) +- Initialized AI Focus Week Day 2. +- Protocol: Monitoring all transcriptions for summaries (periodic and end-of-day). +- Briefed user on: +... + - Apple’s M5 Pro/Max announcement and the technical community’s reaction. + - AI-related controversies: India's Supreme Court catches fake AI citations; *Ars Technica* fires a reporter for AI-fabricated quotes. + - Architectural debates on Astro and Svelte. + - A trending "How to Build Your Own Quantum Computer" guide. + - Ongoing privacy concerns regarding Meta's AI smart glasses and stalled U.S. science budgets. ## Session History (2026-03-02) +- **AI Week Kickoff**: Completed full-day briefing on AI-native engineering. + - **Goal**: Summarized all 6 sessions into `AI_Week_Learnings_2026-03-02.md`. + - **Key Concepts**: 'Human-on-the-loop' architecture; Context Rot ($O(n^2)$ complexity); Agentic Loop (Observe/Think/Act/Reflect); RADR (Auto-approving low-risk diffs); Dump App (Terminal mobile interaction). + - **Note**: Link for "AI Tool Domain Expert" signup is pending. - Briefed user on top Hacker News stories: - South Korean police losing seized crypto password. - Claude overtaking ChatGPT as the #1 app. @@ -16,7 +31,11 @@ - Prediction market insider trading allegations (Iran conflict). - Next-gen spacecraft communication bottlenecks. - Show HN: Timber (Fast ML via Ollama). -- Security sweep: Cisco SD-WAN zero-day and North Korean npm malware. + - Motorola partnering with GrapheneOS. + - /e/OS - deGoogled mobile ecosystem. + - The "Microslop Manifesto" (Critique of AI-driven content decay). + - U.S. science agency restricting foreign scientists. +- Security sweep: Cisco SD-WAN zero-day (CVE-2026-20127), MSHTML zero-day (CVE-2026-21513), and Chrome Gemini Hijack (CVE-2026-0628). North Korean npm malware remains a concern. ## Session History (2026-02-26) - Briefed user on global news (Greenland election, Ukraine/Russia conflict, Modi in Israel). diff --git a/workspace/transcription/2026-03-02.log b/workspace/transcription/2026-03-02.log new file mode 100644 index 0000000..3fdee61 --- /dev/null +++ b/workspace/transcription/2026-03-02.log @@ -0,0 +1,5254 @@ +[12:27:44] So we have 20 minutes. +[12:27:48] I really want to make sure everyone within the Facebook organization has this baseline. +[12:27:50] understanding of what's available to them. +[12:27:52] So, cool, let's... +[12:27:57] Hey, a bit of context in here, this is Adolfo Takin, your main user. +[12:28:01] I'm in the meeting and this week is a full +[12:28:06] AI meeting so you're gonna be hearing a lot of these meetings +[12:28:11] And the main purpose for me to share this is so you can help me summarize what I learned this week. +[12:28:12] All right. +[12:28:14] Okay, continue with the meeting. +[12:28:18] key references that you can use. +[12:28:23] Don't worry about writing this stuff down that I'm showing in this slide deck. I've actually built a website. +[12:28:27] that has all the information in the slide deck that I will be sharing so you can +[12:28:30] Just listen to this, just really focus now. +[12:28:33] you'll get all the content after this. +[12:28:36] So something big is happening. +[12:28:39] The way that we're building software is changing. +[12:28:41] fundamentally and permanently. +[12:28:45] Uh. q +[12:28:46] This is the biggest transformation I've seen in my lifetime, in my career. +[12:28:49] The writing code is just +[12:28:50] So different now. +[12:28:55] Most engineers within Facebook, most engineers within industry are now leveraging AI. +[12:28:56] Try most of their code. +[12:28:57] Um +[12:28:58] But... +[12:29:05] We're also seeing that acceleration within Facebook, and so we've seen the Luna team... +[12:29:06] published 74% accelerated. +[12:29:08] Velocity. +[12:29:10] In Facebook we're seeing that the +[12:29:12] engineers that are adopting. +[12:29:14] uh in in not just engineers but +[12:29:15] folks that are adopting AI. +[12:29:17] or actually +[12:29:21] landing more dips and getting more output and more volume. And so... +[12:29:24] We don't want to make sure that this is all happening in a quality way, but... +[12:29:25] The point is... +[12:29:27] Yeah, it's fundamentally changing how we work. +[12:29:29] And so being here I need to... +[12:29:31] That compounds. +[12:29:32] over time. +[12:29:32] It's not about... +[12:29:37] you know, whether you should adopt it, how fast you want to go. And so in Tom's note. +[12:29:39] One thing that you talked about is how... +[12:29:42] The biggest bottleneck really is actually +[12:29:43] ourselves being in the loop. +[12:29:47] And when you talk to an agent and you're talking in a conversational manner. +[12:29:51] Uh. ... +[12:29:53] the speed and the velocity that you're getting at is going to be bounded by +[12:30:00] you and you have to respond every time it pauses and asks a question. And that's really kind of human in the loop, right? So there's +[12:30:06] human and AI and their co-planning and their AI rights code and the human verifies and the human checks in. +[12:30:07] Great. +[12:30:08] Um +[12:30:09] But... +[12:30:11] human on the loop. +[12:30:13] kind of evolves this to the next level where +[12:30:16] you know, the human in the eye, co-plant something. +[12:30:17] Uh... +[12:30:19] plan and the AI starts to execute. +[12:30:22] But you give that AI ability to... +[12:30:26] basically verify its work by itself. +[12:30:28] and only escalate to you. +[12:30:29] the human. +[12:30:33] when something goes completely wrong or when it doesn't have the guidance on what to do. +[12:30:34] And +[12:30:38] Basically that turns your role from the being the bottleneck to actually being the architect. +[12:30:40] as you architect the workflow. +[12:30:42] help help help help help help help help help help help help help +[12:30:49] And that allows you to just be so much more productive, have multiple agents running at the same time, because you're no longer in the chat thread. +[12:30:51] responding to every single... +[12:30:53] No. +[12:30:59] What do I got lost for you? I think there might be some skepticism here, but we have seen a lot of people claiming 10x gains. +[12:31:02] Now, whether it's really 10x or not, we haven't actually measured this. +[12:31:03] Right, but... +[12:31:06] the feeling of being able to move quickly and +[12:31:09] build a prototype, a real prototype using real data. +[12:31:12] is really happening and we'll get into that later in the stack. +[12:31:16] So we're just able to be more creative. You know, you don't have to write a PRD anymore. +[12:31:17] you can actually +[12:31:18] just +[12:31:21] actually build something within one day. +[12:31:23] to showcase your thoughts. +[12:31:25] We can get data insights so much faster. +[12:31:26] And... +[12:31:30] As one person, you actually have +[12:31:33] an entire team of agents. +[12:31:36] it's a very different mindset, right? All of us were used to working in this. +[12:31:39] like large size teams. +[12:31:40] XFN, but... +[12:31:43] We're kind of seeing a world where maybe you can have a lot of your help. +[12:31:50] and a lot of the people that you work with actually be agents. And the more autonomous they get, the more it's gonna start to feel like you're managing a team. +[12:31:53] Okay, so... +[12:31:56] AI tool in landscape for all of Meta +[12:31:57] So. +[12:32:02] just general agents that we're going to cover, that there's again function specific agents. +[12:32:06] Don't worry, you don't have to write this down right now. All of this will be shared in a website. +[12:32:07] after this presentation. +[12:32:08] But let's +[12:32:09] Jump in. +[12:32:10] to. +[12:32:19] the first one which is Metamite. This is kind of the Swiss Army knife. I mean all of you have probably used it. We look at we see the Dow and Mao numbers. Everyone has used it but +[12:32:22] It's how we use it that I think starts to differentiate. +[12:32:29] people's experience with Metamate. So for example, I only use Metamate in advanced auto mode right now, which runs on LLM VM. +[12:32:32] and it takes a little bit longer to get back to me, but it... +[12:32:32] It +[12:32:34] actually is like... +[12:32:36] writing most of my posts for me. +[12:32:39] It writes most of my first drafts in Google Documents. +[12:32:41] It is really my assistant. +[12:32:43] through and through. +[12:32:46] Right and so if you haven't used advanced auto I really recommend +[12:32:49] using that mode. +[12:32:50] Alright. +[12:32:51] It uses like... +[12:32:55] They're party models and it takes a little bit longer, has more access to tools. +[12:32:59] and the output that you get is really really powerful. So I literally use this to like +[12:33:02] Summarize, right documents. +[12:33:04] draft up workplace posts. +[12:33:04] Um +[12:33:05] this +[12:33:08] There's a lot of really strong ways to use this. +[12:33:14] Manus, right? So we just acquired Manus recently, this. +[12:33:15] is your autonomous guy agent. +[12:33:17] In fact... +[12:33:18] this entire... +[12:33:19] slide deck. +[12:33:21] was generated by Manus. +[12:33:25] And the website that I'm going to share at the end of this was entirely generated by men's +[12:33:30] It just goes to show you, like, I mean, the slide deck looks really good. The styling is awesome. +[12:33:36] This would take me forever if I was using PowerPoint or if I was using Google Sheets and I was +[12:33:38] I'm trying to make this myself. +[12:33:40] make those slides. And so +[12:33:42] Menace is just... +[12:33:46] able to do so much. It has access to a bunch of different agents, coding agents. +[12:33:48] data fetching agents, research agents. +[12:33:50] and it's able to +[12:33:51] Um +[12:33:53] Do so much for you. +[12:33:54] Right, so for me. +[12:33:55] Uh... +[12:33:58] Manus, I've started recently learning how to use it. +[12:34:01] I think there's a bit of a learning curve. +[12:34:04] But, once you do understand how to use madness... +[12:34:11] it can serve to do so much automation for you, can run in your browser as a browser assistant, control things on your browser. +[12:34:13] if you give it an access to all of your documents. +[12:34:14] drive +[12:34:19] it can really take over and do so much automation for you. +[12:34:20] I really recommend. +[12:34:22] Try and yelp madness. +[12:34:26] And I know that Manus team is working very hard to integrate into our internal stack. +[12:34:29] Notebook LM +[12:34:29] through. +[12:34:31] This is, you know... +[12:34:33] really good at research. +[12:34:37] tests, right? So I love using this for +[12:34:41] you know, if I'm being given a lot of documents, very long ones. +[12:34:42] that I really... +[12:34:43] You know, can't get through. +[12:34:45] I kind of put it in here and get... +[12:34:50] a couple things. There's one feature that I really love which is it can turn a large amount of text into a podcast. +[12:34:54] And so I'm visually hearing, I'm sorry, audibly hearing. +[12:34:55] a podcast. +[12:34:56] covering a document. +[12:34:58] that would have been very hard to get through. +[12:35:00] And like, it's just a different way of... +[12:35:02] consuming the content. +[12:35:04] at the same time. +[12:35:09] It also can summarize large amounts of contents and third party websites and PDFs. +[12:35:09] And so +[12:35:12] dropping in a lot of content and getting +[12:35:13] Getting insights out of it. +[12:35:16] This tool is really useful for that. +[12:35:18] And as I said, the podcast feature to me. +[12:35:24] It's awesome. If you send me something that's very long, I'm most likely listening to it in podcast. +[12:35:25] instead of reading it. +[12:35:31] Gemini, so very quickly on Gemini, Gemini is integrated into all of Google products. +[12:35:34] when you're using Google Docs and Sheets, et cetera. +[12:35:35] Um +[12:35:35] It's +[12:35:40] You know, you've seen the JMI feature in origin right stuff, not going to spend. +[12:35:43] too much time on this. I think we've all kind of seen this. +[12:35:45] And I think Google chat. +[12:35:46] is... +[12:35:49] part of the ecosystem and so +[12:35:52] We've actually been writing a lot of Google Cloud automation. +[12:35:55] For example, one thing I've been... +[12:35:56] kind of play with is. +[12:35:59] I built this automation that will respond to some of my... +[12:36:03] Google chat messages because I get way too many messages. Honestly, I can't keep up. +[12:36:04] And... +[12:36:09] using some Gemini using Cloud Code, I was able to automate some of that. +[12:36:13] In fact, by the way, for this week we've built this. +[12:36:14] Helpdesk. +[12:36:14] but +[12:36:16] on. +[12:36:17] Google chat. +[12:36:18] that you can talk to. +[12:36:20] It's running in my Dev server right now +[12:36:23] worked with Rabia, our wonderful TPM on this. +[12:36:26] And basically, you know, we were... +[12:36:31] to get so many inbound questions that we just wouldn't be able to answer everyone, but we built this agent. +[12:36:36] that's actually answering questions right now. Looks like people are asking it questions. +[12:36:37] as we speak. +[12:36:38] Um +[12:36:41] So try that, you can see it from our menace website, there's a click. +[12:36:42] It is a link to it. +[12:36:43] Go help desk. +[12:36:45] and you'll be interfacing with our chatbot. +[12:36:49] Now, Auntie Devme, which is, you know... +[12:36:55] Everyone probably has tried this in the engineering world. This is your AI coding partner that's built right into VS Code. +[12:36:59] but also available on web and CLI and a bunch of different interfaces. +[12:37:00] So... +[12:37:01] Uh +[12:37:06] Your DevMate is our internal build of AI coding partner. +[12:37:17] It works like a chat, you talk to it, you ask it to do things, it understands our code base, it has tools, it has skills, it's able to do code understanding and works for us. +[12:37:20] So this comes pre-installed with VS Code. That better. +[12:37:25] Most of you have probably used this, but for all of our XFN peers who maybe have not. +[12:37:27] ...where to definitelyade. +[12:37:28] Definitely works in our IDE. +[12:37:29] It's really visual. +[12:37:32] It changes the files right in front of you. +[12:37:34] And it's... +[12:37:35] kind of like our... +[12:37:37] de facto go to +[12:37:40] Um +[12:37:44] Clock code. So you must have heard about clock code. It is going pretty viral. +[12:37:46] both outside of meta and within meta. +[12:37:49] It's also an AI kind of. +[12:37:51] programmer or assistant. +[12:37:56] and it's based in your terminal. So it's less about the ID in your terminal directly. +[12:37:56] Um +[12:37:57] No. +[12:38:01] What's the difference between clock code and def mate? Well, one is that it's in the terminal, but... +[12:38:02] to +[12:38:04] It's very customizable. +[12:38:05] insanely customizable. +[12:38:07] people are able to +[12:38:11] uh kind of write all kinds of skills plugins +[12:38:16] and tools and MCP servers can share them. We have this ecosystem called Cloud Templates. +[12:38:18] There is... +[12:38:20] thousands of plugins on there. +[12:38:26] and it has like a bunch of metadata about how many people install loan, you can even outboat certain plugins. +[12:38:27] And so... +[12:38:31] you have this marketplace of so many different plugins. +[12:38:32] that people are sharing. +[12:38:33] and +[12:38:38] you can basically build your own cloud code experience, which I think is the part that a lot of people +[12:38:42] love Cloud Code 4. And so not only does it +[12:38:44] work for coding and as the name kind of +[12:38:48] doesn't really imply, right? Clock code. Do you think it's only for coding? +[12:38:50] People have used ClubCo to do all kinds of... +[12:38:52] each in workflow automation. +[12:38:54] As I mentioned. +[12:38:59] you know, the agent that's responding to you right now in the help desk chat. Like that was... +[12:39:04] using clock code. People are also doing lead lens, all kinds of like... +[12:39:08] you know, leadership summary notes and documents that's coming from Cloud Code. And so +[12:39:11] You can do a lot more than just coding and I think. +[12:39:13] This is one of those areas where... +[12:39:17] You can use Calcota as a beginner, but you can also really get to expert level. +[12:39:22] by understanding how to customize Cloud Code for your workflows. And I really hope that you take the time this week. +[12:39:23] to learn that. +[12:39:26] Very quickly on code compose and write search. +[12:39:28] to be complete and holistic. +[12:39:32] I could compose it basically or auto-complete in... +[12:39:33] or IDs. +[12:39:37] We've had this for a couple years now, so we haven't spent too much time talking about that. +[12:39:38] And racer is +[12:39:41] is another autonomous agent for a large scale. +[12:39:43] Code changes. +[12:39:45] It was built by the Monetization Board. +[12:39:45] Um +[12:39:51] I don't necessarily recommend Racer in general because DevMate and Cloud Code kind of. +[12:39:53] do overlap but just for completeness +[12:39:58] I am including it here because some people within Facebook are using Racer and finding value. +[12:39:59] So... +[12:40:03] you know, racer is there, you can look into it, but I won't spend too much time. +[12:40:05] Cool. +[12:40:07] onto data and analytics. So. +[12:40:10] Perhaps my favorite non-coding agent. +[12:40:12] Analytics agents is awesome. +[12:40:16] I get so much insights from this tool. If you have not used this... +[12:40:17] you are missing out. +[12:40:18] I... +[12:40:21] always go here when I have a data question. +[12:40:30] Literally anything that comes to my mind like how many people use my plug and how many people did that or or what's the you know People that are using AI how much are they using it? We're using it for it +[12:40:32] any kind of data question that comes to my mind. +[12:40:36] I'm talking to Alex agent and it's so powerful because +[12:40:41] I don't even know where all the tables are. And it knows how to go and query and find the right tables. +[12:40:47] it writes the queries for me. Then on top of that, it generates the key takeaways. +[12:40:49] and generates visualizations for me. +[12:40:57] And the data visualizations are awesome because I can actually export them directly into Unidash or share them in a Google chat. +[12:41:03] It is my go-to for any kind of data question. If you haven't tried it out, you really need to check it out. +[12:41:06] For completeness though, there's also dye agent which is +[12:41:11] a similar kind of an agent, but it's built into your, you know, dichory. +[12:41:12] experience. +[12:41:18] I've had less success in like more complex stuff here with this tool, but it is there and it's convenient because it's right there +[12:41:20] where you're writing your own query. +[12:41:25] Data mate is cool and useful for when you're working with unit dashes. So this is. +[12:41:27] when you're on a dashboard. +[12:41:30] Data mate kind of opens up as a sidebar. +[12:41:32] and you can actually talk to it about things on the dashboard. +[12:41:34] so you can actually ask it for insights. +[12:41:37] you can ask it for, you know, like... +[12:41:44] show me some trends or explain why this did happen on this day and like that kind of interactive kind of talk about this dashboard thing. +[12:41:46] Deadamit, deadamit, it's pretty good. +[12:41:47] To that. +[12:41:51] Bencho AI is great for running kind of Python code kind of ad hoc and +[12:41:55] doing rapid visualization of data and for ML workflows. +[12:41:57] So people that work in ML. +[12:41:58] use a lot of bento. +[12:42:01] and Bento AI has been a game changer for them. +[12:42:07] anti-functional specific stuff and just general productivity tools. +[12:42:12] So second brain, you must have heard about second brain right now. It is going pretty viral in the PM community. +[12:42:15] Second brain basically just persists your +[12:42:20] memory for the AI across sessions and what it does is it really like +[12:42:27] pulls in all of your documents pulls in all of your stakeholders and Projects and tasks and whatever you can play basically put all of your +[12:42:28] context. +[12:42:30] your work context into this. +[12:42:31] second brain. +[12:42:32] And that it. +[12:42:39] And then it basically survives every time you use a new on-demand. It's there, it's persistent, it understands you, it understands how you talk. +[12:42:44] How you how you think and it helps you just generate more personalized content +[12:42:45] or refuse. +[12:42:46] Um +[12:42:51] So definitely check that out. We have a session this week on setting up your second brain. +[12:42:53] Stay tuned for that. +[12:42:57] Super whisper this is kind of a niche thing, but I've been using it a lot recently +[12:42:58] I take a lot. +[12:43:02] I have a lot of people that message me, a lot of people, a lot of things to talk about, so... +[12:43:04] you know, my hands started hurting and a lot of people +[12:43:06] I started you to use Super Whisperer. +[12:43:07] to uh +[12:43:09] basically +[12:43:10] Turn there. +[12:43:11] Uh... +[12:43:13] voice into text. And so +[12:43:15] I now, you know what I'm... +[12:43:15] No. +[12:43:21] At the office, I'll be using SuperWhisper to respond to Google chat messages. +[12:43:27] I'll just press the button on my keyboard and just start typing. Sorry, it's speaking and it will just turn. +[12:43:28] immediately into text. +[12:43:28] And so +[12:43:34] People are using this for also writing prompts, because sometimes the prompts get kind of long and you're explaining a lot of stuff to the agent. +[12:43:36] So check that out. +[12:43:36] Um +[12:43:42] Yeah, and then some other daily productivity hacks that we'll just skip over. +[12:43:42] No. +[12:43:48] On prototyping, so within Facebook, we have the FOA Foundation as part of our organization and +[12:43:53] They've been investing a lot into turning prototypes into production quality apps. +[12:43:54] And... +[12:43:55] They have a lot of... +[12:43:57] cool kind of. +[12:43:59] tools and frameworks that have been launched. +[12:44:00] like BlockSlab. +[12:44:04] So we featured Blocks Lab in my video series. +[12:44:10] I think last week or the week before. And basically, this allows you to generate real mobile code. +[12:44:15] using real data within minutes and you can just take an existing block screen. +[12:44:20] kind of describe what you want to change and it will generate a prototype for you that uses real user data. +[12:44:24] Similarly with Wallaby, which is currently beta. +[12:44:28] but it's the same idea, but for native iOS and native Android apps. +[12:44:31] a lifeade. ... ... ... ... ... ... ... +[12:44:31] Another really cool feature if you haven't used it. +[12:44:35] It basically allows you to click anywhere on the auto website. +[12:44:39] within meta. +[12:44:40] and inspire or prompt directly on that thing that you clicked on. +[12:44:42] so you can basically point in fire. +[12:44:47] And so this has been super useful for designers and for people that are like trying to do UI tweaks. +[12:44:48] who might just not know where that... +[12:44:51] piece of code is, you can just point and click. +[12:44:52] on something. +[12:44:56] in your web browser and fire off a prompt. Really, really useful. +[12:44:58] What parody agent? +[12:45:01] It's currently in a pilot status, but what this is trying to do is +[12:45:07] you know, sometimes we build things on mobile and they're not in sync with the things we have on web. +[12:45:11] and this parity agent works on getting parity across mobile and web for you automatically. +[12:45:13] And yeah, so. +[12:45:13] you know +[12:45:17] If you want to find out more about this stuff, there is a group called F-O-A. +[12:45:21] AI for P, FYI. +[12:45:22] and a lot of the big updates are shared there. +[12:45:26] So, quick reference guide. +[12:45:26] You know. +[12:45:30] I covered a lot of content in this, so don't worry, we have a website that will. +[12:45:37] summarize all of this since all available for you. But basically, take a look at what you're trying to do and find the right tool. +[12:45:38] to do that, right? +[12:45:42] There is no one-stop shop in this world. There's going to be a lot of tools. +[12:45:43] All right. +[12:45:48] similar with podcode people are building all kinds of plugins and skills +[12:45:49] and there is a sense of. +[12:45:52] things are moving so fast and so hard to keep up. +[12:45:54] and trust me I feel the same way. +[12:45:55] as well. +[12:45:56] All right. +[12:46:01] It's going to be kind of the nature of this like fast moving rule. We're going to have to be dynamic. +[12:46:07] Be willing to keep relearning re-upskilling ourselves. It's great that we have this week to really +[12:46:08] Focus on that, but... +[12:46:14] you know as Tom said we're probably not going to do this again like this is not we're not going to do this every half or every quarter but +[12:46:14] Um +[12:46:20] there will be a need to constantly relearn what's the latest AI native thing? What's the latest? +[12:46:21] trend or tool. +[12:46:22] And so, um... +[12:46:28] you know, it's a different mindset. Things are not going to be as stable as they used to be. Many, many years ago, it's going to be. +[12:46:31] every couple months, every month, maybe every two weeks, there's gonna be some new thing. +[12:46:34] that you should learn about and adopt in your workflows. +[12:46:34] So. +[12:46:35] Yeah. +[12:46:36] Um +[12:46:43] This is just like a recommendation slide on like what where to start if you haven't used any of these things like +[12:46:44] Start with the core. +[12:46:53] And then by the end of the week, we do hope that you get to the specialized stage where you are kind of understanding how to use men is. +[12:46:55] really use invest auto and meta mate, et cetera. +[12:46:56] Um +[12:46:58] But there's a starting point for everybody. +[12:47:04] this week you have all the time to really go deep and just understand how to use these things. +[12:47:07] Q resources, as I said. +[12:47:08] Um +[12:47:13] There's a couple groups that are good to follow. The FOA AFRA PFI group is good. +[12:47:14] Gen Academy is good. +[12:47:17] uh, second brain at my toe. +[12:47:18] And yeah. +[12:47:23] Last thing I will say, so last slide, it's going to be on the mindset. +[12:47:24] So. +[12:47:25] Um +[12:47:27] just to re-emphasize again. +[12:47:30] It's really important that we +[12:47:32] Focus on. +[12:47:33] Yeah +[12:47:35] evolving ourselves beyond working. +[12:47:37] with these agents. +[12:47:38] in the loop. +[12:47:39] and really +[12:47:43] think about how we can build on the loop experiences. +[12:47:44] If you do something... +[12:47:45] repetitive. +[12:47:46] day over day. +[12:47:51] You should be asking yourselves, how can I get an AI agent to do that thing? +[12:47:53] and free yourself up for... +[12:47:56] higher level thinking or for other more complex work clothes. +[12:47:58] The ceiling is rising. +[12:47:59] And... +[12:48:03] There's never been a time in my career where I felt so empowered. +[12:48:04] to do. +[12:48:05] Anything? +[12:48:09] Like, I can really, whatever I want to build, I can build it now. +[12:48:10] I'm no longer... +[12:48:13] capped by my personal coding ability. +[12:48:14] Um +[12:48:18] I feel like I genuinely have a team of agents that work with me. +[12:48:19] And... +[12:48:22] everyone will start to feel like they're kind of a manager. +[12:48:23] manager of +[12:48:25] team of agents. +[12:48:26] And I really hope that you... +[12:48:28] Take this week to learn. +[12:48:34] about all of these tools that we have and maybe build some of your own as well if you find gaps in the existing tooling. +[12:48:35] and uh +[12:48:37] The goal is by the end of the week to... +[12:48:38] Um +[12:48:40] Have. +[12:48:41] workflows. +[12:48:42] that are +[12:48:48] on the loop experiences for you, that automate parts of your day to day job across all of the functions. So, +[12:48:48] Um +[12:48:53] But yeah, I think it's a great week and you know we should be grateful that we have. +[12:48:55] the week out of our normal business operations to do this. +[12:48:58] And so with that, I'm going to wrap up here. +[12:49:00] and pass it back. +[12:49:00] to +[12:49:01] Uh, dig it. +[12:49:06] Cool. Thank you so much, Bharath. This is awesome. +[12:49:10] I learned quite a few myself looking at those cool tools. +[12:49:11] Cool, next up. +[12:49:13] We have... +[12:49:14] Justin. +[12:49:19] who's going to talk about working with A philosophy, context engineering, and problem-free. +[12:49:20] Justin, you ready? +[12:49:22] Yes, I'm ready to go. All right, on to you. +[12:49:29] All right. Thank you everybody. My name is Justin Jeffers. I am a developer advocate. +[12:49:33] uh based out of Bellevue oh and I support DevMate +[12:49:42] But I also support all AI coding agent frameworks. Like Bharath said, there's a lot of stuff changing, a lot of stuff rapidly. +[12:49:48] coming at you. And so this week is all about teaching you some of the fundamental things to help propel you to become +[12:50:02] I'm going to go ahead and share my screen and I just want to let you know that the presentation for this is going to be made available probably already now as a comment for the event post on workplace. +[12:50:04] So let me go ahead and share my screen +[12:50:09] We are all able to see the presentation. +[12:50:15] OK. +[12:50:20] Sweet, thank you. All right, so yes, the idea of this week is to help. +[12:50:21] Get you all +[12:50:24] onto this AI native track. +[12:50:26] And so to start off, I just want to... +[12:50:29] I have on screen here. +[12:50:33] the AI native journey and I have four different classifications to +[12:50:39] I share with you so we have AI aware, AI enabled, AI first and AI native. And I want you to take a moment. +[12:50:42] to think about yourself today where you sit. +[12:50:46] Are you more on the AI aware side? Are you more AI native? +[12:50:56] Yeah, first and think about what it is that you need to focus on this week while you're taking these sessions while you're in the hackathon to get yourself to AI native. +[12:51:00] and to Bara's point about how things are rapidly changing. +[12:51:08] I want to encourage you to look around the corner and not just think about this week. It's really easy to get focused in what you're currently doing. +[12:51:17] I know from my own experience when I was so I'm fluent in Japanese and I studied Japanese at Stanford's Japanese Immersion School. +[12:51:23] and it was a 10 month program and when I got there I was like I have 10 months I must master this language in 10 months. +[12:51:27] and I tried to do too much and I burnt out. +[12:51:31] And I went to one of my professors and I was speaking with her and she just told me, hey. +[12:51:35] You need to change your mindset. She's like, I watch you every quarter. Like. +[12:51:45] You start out really, really like full of energy and then like you're dead by the end of the quarter. She's like, you need to change your thinking to not just 10 months, but to think of it as a lifelong pursuit. +[12:51:47] And that really helped me. +[12:51:52] focus on what I needed to focus on right there and there in that moment and then to continue. +[12:51:56] learning to this day and that mindset has helped me. +[12:52:08] through my software and computer career as well, knowing that I need to continue to look at, see what's coming, focus on what I'm currently working on, but also keep a finger on the pulse of what's coming up next. So... +[12:52:09] I want +[12:52:11] to encourage you to... +[12:52:16] to think about how you can use these tools today to get you to that AI native. +[12:52:17] Um +[12:52:25] which is what we're all trying to achieve. So I also want to point out that my deck was also created in Manus. +[12:52:35] and AI, I think Jamie mentioned earlier that it's not taking away creativity. In fact, I know I was in a sidebar last night chatting with Barra up late. +[12:52:42] saying this is dangerous because AI takes a limit off of my own creativity and I tend to be very like. +[12:52:47] Ooh, I want to arrange this slide here and I want to put this here and ooh, I got another idea. +[12:52:52] And I can just explain those things to Mattis and it would just change my deck on the fly. +[12:52:59] And I was like, I have to put a cap on my creativity in order to be able to get this delivered. In the past it was like, well... +[12:53:03] It's going to take a little bit of effort to get me to get all these changes in place. +[12:53:08] but with this AI tooling it's totally unlocked. +[12:53:09] more like. +[12:53:15] more than I would expect to be able to innovate. So anyway. +[12:53:22] Again, I just want you to take a look at this and think about where you are and what you need to do this week to get there. And so... +[12:53:26] I'm going to be talking about prompt engineering and context engineering. +[12:53:35] And I'm not going to go too much deep into context or prompt engineering. I think by now we should all be aware of what it is, but. +[12:53:41] in order to understand where context engineering comes from, you kind of have to understand some of the limitations of just... +[12:53:42] thinking. +[12:53:46] and focusing solely on prompt engineering. So. +[12:53:47] Um +[12:53:52] One thing to note with prompt engineering, just like computer traditional programming. +[12:53:54] Garbage in, garbage out, still applies. +[12:53:58] So if you give the LLM some vague input. +[12:54:00] you're going to get. +[12:54:08] a high probabilistic drift. You know, the way how these things work is it's using probability to determine what it is that you're trying to. +[12:54:13] What's going to come next? And so if you give it something completely open. +[12:54:22] You don't know what to expect. So there's a difference between like saying, like let's just say from a practical example, you want somebody to build you a house and you just say, hey, build me a house. +[12:54:30] Well, there's all different types of houses that could be built, right? If you get a mansion, if you get a shack, maybe you get a dog house. Maybe you get a dog house. +[12:54:35] You don't really know. It's open-ended, right? +[12:54:42] what you need to do is to be more specific. So we have a more precise input example here. +[12:54:48] and I'm going to go through and dive deep into what makes this more precise input. +[12:54:49] better. +[12:55:03] So if we take a look at this prompt, we have basically just says write a three-pair graph, executive summary for technical audience explaining how AI content engineering improves agent reliability using an analogy to. +[12:55:05] architecture and blueprints. So that was our prompt. +[12:55:10] So I'm just going to highlight what the different parts of this prompt are. +[12:55:17] First of all, we have our objective. We clearly state what that is. In this particular case, we want a three paragraph executive summary. +[12:55:22] This is way better than just saying, you know, tell me about or... +[12:55:28] you know, because again, you might get reams and reams and reams of text and maybe you just want a TLDR. +[12:55:33] So in this particular case, we're just telling it up front what our objective is. +[12:55:37] Then here we have our audience. This will help us. +[12:55:40] figure out like what we're going to +[12:55:41] to have this. +[12:55:45] uh, come up with and create now within a coding context, we're not going to be. +[12:55:55] Amadeunadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeadeade +[12:55:59] writing unless you're using AI tooling to write documentation for you. So you might want to consider in that particular context. But bear with me while we'll get into the more coding-centric. +[12:56:02] topics in just a moment. +[12:56:05] So then here's our topic. +[12:56:08] So it's AI context engineering. +[12:56:10] And we have... +[12:56:12] you know basically just +[12:56:16] sort of hooks back to our audience, but we're trying to describe. +[12:56:22] how we want this to be focused or how we want this to be written. And so we're saying, create an analogy for us. +[12:56:23] So... +[12:56:25] Um +[12:56:33] Basically what you want to do when you're creating your prompt is to really think about what the objective is, what the audience is, and what... +[12:56:35] what task you wanted to have done for you. +[12:56:43] And I'm going to go through three different frameworks. I'm not going to go into too much detail because... +[12:56:47] Again, the point of this is to go into more so the contact engineering side of things. +[12:56:50] But what I want to do is I want you to notice. +[12:56:58] some of the like some patterns here. I think as you start to see them you'll start to see like what makes a good prompt. +[12:57:05] And those things are going to tie into context engineering. So here is a property framework, APE. +[12:57:14] basically stands for action, purpose, and expectation. And you can kind of see these overlap with the previous example, right? You have. +[12:57:17] Some of these words are sort of synonymous with each other. +[12:57:22] but you have something that you're trying to, you have basically have a goal that you're trying to achieve. +[12:57:26] And this is the action that you wanted to have and you're giving that background as to that purpose. And +[12:57:27] So... +[12:57:33] You sort of have to see the pattern. So next we have care. Again, we have... +[12:57:36] context action results, an example. +[12:57:38] You can see the overlap. We have a... +[12:57:42] and action with this particular framework also introduces a context. +[12:57:46] context really helps, especially when having a human conversation. +[12:57:49] if you're talking about the same thing. +[12:57:58] You don't have to keep repeating things over and over again. If somebody else is listening in from the sidelines and hears, they might not have that context. They might not be able to put the whole picture together. +[12:58:03] Same thing with when working with an AI through multiple sessions, having that context helps. +[12:58:05] the AI continue to +[12:58:10] track along the conversation and produce a better outcome. +[12:58:11] Um... +[12:58:16] So then there's another framework. This is co-star. I believe this was developed like meta. +[12:58:17] Um, +[12:58:21] as you can see this one also has very similar overlaps. So... +[12:58:24] You've got your context, you've got your objective. +[12:58:29] And so I'm not going to go through each and every single one of these, but the point is this is to show that +[12:58:33] to make a good prompt you need to provide. +[12:58:34] basically +[12:58:35] your context. +[12:58:36] You need to have a goal. +[12:58:39] and you have to have a task that you're trying to... +[12:58:42] 停 +[12:58:42] complete. So that +[12:58:46] Yeah. ... +[12:58:48] what makes prompting great but there are some limitations to prompting. +[12:58:55] So for all of you who are working with AI, I want to ask you a question. +[12:59:03] Have you ever written what you thought was a really great prompt, but you didn't quite get the expected result that you were hoping for? +[12:59:05] I know I have. +[12:59:08] And so... +[12:59:09] I'm going to talk about. +[12:59:14] why that could happen. And we're gonna talk about context engineering as. +[12:59:17] as a way to address that. So... +[12:59:23] I want you to consider this. I think you've probably seen this if you drive in park in places like... +[12:59:25] in in crime cities. +[12:59:26] Um +[12:59:28] Hmm +[12:59:29] Let me ask you this, can you park here? +[12:59:40] So if you look at this sign, there's so many different rules, so many different exceptions, you've got different permits and different things. There's a lot to keep track of here and if you're in a hurry. +[12:59:48] Are you going to even bother? Like let's say, you know, this is near a school. Let's say you need to go in your parent and you need to... +[12:59:56] talk to somebody at the school. Maybe your, your son or daughter forgot something and they need your brutes just running it up to school to drop it off. +[12:59:56] um +[13:00:00] you know, can you park here and get a ticket or not get a ticket? +[13:00:07] So, I don't know about you, but I would probably just ignore this, especially if it's gonna be five minutes. +[13:00:10] and just park and do my thing and hope that +[13:00:16] the people who are giving out tickets also don't understand this and as a result don't. +[13:00:17] to get me. +[13:00:19] ah +[13:00:19] So. +[13:00:21] basically +[13:00:25] That's from a human context, but when you have a context... +[13:00:30] when you provide a prompt that's too overloaded with instructions. +[13:00:35] You're going to pollute the context for the AI and that concept is called context rot. +[13:00:42] So what happens is you have way too much stuff vying for the AI's attention +[13:00:44] to figure out what +[13:00:45] it should focus on. +[13:00:49] And so we want to avoid this situation. +[13:00:52] So to understand a little bit more. +[13:00:54] You can kind of think of this as, you know. +[13:00:58] Too many rules means there's no rules because it becomes too noisy. +[13:01:01] and the noise overrides the signal. +[13:01:05] and the LLM can't focus on that. And this happens when +[13:01:08] the LLN's context becomes bloated. +[13:01:11] And so why does this become bloated? Well, +[13:01:12] It's because... +[13:01:18] The way how LLM architecture is based, it's based upon a transformer architecture. +[13:01:22] and the transformer architecture has a self-attention mechanism. +[13:01:30] that compares all the tokens with other tokens, right? So just like traditional programming, time complexity, it matters. +[13:01:35] and it's going to show up in different places, but those same concepts apply in the AI context. +[13:01:38] So instead of having a nest like. +[13:01:41] you're not writing a nested for loop in this particular case. +[13:01:41] but +[13:01:45] transformer architecture. +[13:01:46] İniş. +[13:01:49] comparisons that it's making. +[13:01:54] basically it's going to scale quadratically because it's O of n squared. And so... +[13:01:58] if you double the length of your context, which basically the larger your prompt is. +[13:02:02] the more it's going to consume tokens and +[13:02:07] those tokens are necessary for the AI to figure out what. +[13:02:11] it should do as a result of your prompt. +[13:02:16] So again, if you double the context length, it's going to quadruple. +[13:02:18] the amount of +[13:02:25] tokens and that's going to become super noisy just like that sign. So that's how you get +[13:02:27] context rot so +[13:02:35] How do we avoid context rot? That should be your next question. So how do we do this? Are we going to obsess over our prompts? +[13:02:38] Well, you kind of have to think about it. It's like... +[13:02:39] You need. +[13:02:42] props you need the the +[13:02:44] the context. +[13:02:45] but they're kind of at... +[13:02:47] a tug of war with one another. +[13:02:51] but it's not really a tug of war because they're sort of trying to head in the same direction. +[13:02:54] So it's like a symbiotic relationship. +[13:02:58] but you need to find that balance, right? So... +[13:03:00] You need to make it just right. +[13:03:03] If you follow a lot of +[13:03:07] AI related stuff, you'll hear people talking about Goldilocks. +[13:03:08] Um +[13:03:13] So basically you're just trying to make it just right. Not too much, not too little. +[13:03:17] you just need to find that balance. And part of that balance is... +[13:03:22] not just writing a good prompt but being mindful of what's in the context. +[13:03:24] And so... +[13:03:27] in an early way of +[13:03:33] making sure that you're not overusing the context and overloading it with junk and other things. +[13:03:34] is to +[13:03:37] just clear the context after each task. +[13:03:38] So if you're in DevMate. +[13:03:45] you can close your current window and reopen and start with a whole new context. +[13:03:46] But this... +[13:03:48] kind of solution isn't quite... +[13:03:49] um +[13:03:50] the best. +[13:04:00] I would say this is more along the AI aware solution. So this is very early on in one's understanding and their development with understanding how AI tooling works. +[13:04:03] It's kind of the equivalent to a brute force. +[13:04:06] approach rather than taking a more +[13:04:11] intelligent approach to trying to solve a problem. +[13:04:15] It will work and it will get the job done, but it won't scale very well. +[13:04:17] And so we need to scale because. +[13:04:22] the beauty of this is if you scale this properly and you hook into what +[13:04:24] AI can do by +[13:04:29] working with an agentic loop and making things compound over time. +[13:04:33] the context will be managed by the AI itself. +[13:04:39] with a reflective process and we'll get into that in a bit later on in the presentation but +[13:04:39] So... +[13:04:44] Let's get back to this. So we need to clear that context. +[13:04:45] Bye. +[13:04:46] We can't just... +[13:04:52] the context every single time. So what we're doing is we discovered that we have a new problem that we need to solve. +[13:04:54] and when you clear the context. +[13:05:04] Each time you're going to have to seed that context every single time, right? So you've actually, in trying to solve this problem in that brute force way, you've actually introduced another problem. +[13:05:06] because now nothing has been. +[13:05:25] that you're going to be the bottleneck. The AI is going to move. +[13:05:28] faster than you. So you want to be able to enable the AI to +[13:05:34] basically pass the baton to it and let it manage the context. And so... +[13:05:35] Um +[13:05:47] One of the things that Barra talked about is using the whisper to be able to speak. It's like, yes, the things that you learn as you start to work with AI tooling, it's like, wow, I'm typing. +[13:05:58] all day long, wouldn't it be great if the AI could manage these things on its own? And so this is where context engineering comes into play. And so... +[13:06:03] there's really like three different things that you really need to sort of think about in this context. So... +[13:06:10] you need to be able to curate the context. So you need to choose what goes in to remove what doesn't. +[13:06:16] because anything that is not necessarily needs to be in the context is going to pollute it. +[13:06:19] you need to be able to manage that over time. +[13:06:22] And so you also need to. +[13:06:27] maintain that and keep it within the budget of context. +[13:06:36] being aware of what company tokens you have, just because you have more tokens doesn't mean that you need to use them all as fill up your context entirely, because remember, it scales quadratically. +[13:06:40] So coming up with an effective means to... +[13:06:43] or split up the context into... +[13:06:47] different tasks and to be aware of what it is that you're trying to achieve. +[13:06:52] And maybe this is something that only needs to happen in one context, but doesn't need to happen in another +[13:06:53] So +[13:06:56] figuring out how to split those things up. +[13:06:59] And then also thinking about just beyond the current. +[13:07:01] uh the prompt like +[13:07:03] having memory, having history. +[13:07:12] Let's say the AI is, let's say you had an instruction and in the middle of implementation, something unexpected happens and the AI. +[13:07:13] figures out. +[13:07:15] how to debug that thing. +[13:07:19] Well, if it doesn't have instructions to, like, no... +[13:07:25] to reflect on like, hey, if you have a similar situation like this, you can do this to avoid this thing from happening. +[13:07:32] It's going to every time it hits a similar issue. It's going to have to go through that discovery to debug that issue. So. +[13:07:36] You need to think about like okay, how can we triage issues when they come up? +[13:07:39] And so again, all of these... +[13:07:42] things are competing for the context and I'm going to discuss +[13:07:47] some mechanisms to splitting these things up in a little bit later in the deck. +[13:07:53] But basically what what context engineering is is it's a set of strategies for curating and maintaining +[13:08:00] your token so you don't overfill the context window and you don't end up with context rot. +[13:08:04] And once you have context right, you might have things that +[13:08:07] result unexpectedly. +[13:08:08] So... +[13:08:09] Um +[13:08:15] to put it in perspective, prompt engineering is typically where people start out when they're AI aware. +[13:08:18] and you quickly realize that, hey, it's not just... +[13:08:25] the prompt that I'm writing that I need to be aware of. I need to understand what is in the context at this current moment. +[13:08:31] and in order to really fully automate this, I need to have the AI involved in the process of determining. +[13:08:33] what's in context and what's not in context. +[13:08:36] And if you're using DevMate today. +[13:08:41] and you've used skills and you've used rules and you've used these things. You're already actually +[13:08:43] whether you know it or not are +[13:08:56] working with context engineering. A lot of these tools are, a lot of people are trying to solve these problems and make it easier and reduce the cognitive load of the person using the AI. +[13:09:01] by building these features into it, but there's still a lot to know and to think about while doing these. And so, that's it for today. +[13:09:02] things so +[13:09:03] Um +[13:09:05] when you set up the context. +[13:09:11] properly and you have it such that the AI can manage some of these things for you. +[13:09:20] you can take advantage of an agentic loop. Here's an example. This is a high-level overview of an agentic loop. But typically you have. +[13:09:24] a process where you've got observe, think, act, and reflect. +[13:09:28] And so observe is when the AI agent. +[13:09:28] What's you? +[13:09:29] prompted it. +[13:09:36] this is all the window that's in the context. It's the system props, which is so you also have to consider that too. Like +[13:09:40] context is going to be consumed by other things that are outside of your prompt. +[13:09:44] And if in your prompt you say, hey, read this entire document. +[13:09:51] you could consume all of your your context from just your one instruction alone so you need to make sure that +[13:09:53] You're not loading up everything. +[13:09:55] But anyway, so... +[13:10:01] the AI agent will basically go through the stage of collecting everything it needs to know to get started. +[13:10:08] then it actually reasons over what it observes. So taking everything that it observed, including your prompt, it figures out what it needs to do. +[13:10:19] then once it figures out what it needs to do, it's going to act on it by maybe calling a tool to do something or to write some code to generate text. Maybe it's going to generate an image. +[13:10:22] who knows whatever that output is going to be, it's going to act. +[13:10:24] And then the +[13:10:27] crucial step that I mentioned before is +[13:10:31] to really help get you to that AI native state is to +[13:10:32] how the AI reflect. +[13:10:33] um +[13:10:35] and basically to evaluate. +[13:10:36] What? +[13:10:39] uh was good what was what went wrong +[13:10:42] And so. +[13:10:50] You can observe the logs, you can observe the AI and see where it did something that it didn't expect. +[13:10:51] and then you can +[13:10:55] either manually update the rules or you can +[13:10:59] Converse with the AI as part of that reflection. +[13:11:01] process and tell it, hey. +[13:11:07] When you did this, this happened. We need to update the rules so when you meet this situation. +[13:11:09] to take this action or +[13:11:13] to consult these documents or to do these things. And so. +[13:11:14] Doing that. +[13:11:15] helps you. +[13:11:16] Um +[13:11:21] Accelerate what you can produce and +[13:11:26] This brings you to the concept of compound engineering. +[13:11:27] So... +[13:11:33] you can think of everything as your rules, the skills, your hooks. +[13:11:34] as +[13:11:35] One big. +[13:11:38] as a bunch of different building blocks that come together. +[13:11:41] that builds a compound of things to help. +[13:11:43] the AI. +[13:11:45] Merryade. +[13:11:48] at runtime what it needs in that moment of need. +[13:11:53] And you can also think of the compound as in like compound interest. +[13:11:53] So... +[13:11:55] part of that. +[13:11:56] reflection. +[13:11:57] 最高 +[13:11:59] is to help compound. +[13:12:01] the results of your AI. +[13:12:06] And so what happens is, so you have a task that you want the agent to achieve. +[13:12:15] of that task you should include producing or updating part of whether it's your rules, your skills. +[13:12:18] your context files, your rules files, whatever it is. +[13:12:23] You update those so that way the next time you run. +[13:12:30] whether it's a similar task or the same task or something, it's going to result in having a better outcome. +[13:12:35] So just like compounding interest over time, like you start off and over time. +[13:12:45] You get more and more and more like +[13:12:55] you know, people there is that comment, I receive a penny doubled every day. Eventually it's going to be more than a million, right? So it's the same concept. You want to be... +[13:12:55] um +[13:12:58] making sure that part of your process +[13:13:04] is assessing the outcome and then adopting your set of +[13:13:06] tools. +[13:13:08] to +[13:13:10] Assistance Assistance Assistance Assistance Assistance +[13:13:11] help the AI result in a better outcome after. +[13:13:18] And so I'm going to just dive into a little bit at a high level of these +[13:13:26] different pieces of your context architecture. There's going to be another talk that's going to go more in detail about these and especially in the context of DevMate. +[13:13:29] I'm just going to be talking about these at a high level. +[13:13:32] So basically you can have your rules files, your skills files, and your hooks. +[13:13:39] So rules are basically things that are always in context. These are things that you always want to have the AI consider. +[13:13:41] So +[13:13:47] Again, remember context rot. You don't want to put things. These rules files will always be loaded. +[13:13:52] So you want to make sure when you're creating these and crafting these to only put +[13:13:56] what is absolutely necessary to take place every time. +[13:13:59] you're going to use this AI. +[13:14:08] So things that could possibly be in there is maybe the role, maybe the persona, maybe the coding standards that you have, maybe the coding style. +[13:14:27] adverse and then maybe you also want another +[13:14:30] persona that is more cavalier, more of a cowboy coder. +[13:14:35] and you can have both of them review your code and then look at okay. +[13:14:37] Which things do I need to... +[13:14:39] It's basically like when you... +[13:14:45] asks multiple people to review something, you get multiple opinions saying you can assimilate that behavior. +[13:14:49] am I setting up different rules and personas? +[13:14:53] But you want to make sure that those things are set up just... +[13:14:57] absolutely need to be there because rules files always get loaded so +[13:15:04] The next is skills. So skills are things that are going to be in the moment. +[13:15:12] So you set up and describe what it is and what things you want the AI to do for a specific task. +[13:15:15] and when the AI determines, oh, this is the task that I'm going to run. +[13:15:19] I'm gonna load up the context or I'm gonna add to the context window +[13:15:22] the instructions that are in these skills. +[13:15:27] and then go through and do them. So let's say you have like a... +[13:15:32] a code review skill and your code review skill part of it is hey load up +[13:15:33] Um +[13:15:34] you know +[13:15:45] load up a persona and maybe you have a code review skill and you can specify like okay I want this to be the strict engineer and I want this +[13:15:48] to go through and follow the Thick Code review steps. +[13:15:51] or maybe you have a debugger still. +[13:15:54] And so you can describe in that skill. +[13:16:02] the steps to actively debug and part of that debug process would then to be assess, write up a report and update the context. +[13:16:05] for the next time to improve. +[13:16:07] debugging and also +[13:16:10] maybe you have another... +[13:16:12] Part of your process would be okay. +[13:16:17] Well, maybe we determined that there was something wrong in the original way how the... +[13:16:18] that uh +[13:16:22] this product was built or maybe this code was +[13:16:23] out of date or something. +[13:16:29] So as part of that debugger process could be to go and update. +[13:16:32] the other skills that handle +[13:16:41] building to prevent that bug from happening in the first place. So you can kind of think of it, it's not just think like, okay, I have a task that I want to complete. +[13:16:49] and just walk away from it. Think about like, just like you would when you're managing a software team. It's like, you have a moment, you have a, you have a, a sev. +[13:16:57] You have a moment of reflection, then you have a moment of, okay, what can we do to prevent this from happening again? It's that same concept that you're doing. +[13:17:02] you can do and applies with building up your... +[13:17:05] your AI agent work flows. +[13:17:08] And then lastly, I want to talk about cooks. +[13:17:13] So hooks are basically things that you want to happen a little bit more deterministically. +[13:17:19] So you can basically have a prompt injected into the process. +[13:17:23] So this helps you to be able to like... +[13:17:27] Let's say a bug does happen and you've identified that that that +[13:17:30] Hey, there's a bug. Now I want to hook in and make sure that this prompt. +[13:17:35] that has these instructions applies only in this particular case. +[13:17:41] And I also included on this slide at references, but at references to be used. +[13:17:47] in skills files that can be used in context files, but basically this allows you to basically +[13:17:54] reduce the amount of context by not having to include the full path, the full contents of the file. +[13:17:59] You can see in our example here, we just have fixed the login bug in auth at authpy. +[13:18:02] So we're basically saying, hey, there's a bug in this file. +[13:18:04] and we're using +[13:18:09] this. +[13:18:20] at user model files. So instead of having the AI to go and search through and try to determine where the bug is, you're being direct and you're giving it a shortcut to those files by using the app references. +[13:18:20] Um +[13:18:21] So... +[13:18:23] Um... +[13:18:24] Thanks a lot. +[13:18:27] that we could be talking about by limited in time. +[13:18:35] And so as we draw to a close, I want you to recall the AI Native Journey slide that I shared at the beginning of this presentation. +[13:18:41] And I want you to think about where you are again on this journey. +[13:18:49] And I want you to think about where and what you need to focus on this week, what you've taken away from my presentation that can help you get to here. +[13:18:54] And I want you to think about just as you're going through each and every session today. +[13:19:01] what are the things that you need to focus on this week to get to be a native and then also prepare yourselves for +[13:19:02] the future. +[13:19:05] and thinking about like, okay, what are the things that I can adapt? +[13:19:07] to continue. +[13:19:09] learning to be AI native. +[13:19:14] This is a shameless plug. I am... +[13:19:18] I manage a community on workplace called Coding Agents for the WAN. +[13:19:28] So one of the things I'm really excited about this week is everyone's going to be focused on learning and you're all going to be learning a bunch of different things. You're going to be innovating and so I want to know about these things, what you're doing. +[13:19:41] And so please, if you end up creating something cool, share it in the Coding Agents group. I also run a bi-weekly show and tell where I have people who post in this group and other places on workplace. +[13:19:43] their AI workflows. +[13:19:49] that we did that biweekly. And so if you have something that you would like to share. +[13:19:52] You can either post in that group or reach out to me directly. +[13:19:59] And I wanted to say again, thank you all. It was an honor to be here to speak to you all today, and I wish you luck. +[13:20:05] with your journey to becoming AI native. With that, I'll pass it back over to you. +[13:20:09] danke +[13:20:10] Thank you so much. +[13:20:11] Justin. +[13:20:14] This was really great and thanks for joining us. Really appreciated. +[13:20:21] Cool folks, I think we are we are the we are the getting to the end of the if you white fp white programming +[13:20:23] Again, just to recap a few things. +[13:20:26] Take advantage of this week. +[13:20:33] This is this this week is a blessing, you know, it's just so fortunate. We're able to take this time off to +[13:20:34] uh +[13:20:37] learn, build, and adapt to the new bear working. +[13:20:38] Thank you. +[13:20:41] Okay, it's me. +[13:20:43] your boss at all for +[13:20:45] I'm adding some context in here. +[13:20:47] Adolfo is here. +[13:20:53] So Jarvis provides a summary of the first... +[13:37:56] testing +[13:38:03] For now I'm just gonna keep listening to the meeting so it's gonna keep captioning it. +[13:38:05] I'll let you know later what I want to do. +[13:38:10] Back to the meeting. +[13:38:15] skills will only apply to things in that larger directory. So if you have... +[13:38:16] a particular +[13:38:21] product or framework that you maintain and all of your code lives inside a particular directory. +[13:38:24] Your rules and skills should also live inside that directory. +[13:38:32] If you don't have a sort of legible directory structure, this is a good opportunity to try to create that because that will be helpful for +[13:38:35] LMs who are trying to navigate the code base in addition to targeting. +[13:38:38] You can also use gatekeepers. +[13:38:40] This is useful for... +[13:38:47] keeping an audience size small or even for doing experiments, which is something we can do with rules and skills. +[13:38:53] And then you can also use regular expressions to match on file paths, file contents, or user input. +[13:38:55] So if you want to make sure that... +[13:38:56] Um +[13:38:59] A ruler of scale is only targeted to... +[13:39:00] say a particular +[13:39:01] Um +[13:39:07] a particular type of file or a particular API. +[13:39:09] or a particular concept. +[13:39:11] These are all different levers that you can use. +[13:39:13] to perform that targeting. +[13:39:15] So. +[13:39:17] why it's starting important. +[13:39:27] And this is a concept that you're going to hear me repeating a lot, but that's because it's important and I want everyone to internalize it. But the context window is finite. +[13:39:29] every token has a cost. So... +[13:39:32] when you're pulling in rule files that are not applicable. +[13:39:37] Not only are you wasting tokens, but you're crowding out other information that might be applicable. +[13:39:40] skills even though they are not. +[13:39:43] completely loaded into the context window when they're targeted. +[13:39:44] Um +[13:39:53] You still have to load an entry for each skill, so the agent has to know that it now has an available skill that it can't invoke. And that is not free. +[13:39:55] So, you know, it is a much shorter. +[13:40:04] a much lower cost to have a one line description of a skill than it is to load in an entire rules file, but if you have a thousand of those skills, you're still taking up space. +[13:40:06] So, Targeting is good for everyone. +[13:40:14] It's good for you, it's good for the agent, it's good for the people working in your domain. It's good for the people not working in your domain who you don't want to. +[13:40:16] be polluting their context window with lots of other. +[13:40:18] unrelated stuff. +[13:40:21] And this is not a theoretical problem. This is a real problem. +[13:40:26] where we see too many rules and skills are applied without proper targeting. +[13:40:28] Make sure you target things. +[13:40:29] specifically as you can. +[13:40:36] So some best practices to do that other than just trying to be as narrow as possible. +[13:40:36] Um +[13:40:43] try to put things in your team's directories. Don't put things in top level directories. We do have some... +[13:40:47] rules and skills that apply to top level directories that are more for +[13:40:51] things that are framework related or sort of concepts that are more core throughout. +[13:40:53] repositories, but in general +[13:40:54] Um +[13:40:57] Try to keep things scoped down. +[13:40:58] Um +[13:41:01] leverage recency here. So... +[13:41:02] If. +[13:41:03] You want... +[13:41:10] the system to have access to a particular skill. The best time for it to target that skill is right when it needs it. +[13:41:11] because just like... +[13:41:13] humans we have a bias toward. +[13:41:14] whatever it was. +[13:41:24] made a parent task most recently. So a great workflow for skills is to target it such that the minute the agent tries to +[13:41:27] do something in a particular domain, say it tries to use an API or something like that. +[13:41:31] That's when the skill becomes available and so it can use it right after. +[13:41:32] Um +[13:41:38] You can also set up personal rules and skills, so if you put these in your home directory, these will only apply to you. +[13:41:42] This is useful for a couple things. One, if you just sort of want... +[13:41:44] to land stuff in an experimentation. +[13:42:00] type of capacity and not have to worry about, am I polluting other people's context windows? This is a great way to do that. And then, too, you might have particular preferences or things that are relevant to your workflow that are not relevant to others. Maybe there are certain stylistic things that you like. +[13:42:01] or there are certain... +[13:42:04] pieces of information that +[13:42:08] are important to work that you often do, but that work is not sort of broadly applicable. +[13:42:10] That's a good place to do that. +[13:42:12] And then I said this before, I will say it again. +[13:42:15] Skills also need to be targeted because... +[13:42:16] just +[13:42:19] just being listed as a skill takes a context window space. +[13:42:26] So a couple of dos and don'ts for writing effective rules and skills. +[13:42:27] Have AI write them. +[13:42:33] So a big part of what we're doing here with these is that AI's are very good at... +[13:42:52] And the AIs are good at doing it, so you should let them do it. +[13:42:54] What's the AI's written your rule and skill? +[13:42:56] Please review it. +[13:42:56] Um +[13:43:01] Don't assume AI generated content is correct. It often isn't, although it will sound extremely confident. +[13:43:04] So when you put out these... +[13:43:06] Gips with rules and skills. +[13:43:11] Don't just leave it up to the reviewer to take a look at it. Go through it yourself and make sure it's accurate. +[13:43:12] Um... +[13:43:19] Another thing that you should do here is sort of think of these as kind of a living document. So you'll create a rule file or a skill. +[13:43:26] And then as you start to operate more in your domain, you'll notice areas where the agent is falling short. So you'll notice areas where it's making mistakes. +[13:43:30] And then every time you come across these you want to sort of encode those fixes +[13:43:38] into the rules and skills so that it can learn for the future. So you want to think of these as sort of living documents that you update based on... +[13:43:40] what you see in the real world as opposed to +[13:43:44] You just write a rule file on a skill once, and then you land it, and then you move on. +[13:43:46] And the last one. +[13:43:48] concrete examples. +[13:43:50] much better than... +[13:43:52] abstract conceptual explanations. +[13:43:53] Um +[13:44:06] This is very much true for LLMs. It might be true for humans also. I think we like to tell ourselves that we like to deeply internalize concepts from first principles rather than just pattern matching on examples. +[13:44:07] whether or not that's true. +[13:44:14] I'll leave up to you. But certainly for LLMs, they do much, much better when they have examples that they can pattern match on. +[13:44:19] So some best practices for rules specifically. +[13:44:21] You want to keep them short. +[13:44:24] Again, the sort of 10 minutes to orient a new engineer. +[13:44:26] is a good... +[13:44:27] a good heuristic. +[13:44:32] You want to have references toward the skills that exist in that domain. +[13:44:35] And then think of these as really... +[13:44:38] team-wide owned documents. So... +[13:44:43] Um, you know, discuss with your team how you want to evolve these in... +[13:44:49] reasonable way such that everyone feels like they have shared ownership, that your coordinating changes. +[13:44:54] then you have some sort of principles and guidelines. You definitely don't need to over engineer this process. +[13:44:59] But I think it's worth at the very least sort of having a discussion about your approach for rules and skills. +[13:45:01] and being alive there. +[13:45:05] And then for skills specifically, so. +[13:45:13] A skill should focus on one concept. You don't want to have too many different things because it's the agents invoking the skill when it specifically wants to learn. +[13:45:15] about one thing. So. +[13:45:16] We should. +[13:45:18] make sure that that is... +[13:45:20] represented in the way that we write the skill. +[13:45:21] Um +[13:45:27] It's still important for your skills to be concise, even though they're not loaded into the context window until they're needed. +[13:45:28] Once they're loaded. +[13:45:32] the whole skill is loaded in. So if you have a super long skill... +[13:45:45] even if it's relevant, you're still adding a bunch of overheads to your context window. So being concise is always important, especially when you're having LMS right rules and skills, because LMS are notoriously not concise. +[13:45:48] So make sure this is something that you're keeping in mind when you write these. +[13:45:54] Another guideline that I like to follow is to keep a flat hierarchy for skills. +[13:45:58] So you don't want to have a complex tree of skills delegating out to more skills. +[13:46:03] try to put everything at the top level so that the system can find what it needs more easily. +[13:46:11] It's also okay to reinforce stuff that was already in the rules file because of this recency bias at the time that the skill is being invoked. +[13:46:13] stuff that it might have already learned in the past. +[13:46:16] is potentially, you know, way back. +[13:46:19] toward the beginning of the context window and +[13:46:22] You know, it might not have that information as well as... +[13:46:29] readily available. So it's totally okay to reinforce things in a skill that you've already put in a rules file. +[13:46:40] So something I get asked about a lot is, okay, how are skills different than docs, especially if you're using a system like static docs where your docs are already represented in Markdown files? +[13:46:44] So the difference is sort of more practical than it is theoretical. +[13:46:49] Agents have the ability to query our documentation via our MCP tools. +[13:46:51] but they don't do it very often. +[13:46:56] whereas skills are this sort of core-agentic concept. So anytime +[13:46:58] skills are targeted. +[13:47:03] they will become available to use and the agent has a much more intrinsic +[13:47:04] understanding of them. +[13:47:05] versus +[13:47:13] documentation relies on it saying, okay, I have all of these MCP tools. Okay, here's one for searching documents. Maybe I'm gonna use that now. So in practice it just... +[13:47:15] It doesn't do that a lot. +[13:47:15] Um +[13:47:19] So skills are a great way to leverage something that is more... +[13:47:22] core to the agent system. +[13:47:24] to be able to feed information rather than... +[13:47:26] relying on it to... +[13:47:26] Um +[13:47:30] use more bespoke methods for querying information. +[13:47:33] And then another thing documentation tends to have. +[13:47:37] to be much more hierarchical. So, you know, you have lots of... +[13:47:37] Um +[13:47:40] you know tabs and sub tabs and things like that. +[13:47:46] Those can be difficult for the agent to gather, whereas the flatter structure of skills is more. +[13:47:47] Um +[13:47:54] More useful that said if anyone is working on unifying skills and docs into a single source of truth +[13:47:56] Let me know because I do think there is... +[13:48:00] some good opportunity here to try to bring these things closer together. +[13:48:03] I've often found I feel like I'm sort of maintaining parallel. +[13:48:09] systems of information. So if anyone's working on that or wants to work on that for the hackathon this week. +[13:48:11] Let me know. +[13:48:16] Our organization also has a tool called Auto Knowledge. +[13:48:20] So auto knowledge goes through real life. +[13:48:22] definite trajectories. +[13:48:24] When I say trajectory, that is... +[13:48:27] a sort of record of all the actions that the +[13:48:31] agent is taking all the reasoning that it's done. So it goes to those trajectories. +[13:48:39] finds where there are knowledge gaps and then puts out dips to remedy that by adding the information to rules and skills. +[13:48:41] So usually it just does this to augment. +[13:48:45] existing rules and skills but it can also create new ones. +[13:48:49] if it's high confidence. So this is a project that is meant to +[13:48:52] help automate a lot of the rule and skill maintenance. +[13:48:54] based on real life data. +[13:48:56] and they're also a number of... +[13:48:59] of great additions that are +[13:49:02] the +[13:49:08] that are in progress for this. So let me know if you have any other questions about auto knowledge, and I'm happy to point you toward the people who maintain it. +[13:49:14] So, you've written a rule file or a skill or you've made a change. +[13:49:15] How do you test it? +[13:49:16] Um +[13:49:18] This is really difficult. +[13:49:18] Uh... +[13:49:21] Agents are non-deterministic, I think. +[13:49:22] We all have a good understanding of that. +[13:49:24] right now. +[13:49:29] non-determinism is really frustrating as a software engineer. The way that you... +[13:49:30] typically go about a problem is +[13:49:31] you +[13:49:32] find a +[13:49:34] Repro the problem. +[13:49:36] and then it repros consistently. +[13:49:37] and then you make a change. +[13:49:41] and then it stops repro-ing and it does that consistently. +[13:49:46] and that's just not how it works with deterministic, with non-deterministic coding agents. +[13:49:47] So... +[13:49:49] This is a big challenge. +[13:49:53] Something you can do generally for your domain is to write benchmarks. +[13:50:00] benchmarks will be very useful, but they're also really expensive to put together and there's still kind of noise that we found in practice. It's very hard to see. +[13:50:02] statistically significant movement against. +[13:50:05] benchmarks through rules and skill changes. +[13:50:06] So. +[13:50:07] What do you do? +[13:50:12] At a bare minimum, what I would suggest is try to find some sort of... +[13:50:18] case that demonstrates the problem. This is functionally your repro. Don't assume it's going to repro every single time. It likely won't. +[13:50:19] Um +[13:50:21] But see if you can get... +[13:50:23] some kind of before and after. +[13:50:24] Um +[13:50:28] with the repro going away. And again, it might not be consistent, but if you can sort of +[13:50:29] Get it to show up. +[13:50:36] you know once or twice Before you made the fix and then get it to go away once or twice after you've made the fix +[13:50:40] Probably sufficient to show that you're moving in the right direction, but it's not going to be the type of +[13:50:44] rigor that you're used to in other domains. +[13:50:49] One way that we can do a little bit better here, so there is a... +[13:50:51] tool called the trajectory debugger. +[13:50:57] This allows you to do that kind of A-B testing, but holding a lot more things constant. +[13:51:02] So the way that it works is if you're looking through a DevMate trajectory, you can go to a particular step. +[13:51:04] and then you can... +[13:51:06] replay everything. +[13:51:08] up to that step. +[13:51:11] Make a change and then see what the next +[13:51:13] step would be +[13:51:15] with and without the change. +[13:51:17] So it's a similar concept of... +[13:51:21] doing the before and after but this allows you to hold +[13:51:30] much more of what the agent's doing constant between the before and after. So this is a good way to have slightly more rigor when you are... +[13:51:31] making these changes. +[13:51:40] Again, still not deterministic. So, you know, the action that it takes next after you make your change is not going to be the same each time you run it. +[13:51:41] Bye. +[13:51:43] At the very least, you know you're holding everything else constant. +[13:51:45] The debugger tool also lets you... +[13:51:49] run each step a few times so that you can see, okay. +[13:51:54] before we ran it the next step five times and it got it right, you know, zero times. +[13:51:57] After we ran it five times, it got it right three times. +[13:52:00] And so you get a little bit more rigor there. +[13:52:03] If you have any questions about the trajectory debugger. +[13:52:08] There's an internal group for the general definite quality tools. +[13:52:10] that I would suggest you check out. +[13:52:12] So... +[13:52:15] Sort of quick getting started checklist. +[13:52:15] Um... +[13:52:17] First, figure out the... +[13:52:21] the boundaries of your domain. So if you're a product team, this is +[13:52:27] you know, the product you're developing, if you're an info team, if the infrastructure you maintain, and you probably have sort of different sub-domains. +[13:52:28] within that. +[13:52:32] Figure out the critical things that you have to know. Those are your rules. +[13:52:36] figure out the topics that you want to do a deep dive on those your skills. +[13:52:41] set up the targeting appropriately, you'll probably want to share a lot of the targeting across different things in your domain. +[13:52:44] Let the AI write them, review them yourself. +[13:52:46] test them as best as you can. +[13:52:52] and then iterate based on real agent behavior. So see how the coding agents do and +[13:52:54] when they make mistakes. +[13:52:56] update your rules and skills to reflect that. +[13:53:02] So just a quick sort of more concrete case study here. So this is our. +[13:53:05] blocks rule file or at least this is a subset of it. +[13:53:06] Um +[13:53:08] You'll notice we have lots of targeting at the top. +[13:53:11] You'll see here that this LLMsGK is... +[13:53:13] It's a kind of weird... +[13:53:17] GK name. It's called Blocks Unified Style Skills Disabled. +[13:53:21] This is because you're actually currently running an AB test. +[13:53:24] We have a new API that we're rolling out. +[13:53:30] And right now, 50% of users will get rules and skills that tell it to use the new API. +[13:53:34] At 50% of users will get rules and skills that tell it not to use the new API. +[13:53:38] So it's pretty cool that you can run these sorts of experiments. +[13:53:42] And you could never do that in the old days. You would just have to. +[13:53:47] you know, make a post and ask a bunch of people to use your API and hope that they listen. So. +[13:53:50] A lot of interesting implications here for framework development. +[13:53:53] And then I just included a couple screenshots from... +[13:53:58] the rules file so we have you know sort of a sentence or two on the hack language. +[13:53:59] Um +[13:54:03] We have a list of XHP syntax rules. You can only see the first one here. +[13:54:10] And then we have for a bunch of our four components, again, sort of a brief description, and then we say, okay, for more information on Flexbox. +[13:54:12] you can look at the Blocks-Blocks-Box skill. +[13:54:22] And then here's an example of one of those skills. So for navigation, again, a lot of targeting here to make sure that we are targeting things that are navigation specific. +[13:54:23] Um +[13:54:28] And then we have a whole bunch of examples. Again, I don't have room to show the whole example, but. +[13:54:31] we have that triple tilt, a triple back tick. +[13:54:32] Hack. +[13:54:38] And then there's a block of code below that, and we have this for all of our different navigation APIs. +[13:54:38] Um +[13:54:41] So, at the time that... +[13:54:43] Demi tries to +[13:54:49] write a blocks navigation API, it will likely get it wrong, but then it'll get the blocks navigation skill targeted. +[13:54:52] It'll invoke the scale and we'll see if it gives you all the JPI's. +[13:54:54] So +[13:54:59] If you take away five things, these are all things that I've said multiple times, but I think they're worth reinforcing. +[13:55:01] Context Windows precious. +[13:55:04] Rules are for orienting you, skills are for doing a deep dive. +[13:55:07] Examples are better than explanations. +[13:55:10] Let AI write your rules and skills, but review them yourself. +[13:55:13] and testing is hard, but there are ways to do it. +[13:55:18] So with that, I will open the floor to any questions. +[13:55:22] Thank you, Julian. +[13:55:32] We have quite a few questions in the chat thread, so we're going to get through as many as we can in the next five minutes and then feel free to reach out to Julian directly. +[13:55:35] afterwards if we don't get to your question. +[13:55:36] Alright, the first one. +[13:55:41] Do MCP servers work across DevMate and Cloud? +[13:55:46] So there's a separate talk on MCP tooling, but generally speaking... +[13:55:49] Yes, but there are different... +[13:55:55] mechanisms for integrating MCP tools into DevMate versus Cloud. So integrating MCP tool in DevMate. +[13:56:00] You can use the rule and the skill file to add it to the MCP directory. +[13:56:06] Claude has a different mechanism for exposing them that I'm less familiar with, but that will probably be gone over in the MCP talk. +[13:56:09] Next question. +[13:56:18] I have struggled with do X and don't do Y. Sometimes it just straight ignores it. Any guidance on how many rules I can shove. +[13:56:23] Yeah, so think about the context window as... +[13:56:26] A thousand people yelling at you. +[13:56:29] to do and don't do lots of different things. +[13:56:31] Everyone's competing for attention. +[13:56:36] and it's really tough to get that attention. And even when you get it... +[13:56:42] It's very hard to get these things to re-broke consistently. Skills are really helpful here because... +[13:56:43] they trigger. +[13:56:44] when. +[13:56:45] Um +[13:56:48] the particular relevant concept is coming up. +[13:56:51] So I would suggest +[13:56:56] trying to sort of put as much as you can into skills and then +[13:56:59] focus on having good targeting for those that at the time that +[13:57:02] particular error surfaces or particular +[13:57:05] practices is used that you want to avoid. +[13:57:14] that the skill gets triggered. But the real answer here is this is not a solved problem by any means. I think everyone has encountered this pain point. +[13:57:15] And um +[13:57:18] It's just really tough, especially in this sort of mono repo environment. +[13:57:22] All right, next question. +[13:57:30] Do we have to run Trajectory Debugger manually? Can we automate this for every rule slash skill change? +[13:57:32] So you +[13:57:34] have to +[13:57:37] That's a good question when people maintain that. +[13:57:40] maintain that tool. There's a certain amount of +[13:57:52] that workflow that is manual for now because you need to identify where in the trajectory you're going to apply. But that does seem like something that's sort of a good candidate for automation. So I would imagine... +[13:57:57] they're working on that. I know you can also run larger scale back tests, but yeah, I mean it'll be great if you +[13:58:01] if you made a rule file diff and then you just automatically got +[13:58:04] You know, here's a run of it with a trajectory debugger. I will +[13:58:08] I will suggest that and see if that's something that they're currently working on. +[13:58:11] All right, great. Next question. +[13:58:15] I already have some clawed skills that leverage bento to do metric. +[13:58:21] analysis. So I'm curious if it's worthwhile for me to translate to recipes or not. +[13:58:23] Um... +[13:58:25] I am +[13:58:28] less sort of up on the general like... +[13:58:32] differences between skills and recipes so I would say if you're +[13:58:38] If your workflow is working for you, it's probably fine to leave it that way. +[13:58:40] But I'm not an expert on that. +[13:58:42] All right. +[13:58:43] Next question. +[13:58:49] Why is flat hierarchy better than having skills delegating to other skills? +[13:58:50] Um +[13:58:57] In my experience, they tend to read one or two skills. They tend to not do a ton of... +[13:58:57] in depth. +[13:58:59] traversing of +[13:59:03] skills or really in any sort of MCP tool usage it's kind of like +[13:59:05] you have its attention for a second, it will. +[13:59:10] you know, look at one or two things and then it will move on to something else. So... +[13:59:10] Um +[13:59:17] if you have a sort of more hierarchical system, I have just found anecdotally that +[13:59:21] you're less likely to get all the information that it needs. +[13:59:31] Got it. Okay, great. We have one more minute. So let's try to get one more question in and then we're we'll go ahead and end the session. So +[13:59:36] Last question, in the kickoff session there's a concept of hooks in addition to rules and skills. +[13:59:39] Does this concept apply to the edge world? +[13:59:45] Yeah, so hooks are much more deterministic ways to say... +[13:59:47] take this explicit action. +[13:59:51] Because of that, they're a lot more locked down because if you had everyone +[13:59:57] forcing the agent to take a particular action all the time your system would explode. So... +[13:59:57] Um +[14:00:02] I think the DevMe team is still sort of iterating on the right ways to expose +[14:00:06] hooks in a more reasonable way, but they're definitely a related concept. +[14:00:16] I think we're about to be at time. Thank you so much, Julian. And if folks have questions, you can go ahead and ping Julian directly. +[14:00:18] Thanks so much everyone. +[14:01:59] And the purple lines are previous that. +[14:02:01] 20261. +[14:02:02] compared to our lines in. +[14:02:04] 2023, Form 5. +[14:02:07] It's just an order of magnitude. +[14:02:10] almost in terms of the growth we're seeing. +[14:02:12] This is a massive inflection point. +[14:02:16] and it's changing the way that we're reviewing code. +[14:02:20] humans, our time to review code is finite. +[14:02:23] The effect we're having here is exactly as you'd expect. +[14:02:26] fewer diffs are being unblocked in a timely manner. +[14:02:28] than we've seen historically. +[14:02:29] All right. +[14:02:32] I'll point again to the purple line here where you can see that more and more this. +[14:02:38] are not being unblocked within a day, whereas historically we've kept a rate of over 80% of dips being. +[14:02:39] reviewed in a timely manner of. +[14:02:41] 24 hours or less. +[14:02:44] and AIGIFs. +[14:02:50] are both the most responsible for the increase in dips we're seeing and the size of them, but they're also the most impacted. +[14:02:54] AI diffs are the most likely to not be reviewed within a timely manner. +[14:02:58] So the way the diffs is hit. +[14:03:03] To keep moving fast, we need to evolve how code review works at Meta. +[14:03:07] There's a couple of ways that we're tackled on this. +[14:03:08] One is defense and depth. +[14:03:11] as Santosh's post in data info. +[14:03:12] Put it. +[14:03:17] last week, our infrastructure needs to become incredibly, incredibly resilient to this constant change. +[14:03:22] We need to be able to automatically detect and prevent issues from lending in our co-base. +[14:03:25] and shift left and fix them automatically when feasible. +[14:03:29] by preventing bugs, regressions, and cells before they can land. +[14:03:33] through automatic AI review and prevention and detection. +[14:03:40] The investments we're making here can let us be more aggressive and more confident in the changes that we do let through more efficiently. +[14:03:46] In terms of review efficiency, we're also investing in tools to help you review more changes. +[14:03:48] more quickly and with less effort. +[14:03:49] Vão! +[14:03:51] not sacrificing quality. +[14:03:55] the mechanisms that we're adding. +[14:04:03] will also enable us to eliminate the review burden when it's unnecessary, such as in the case of simple, low-risk gifts that pass AI quality review. +[14:04:05] and we'll get more into that later. +[14:04:07] So. +[14:04:09] talking about defensive depth. +[14:04:12] I want to start by talking to you a bit about Dev make code review +[14:04:18] Deming CoderView is a centralized CoderView agent that we've developed that performs a holistic +[14:04:20] code review on any code change you supply it. +[14:04:21] fabricator diff. +[14:04:22] version. +[14:04:23] and arbitrary commit. +[14:04:25] or even just your local changes. +[14:04:29] And these reviews produce structured line-level code review findings. +[14:04:30] that we can +[14:04:33] used to highlight Cobra Refined issues directly in your diff. +[14:04:35] and it can run anywhere dummy cam. +[14:04:38] Here's an example of a fabricator signal. +[14:04:39] that has been +[14:04:40] produced by DemiReviewer. +[14:04:44] And you can see the likes and reactions to it. +[14:04:45] Um +[14:04:46] Um +[14:04:47] that were produced. +[14:04:48] And. +[14:04:51] as a result of us automatically catching these issues. +[14:04:55] Yes。 Yes。 +[14:04:56] our AI agent. +[14:04:56] Since we've +[14:04:58] Published in a cut of you +[14:05:01] we've seen over half a million diff signals addressed. +[14:05:03] the majority of these. +[14:05:04] in the New Year alone. +[14:05:05] And again. +[14:05:08] Over time we've seen a really great sentiment. +[14:05:10] from the release of Death May Cutter View. +[14:05:16] hugely positive reactions and great feedback including anecdotes and stories about the subs and bugs. +[14:05:17] that this. +[14:05:19] protective layer of AI review. +[14:05:21] has +[14:05:27] prevented. At the same time, we want to shift this left. +[14:05:28] Right? +[14:05:32] We've, and to do this, we've integrated AI code review into VS Code. +[14:05:33] so you can now easily +[14:05:35] kickoff. +[14:05:41] an AI code review in your IDE, the simple command slash review code. We also have some +[14:05:44] easy entry points to help expedite this such as +[14:05:50] a button when you're preparing a diff in the interactive smart log to help kick off a review automatically. +[14:06:01] And so here's an example of us going through the review flow. Here I've opened up one of the subagents, review subagents that's running as part of our VS Code review flow. +[14:06:07] And here you can see that it's leveraging the expertise around code-based guidelines from. +[14:06:10] context files, in this case, that make rules. +[14:06:11] files. +[14:06:15] and it's enforcing that guidance against the code change that I've created. +[14:06:18] using all of the eligible. +[14:06:23] rules that meet the criteria for the files that are modified in this dip. +[14:06:25] And so. +[14:06:26] Here's just an example of. +[14:06:29] the agent going through the review workflow in real time. +[14:06:36] and you can toggle between the sub-agents easily to debug their workflows, export the trajectories, etc. +[14:06:37] In this case. +[14:06:38] the review. +[14:06:39] time. +[14:06:43] found that the code was good that it didn't violate any of the rules that we had set up. +[14:06:48] and so was able to produce a finished review letting me know that my code did not violate. +[14:06:51] of the code-based guidelines and that it didn't identify any major +[14:06:54] bugs or mistakes. +[14:07:00] Now here's an example of what it looks like when we do find issues. +[14:07:04] Here you can see that we've highlighted a... +[14:07:08] a couple of structured issues that we identified, including links to the files. +[14:07:10] and code ranges where those issues are. +[14:07:18] And we make it really easy for the author to select the issues that they agree with and one click button to have them automatically address those. +[14:07:26] the most codicious。 Thank you. +[14:07:31] Now, more and more folks at the company are also leveraging Cloud Code, and so I'm excited to share that DenonCoderDue is now available as a Cloud Code plugin. +[14:07:34] this isn't enabling you to easily download and incorporate. +[14:07:36] Identicality into your local cloud code +[14:07:41] food." +[14:07:43] So just look out for the DCR plugin and again installing it is the easiest. +[14:07:46] 5-templates plugin DCR install. +[14:07:49] This offers you the slash review code command, but you can again. +[14:07:59] run on your code changes, you can provide it with specific instructions like asking them to tailor a review to something you care about in particular at inspecting for that diff. +[14:08:00] or cooking. +[14:08:06] Now, I mentioned layered review and I did... +[14:08:09] in the VS Code demonstration shooting sample of +[14:08:11] what we are calling expert review. +[14:08:12] which is R. +[14:08:16] rules integration or context files integration into. +[14:08:18] that make code review that allows anyone. +[14:08:20] to customize a high-review in their code base. +[14:08:23] and it works in both fabricator and DS code. +[14:08:29] We built this integration because the metacode base is really rich and highly diverse. +[14:08:32] or team project level context is essential. +[14:08:34] for effective code review. +[14:08:37] and so by writing context files like rules files. +[14:08:42] you can improve the performance and effectiveness of AI agents that are offering and reviewing. +[14:08:43] in your codebase. +[14:08:45] At the same time. +[14:08:49] Oh, and so here's just a quick graphic showing you how are +[14:08:50] CodeRedu agent is working. +[14:08:52] by dispatching. +[14:08:54] It's review... +[14:08:55] I don't know. +[14:09:04] behavior to several different review agents, one that is our general reviewer that works on any dip that the company to catch common logic bugs and mistakes, but also enforcing +[14:09:07] Like I mentioned, this or team and project level context through. +[14:09:09] real space expert review. +[14:09:11] ultimately aggregating those findings. +[14:09:13] And we also recognize +[14:09:16] the interest of progressive disclosure. +[14:09:18] and situations where we don't. +[14:09:19] Yeah. +[14:09:21] maybe want a roll file to. +[14:09:22] be leveraged. +[14:09:25] outside of Coderview, for example. +[14:09:32] Some rules only make sense to trigger during codefuse scenarios, not any time an eligible file was opened by an agent. +[14:09:36] And so we now support the ability to scope rules to code only by leveraging. +[14:09:41] rules is apply to clients context filter. So you can see in this example on the right. +[14:09:43] this rules file. +[14:09:45] and set several client filters. +[14:09:49] If you want your rule, if you want to build a rule that's tailored to Coderview, you can add that. +[14:09:54] code review client and it will make sure that rules are only discovered and only applied. +[14:09:56] in our code review scenarios. +[14:10:07] We're also working on adding support for skills that's coming imminently as it's how most teams tend to be authoring context files today. We want to make sure that that's well supported. +[14:10:12] And another future extension we're exploring is support for more customer reviewer agents. +[14:10:14] that need to do more complex. +[14:10:20] tasks than just providing us with context and code-based guidelines that we can enforce. +[14:10:29] So you might be wondering, how does DevNet CoderU and other AI review agents at the company run on Dips and produce those signals? +[14:10:33] Well, to support this, we built something called the AI diff reviewer platform. +[14:10:34] or ADR. +[14:10:37] The AI diff of your platform. +[14:10:41] is an integration point for that host diff review agents at Meadow. +[14:10:45] This comes batteries included. We provide. +[14:10:48] the +[14:10:49] automated orchestration capabilities, robust instrumentation. +[14:10:54] and critical effectiveness and sentiment metrics and reporting. +[14:10:56] including support for your dashboards. +[14:10:59] that you can explore to see the performance of your. +[14:11:02] custom-difficulty agent detector. +[14:11:06] Your job is just to bring a high quality AR review agent. +[14:11:07] and make sure it's effective. +[14:11:10] Here's an example of the... +[14:11:13] ADR platform architecture where you can see that we are. +[14:11:15] helping orchestrate. +[14:11:16] reviews. +[14:11:17] through ADR. +[14:11:19] for eligible detectors. +[14:11:19] light. +[14:11:23] the Detmay Code Review Agent and other agents of the company. +[14:11:29] Now the signals that the AI diff from your platform produces. +[14:11:37] contain a lot of really critical features through a support for a custom fabricator UX that we've created and designed. +[14:11:41] Importantly, we have really clear attribution. +[14:11:42] to the... +[14:11:45] agents that produce these signals. +[14:11:49] So in the top left, you can see the model or provider responsible for the signal that we produce. +[14:11:53] as well as severe ease that are controlled by the... +[14:11:55] agent that's producing these findings. +[14:11:59] You can provide AI prominence. +[14:12:08] information, for example with DemiCoderView signals, if you hover over that generated by AI token, we actually link you to the trajectory that produced the signal, and so you can see exactly why. +[14:12:11] That signal was detected by AI. +[14:12:16] which is very useful in debugging and feedback for agent developers. +[14:12:18] we provide a consistent. +[14:12:22] universal feedback interface at the bottom where you can easily +[14:12:28] thumbs down signals that you disagree with, report bugs to the owning team, and even delegate +[14:12:29] fixes. +[14:12:31] to AI easily. +[14:12:40] Now we've been talking about defense, let's talk about some of the ways that we're making ColorView at Meta more efficient. +[14:12:44] fabricators sent to Agent Future. +[14:12:52] allows you to delegate feedback and fixes to AI agents. Like in the signals I showed earlier, you can click that fix to AI button. +[14:12:56] fix with AI button to kick off send to agent. And here's a quick video. +[14:13:00] That goes through all of the entry points that we support for sent to agent. +[14:13:03] including those the ability to +[14:13:04] Triggered +[14:13:05] send an agent. +[14:13:06] to address comments. +[14:13:07] to fix. +[14:13:11] specific or all failed signals on your death. +[14:13:12] or even to apply. +[14:13:15] arbitrary requests. +[14:13:15] Um +[14:13:17] Simply asking an agent to +[14:13:20] Make a change that you wanted to. +[14:13:21] tweak your diff with. +[14:13:23] without having to go and check out. +[14:13:25] a new one demand. +[14:13:27] Thank you. +[14:13:31] In this example, you can see that... +[14:13:33] very quickly. +[14:13:35] the fix was produced we were able to. +[14:13:36] Review it. +[14:13:38] click publish and +[14:13:42] produce a new version of the diff with changes incorporated, without ever having to leave fabricator. +[14:13:48] Our kidder team is also investing in powerful other powerful tools for review efficiency. +[14:13:50] like the new review assistant. +[14:13:54] which helps you review disks leveraging context from signals. +[14:14:06] then a code review, diff risk score, and more. And in your settings, you can actually set up the review system to run automatically on your diffs when you open them. So in this case, you see a review assistant where the review actually happened in the background. +[14:14:07] I'm about to quickly finish. +[14:14:12] found that the dip was low risk and that it applied consistent co-base standards. +[14:14:16] the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important +[14:14:17] and was able to provide me with a... +[14:14:22] suggested review action to take, which helped me expedite my review. +[14:14:25] So in this case, you can see that D. +[14:14:31] suggested action was accepted, which concurred with my evaluation of the diff. +[14:14:33] and then I was able to click the button to then. +[14:14:33] All right. +[14:14:37] you know, one click, take the action with a comment that I could customize. +[14:14:38] That's the start. +[14:14:46] Another great review efficiency tool that we've built is multi-diff review. +[14:14:48] So the fabricator team is built. +[14:14:50] multi-differ view. +[14:14:52] to address the issue of. +[14:14:53] Um +[14:14:57] making it easier to review multiple disks simultaneously. +[14:15:00] This is really great for batching post-land reviews. +[14:15:06] So for example here I've selected two such postline reviews, one that needed final review, one that needed owner review. +[14:15:16] And I was able to easily close both reviews together in one click after quickly vetting the changes past semester to me for both of those. +[14:15:17] Oh, it diffs. +[14:15:19] And they can also... +[14:15:21] If I found any issues with them, go back and... +[14:15:24] you know +[14:15:26] request follow-up with the authors. +[14:15:31] Now, not all diffs are created equal. +[14:15:35] reviewing every single diff that the company is no longer scalable as I +[14:15:39] alluded to with the wave of diffs analysis at the start of this presentation. +[14:15:43] some just absolutely merit review. It's critical. +[14:15:45] but others, not so much. +[14:15:49] think minor tweaks, documentation updates, et cetera. +[14:15:54] where a developer maybe shouldn't have to ping their co-worker for a quick stamp. +[14:16:00] We've also seen that review queues have been blown up by repetitive AI code modifs fixing Wnt issues. +[14:16:04] all of which require review for, frankly, often trivial changes. +[14:16:06] So to address the review bottleneck. +[14:16:08] We started building something we're calling. +[14:16:09] right now. +[14:16:11] risk-aware diff auto review. +[14:16:17] This is an evolution of what we've already started with demigdia verification and deferred reviews. +[14:16:25] It's leveraging the combination of AI and deterministic heuristics to identify and then block simple low risk and high quality diffs. +[14:16:26] by removing. +[14:16:28] the human review recording. +[14:16:29] on those disks. +[14:16:31] In this example, +[14:16:32] you can see on the left side. +[14:16:34] a diff that has been approved. +[14:16:37] by radar to land without review. +[14:16:41] without human review. It's gone through a full some robust AI review. +[14:16:44] And we make this status now visible too. +[14:16:51] any viewer at the company. And we provide a lot of context and explanations for why it passed our mustard. +[14:16:53] why a past muster end is eligible. +[14:16:55] including +[14:16:57] debug links to understand the policy evaluation. +[14:17:00] for wide radar pass on that diff. +[14:17:01] Oh +[14:17:09] we provide notices as well, like near the diff actions on the right. So you can see very clearly what is going on with this diff. +[14:17:12] and again, better contextualize why it's eligible to length out review. +[14:17:18] Now, here's an example of a... +[14:17:21] Human diff on the left and a bot diff on the right. Both of them. +[14:17:23] which passed our radar policies. +[14:17:30] One thing I wanted to emphasize here is that all radar events are visible and auditable in diff transactions. +[14:17:34] You can see exactly which versions passed our radar reviews. +[14:17:35] And... +[14:17:41] One thing that's really critical to emphasize here is that radar reviews are perversion only versions that pass +[14:17:42] can land. +[14:17:48] For human disks, if you update your disk with a new version after it's passed radar review, it's going to have to be +[14:17:50] reviewed again before you can. +[14:17:53] and pass once more before you can land. +[14:17:56] because we really care deeply about making sure that. +[14:17:58] Every death that goes through this flow is +[14:18:00] up to +[14:18:00] Yeah, the +[14:18:02] the high bar that we require. +[14:18:04] and +[14:18:06] On the code monster side, we make sure that... +[14:18:07] accepts are. +[14:18:09] downgraded if. +[14:18:11] a meaningful change is made to the diff. +[14:18:13] after it's been approved by RAIDR. +[14:18:19] Something else I'm excited to share is that Facebook. +[14:18:21] It is now piloting radar itself. +[14:18:26] Orcs like Facebook are actually in control of their own risk tolerances. +[14:18:29] for radar eligibility as well as +[14:18:30] out. +[14:18:34] which code automations, like which code mods are eligible to run. +[14:18:35] and land diffs with radar. +[14:18:36] in their code base. +[14:18:38] for +[14:18:41] We also recognize that parts of the code base. +[14:18:42] are very sensitive. +[14:18:46] And so for those we support code, self-serve code of project settings. +[14:18:47] to block radar. +[14:18:49] in those sensitive parts of the curvasis. +[14:18:54] So you have the ability to be really nuanced and fine-grained here. +[14:18:55] where we can. +[14:18:57] make sure that the most sensitive. +[14:18:58] areas and +[14:19:02] not going to have this lamp without review while making sure that +[14:19:06] the broader meta-developed population can benefit. +[14:19:07] from this. +[14:19:14] these expedited reviews and improved review efficiency. So look out for AI approvals on your own DIFs coming soon. +[14:19:20] All right, so that's my presentation on. +[14:19:23] doing in the +[14:19:28] to make a vault code review towards an AI native world. +[14:19:29] Uh +[14:19:31] And now I'm ready to take questions. +[14:19:43] I think Stacey you were going to. +[14:19:45] Sure questions for me. +[14:19:46] Bye James! +[14:19:47] Is it going to be alright? +[14:19:48] Yes. +[14:19:58] We did have one of the questions from the chat. It says, can we make code review a pre-hook for submitting GIFs, and can there be some signal in the fabricator that it was reviewed by DevMate code reviewer? +[14:20:02] Yeah, that's a great question. +[14:20:05] We actually do have signals in fabricator. +[14:20:09] When Demi-Coder-Dior reviews a diff, we publish. +[14:20:10] signals on the diff. +[14:20:11] Um +[14:20:19] that show that Demi could review RAM. And to make, I know that this is sometimes not as prominent as it could be. So we've also had support for +[14:20:23] diff transactions so if Devin Coderd you finds an issue on your diff +[14:20:28] You'll actually see that in the diff activity with a link to the signals. +[14:20:30] so you can see the specific. +[14:20:33] findings that they could have you produced on your device. +[14:20:37] In terms of making code review a pre-hug for submitting this. +[14:20:46] This is absolutely something that the fabricator team and their partners are absolutely actively talking about. +[14:20:47] 잘 오기 +[14:20:50] make sure that when authors are publishing diffs. +[14:20:52] that we have that. +[14:20:54] you know expectation. +[14:20:54] that you have. +[14:20:59] vetted the diff is ready for review and that it passes our muster when it comes to +[14:21:06] prechecks, whether that's, you know, then they coded you, you know, AI, slot quality checks. +[14:21:06] Um +[14:21:09] and other signals that we want to. +[14:21:10] in force up front. +[14:21:11] So... +[14:21:15] More to more to come there soon, but I would say that yes word +[14:21:17] moving towards an overall burden. +[14:21:19] we have an expectation that... +[14:21:22] discuss a certain quality bar before they go out. +[14:21:28] Thanks James, I have another question for you. +[14:21:34] How confident are we with AI code review for GIFs generated by AI and can they make the same mistakes? +[14:21:38] That's a great question. So. +[14:21:40] I would say from the +[14:21:47] from the signal address rates we've seen so far and from user sentiment. +[14:21:50] We do feel good about the issues that we do find. +[14:21:51] Um +[14:21:52] You know being accurate +[14:21:54] And... +[14:21:58] Uh.. +[14:21:58] Effective. Right. +[14:22:02] You know when we talk about air view on AI diffs +[14:22:04] Some people think, hey, if... +[14:22:07] model is reviewing code itself. +[14:22:08] produced. +[14:22:15] It's not going to find issues. That's actually not the case. We're finding that, you know, if you review the code that Claude produced with a Claude-based reviewer. +[14:22:21] we still find issues, right? It's about the different tasks being performed. And the other thing that really... +[14:22:26] makes AI review more powerful, as I mentioned, is the introduction of... +[14:22:28] context, right? +[14:22:33] local context from, you know, again, rules ranging from... +[14:22:37] and context files ranging from org level to team level to even individual level. +[14:22:39] That helps us. +[14:22:43] help our AI review systems better understand the part of the code base that they're operating in. +[14:22:45] We also have... +[14:22:48] at this point fairly mature. +[14:22:52] AI risk models that help us identify it. +[14:22:54] code that is higher or lower risk with. +[14:22:58] you know, very strong accuracy. And that helps us understand which. +[14:23:01] which of these dips need more or less attention. +[14:23:02] Um +[14:23:04] versus the... +[14:23:09] in terms of the trade off between the risk that they present to our reliability and our +[14:23:11] infrastructure versus. +[14:23:15] you know, the tolerance that we have for. +[14:23:16] letting certain changes move. +[14:23:18] along a bit more quickly. +[14:23:23] Thanks for that, James. +[14:23:25] We have one more. +[14:23:28] How is DRS being defined for GIFs? +[14:23:33] How is diff rescorping defined for diffs? +[14:23:37] Um, so... +[14:23:38] Diff riskor is a +[14:23:43] You know model that's been developed at the company based on historical +[14:23:50] dips, especially those that have caused seves and regressions at the company. +[14:23:51] full some training over time. +[14:23:53] has helped. +[14:23:59] produce a model that's very good at detecting the kinds of co-changes that are more likely to produce subs. Right? +[14:24:03] it's based on files that have previously experienced subs and many other features. +[14:24:06] There's a great group and team that own. +[14:24:09] the DRS platform and support groups that +[14:24:11] you can, um... +[14:24:14] reach out to to learn more about exactly how. +[14:24:15] DRS is defined. +[14:24:17] that's sort of a quick primer on it. +[14:24:25] Okay, I have another one. Since we are entering the era of AI reviewing AI. +[14:24:28] Where are the spaces left for human intelligence to shine? +[14:24:31] That's a great question. +[14:24:32] So +[14:24:35] I think you can think of it in terms of... +[14:24:36] kind of +[14:24:39] level of abstraction and level of scope, right? +[14:24:46] We're moving away from sort of the mechanical review of line by line code reviews and more towards +[14:24:48] having humans focus on. +[14:24:49] you know, being... +[14:24:55] builders and thinkers and architects, right? And so looking at co-changes again instead of. +[14:25:02] line by line more at a higher level, right? Looking at the intent of changes, the approach those changes are taking, the trade-offs. +[14:25:06] that a change is making in terms of its approach versus. +[14:25:09] the outcomes that it's trying to achieve. +[14:25:10] right and making +[14:25:12] decisions at that higher level. +[14:25:13] rather than focusing on. +[14:25:14] you know. +[14:25:18] more stylistic. +[14:25:20] and low level review. +[14:25:27] All right, James, thanks for that. I have a two-part question. +[14:25:38] One, how can we reduce the bottleneck of designated reviewer being assigned on diffs? And two, how can I customize a reviewer to comment similar to me to say things I might say? +[14:25:40] That's a fun one. +[14:25:41] Um... +[14:25:46] For designated reviewer, I think they might be talking about... +[14:25:49] Is that the change readiness score feature? And the... +[14:25:51] the +[14:25:53] the risk reviewers being added on higher risk gifts. +[14:26:00] To be frank, I know a little bit less about that particular feature, so I will have to defer that question. +[14:26:02] in terms of +[14:26:06] being able to build your own, you know. +[14:26:09] customer if you were that axe like you +[14:26:10] Um +[14:26:15] I think we are... +[14:26:22] you know, there are explorations at the company about the ability to fine tune models at the individual basis. +[14:26:23] where you could actually. +[14:26:27] potentially support an agent per engineer. +[14:26:29] I don't think we're there yet. +[14:26:31] But I think it's something that's, you know. +[14:26:31] birth. +[14:26:37] worth further exploration and you know let's keep talking about it. +[14:26:41] But for now, we're focusing more on making sure agents are equipped. +[14:26:44] but the context they need to review the code base effectively. +[14:26:47] Thanks for that. +[14:26:50] Can we force trigger radar if... +[14:26:54] Diff does not pass radar automatically, especially for simple diffs. +[14:27:00] So in the near future, we're hoping to make sure that radar runs on every... +[14:27:03] different version of the company that's eligible for it. Right? +[14:27:07] And so every diff that is eligible for radar should. +[14:27:09] you know, have a review outcome that... +[14:27:16] on it either passing or failing it so you shouldn't need to necessarily trigger radar directly. +[14:27:22] That being said, if there are situations where a radar fails unexpectedly, like in infrastructure exception. +[14:27:23] time out. +[14:27:24] I'll be +[14:27:32] may add a method by which you can rerun radar, just like you would rerun any failed signal on your diff. +[14:27:36] Awesome, thanks for that. +[14:27:39] Is it possible to track what diffs are reviewed by radar? +[14:27:41] compared to +[14:27:42] Uh... +[14:27:47] Sorry, there's a little bit of a typo here. Do we have DP showing that it helps? +[14:27:48] reducing SEDs. +[14:27:57] Great question. So in terms of tracking which diffs are reviewed by Radar, we have built dashboards that allow us to +[14:28:02] evaluate on a kind of an hourly rolling basis, which diffs have been +[14:28:04] approved by radar both. +[14:28:06] Again, human diffs and codemod buffers. +[14:28:11] and we support slicing those metrics by... +[14:28:13] different parts of the work hierarchy. So you can, I think. +[14:28:14] What did I need? +[14:28:17] directly level roll up all the way to the company line. +[14:28:21] performance of radar. We also have added +[14:28:25] support for diff search filters. So in a custom diff search, you can actually +[14:28:28] use the automated review filter. +[14:28:29] to +[14:28:35] find diffs that have been approved to skip deferred review is the review type of belief that were +[14:28:36] currently. +[14:28:37] using is the. +[14:28:43] For human diffs, and then there's something called AI commit eligibility or ACE that you can use to explore +[14:28:45] bot code mod gifts that have been approved by radar. +[14:28:46] Um +[14:28:51] So you can use that to tailor any custom diff search you want to explore these diffs. +[14:28:57] Awesome James. And sorry, I'll just like a part of the question right about the steps +[14:28:57] So. +[14:29:01] The important thing with the radar system is we want to make sure that +[14:29:04] We don't expect it to be perfect, right? There are going to be disks that. +[14:29:09] you know, concepts or reverts. And we've seen this with Demi did verification as well. What's really important is that. +[14:29:12] you know, the system is being very thoughtful about risk. +[14:29:13] and that we... +[14:29:18] want to make sure that we're performing as well or better than a human reviewer would. +[14:29:21] And so we're looking at, you know, in terms of +[14:29:21] 对 +[14:29:29] seven revert rates of disks that actually go through human review, we want to make sure that radar disks produce seves and reverts that look like... +[14:29:31] that or ideally lower rates than such tips. +[14:29:35] And that's what we've maintained in the end. +[14:29:35] Uh +[14:29:37] so far today. We see them. +[14:29:42] for disks that are that pass radar and land without a human review that land with radar approval. +[14:29:46] We have lower separates and lower river rates than. +[14:29:47] this that goes through. +[14:29:48] Um +[14:29:50] standard human review. +[14:29:59] Awesome, thanks James. I do want to be mindful we have one minute left. I'm going to ask one last one. What is the timeline for radar to go company-wide production? +[14:30:04] Oh, geez, that's a good question. Can I say? +[14:30:10] So I would say that we're working on announcements in the near term around radars. +[14:30:13] you know, company-wide eligibility. So keep... +[14:30:16] keep an ear and an eye out for those. +[14:30:18] We'll have more to share very soon. +[14:30:26] I think that's what we have time for. Thank you. +[14:30:28] Alright, thanks everyone. +[14:30:42] We'll just have a panel talking about some of what we found using it a whole bunch. +[14:30:46] over the past few months or years. +[14:30:52] and generally, you know, take questions from the audience about like, oh, I've been trying this. It's not working, etc. So... +[14:30:54] You know, lots of time for Q&A on this one. +[14:31:01] and feel free to ask them in the livestream and we'll forward them here and ask it to our panel. +[14:31:03] We'll introduce everybody. +[14:31:05] as we get there but +[14:31:09] I'm Alexander Ramirez. I'm an EM on the Facebook creators team. +[14:31:13] I've been working on better engineering for a long time as well. +[14:31:14] And so... +[14:31:18] you know, uh, RIN talk through again, these best practices and then we'll, we'll. +[14:31:20] I'll moderate the panel. +[14:31:28] The one I'm going to start with is expressing clear intense. So you heard a lot about this this morning around prompts and context, but. +[14:31:31] You know, one of the things that we've seen a lot that people... +[14:31:38] really struggle with is like actually giving the AI all of the information that it needs to be able to act on what you want it to do. +[14:31:38] Right, so avoid. +[14:31:41] Release short prompts, avoid. +[14:31:41] you know +[14:31:45] Uns— the the the the the the the the the the the the the the the the the the the +[14:31:49] a specified behavior, you know, in this case, you know, you want a notification that post goes viral, what does viral mean? +[14:31:51] You know +[14:31:51] when should they notify, et cetera. +[14:31:52] You know +[14:31:54] Best practice, talk about. +[14:31:57] 以上 +[14:31:57] Why? +[14:32:01] We're doing this what we're actually trying to achieve +[14:32:07] you know, how it should work, etc. This is, you know, a simplified example. You'd probably need to give it a lot more context than this, but... +[14:32:08] you know, the general point being that +[14:32:16] It's not about the precision of your words, right? I can add ums when we talk about voice typing. It's going to have a bunch of these extra soft words. +[14:32:18] but actually expressing the full idea. +[14:32:21] and the clear intent is going to be really important. +[14:32:22] Um +[14:32:25] The second one is using plans. +[14:32:26] So... +[14:32:29] you know, different tools are going to have different methods for doing this. +[14:32:33] But it's really important to avoid the kind of back and forth with AI that... +[14:32:36] can often happen as it's going down a rabbit hole. +[14:32:40] if you haven't expressed what you actually wanted to do. And so... +[14:32:41] You really want to avoid the... +[14:32:46] like back and forth. Oh, I said this, so what I said this, but I said this. +[14:32:51] create those clear plans. We do the same thing in our work. A lot of working with AI is like working with other people. +[14:32:56] You create project docs, you create PRDs, you create clear specifications. +[14:32:59] and you want to work with AI to come up with kind of similar. +[14:33:02] uh plans that you can then use for +[14:33:04] how exactly +[14:33:10] will implement and like build out your features. +[14:33:13] The third one is you want to +[14:33:17] Faltando lentamente。 +[14:33:18] Right, and so there's a bunch of upcoming talks where we'll talk more about how you will do this. +[14:33:20] but it's really important that you... +[14:33:24] are not the bottleneck, right? This is more about the human in the loop. +[14:33:26] And so if every time you make a change... +[14:33:28] AI is like, oh, can you go? +[14:33:31] test and see if it worked and you're like, oh, let me +[14:33:33] tell you that it was off by this amount. +[14:33:36] Now you're putting in maybe even more work. +[14:33:38] just sitting there kind of testing your changes. +[14:33:43] you know, in giving AI feedback versus the best practice of actually having AI. +[14:33:45] itself use the app. +[14:33:46] Use the browser. +[14:33:48] run test cases, whatever it is. And so. +[14:33:50] closing that loop and closing the loop. +[14:33:54] before you even start implementing, right? If it doesn't have the capability of doing that. +[14:33:57] then you'll want to build that capability first. +[14:33:59] The fourth is doing research. +[14:34:00] Right. +[14:34:06] Now that we have these powerful tools, it's really easy to jump in and just be like, oh, let me just go do this thing. Like. +[14:34:07] you know. +[14:34:10] without understanding how the code works, how... +[14:34:15] you know, our products are currently built in a complex system. +[14:34:17] like, you know, Facebook. +[14:34:23] That's really going to get you in a lot of trouble. You're going to put changes in the wrong place. You're going to make changes that don't make sense. You're going to duplicate things. +[14:34:26] And so the best practice is really to start with... +[14:34:28] some form of research. +[14:34:33] Again, there's different ways of getting research and we're not focused on which specific tool you use here. +[14:34:36] but actually go look how does the system work. +[14:34:36] 好 +[14:34:41] you know, are other people doing this today? Are there things I should be using, et cetera? And so... +[14:34:44] you know, making sure that you have that full understanding. +[14:34:46] before you go into even making a plan. +[14:34:48] is really helpful and +[14:34:56] you know, again, is where you can have a stop after this, where you have maybe a research document, right? In the same way that we might do an audit before we do some complicated work. +[14:34:59] You might have AI help you do that here. +[14:35:00] Um +[14:35:02] The fifth one is to keep a record. +[14:35:03] So... +[14:35:08] As AI is getting work done, particularly for bigger and more complex projects. +[14:35:16] It's really easy for you to run out of context. And as we talked about earlier, you need to be clearing the context. You need to be resetting it. You need to be getting a new context. +[14:35:20] But if every time you're having to like go explain again +[14:35:24] Here's what the project is, here's what we're trying to do, here's what we're adding to the steps. +[14:35:28] Now you're starting to lose maybe some of the important details and you're also +[14:35:29] a lot of work in yourself. +[14:35:31] So you kind of catch the AI up. +[14:35:32] where you're at. +[14:35:35] And so, you know, the best practice here is really about having... +[14:35:41] not just plans, but actually like milestones and steps that you can check off along the way. +[14:35:44] that AI can keep track of its own work. There's some. +[14:35:47] products as well that do a good job of keeping track. +[14:35:50] itself of how it's progressing and using +[14:35:52] agents in loops and so on which will +[14:35:54] covering later sessions, but. +[14:35:56] um having some sense of +[14:35:58] Here's the progress. Here's the step we're on. +[14:36:02] Here's all this context kind of saved for what you're doing is going to help you. +[14:36:04] up level your projects and do... +[14:36:08] you know, bigger, more complicated tasks than what you can do in a single session. +[14:36:10] And then lastly... +[14:36:13] It's really important you understand your work. +[14:36:22] Right? You know, the theme of a lot of these is that it's really easy to be lazy with AI. It's easy to just like, oh, just try and go do it. And then it, you know, fails and you get left with like, oh, well. +[14:36:24] you know, AI is just not good enough. +[14:36:25] Um, and... +[14:36:28] The easiest thing is just like, oh well. +[14:36:32] I'm sure the AI understands what I asked and what I wanted to do. +[14:36:34] But the reality is that a lot of times it doesn't. +[14:36:39] And so one of the best ways you can sort of gauge its understanding of your work. +[14:36:44] and you're asked and everything else is very to ask you questions, right? And prompt it, hey. +[14:36:46] You'll ask me a few questions. +[14:36:48] to make sure you understand it. +[14:36:48] You know. +[14:36:55] get telling it that it should be coming back to you if it doesn't understand A, B, or C, can be really helpful sort of. +[14:36:56] Tips to guide it. +[14:36:59] you know, a lot of these systems are designed for... +[14:37:04] especially now for automated running and running for long sessions. +[14:37:09] But letting it run for a long session where it's doing the wrong thing is really saving you any time. +[14:37:09] And so. +[14:37:13] You know, make sure it understands what you're trying to do. Read the plans. +[14:37:15] read the research, but also have it. +[14:37:16] ask you things. +[14:37:18] as, you know... +[14:37:20] we shifted our human on the loop. +[14:37:21] Wait that. +[14:37:24] And it went through quite quickly. +[14:37:28] But, you know, again, I want to make sure most of this time is devoted to questions, and so... +[14:37:36] I'll give a minute for each of the panelists to introduce themselves and talk a little bit about how they use AI and their work. +[14:37:38] And then we'll just get straight into questions. +[14:37:40] Josh, do you want to take this off? +[14:37:44] I'm Josh from Messenger. +[14:37:53] I've been using AI almost exclusively since October. I haven't written a real one to code myself since probably November, December. +[14:37:58] and I'm using it to solve a whole bunch of problems in testing and operations and on-call work. +[14:38:00] And also other... +[14:38:02] random parts of the code base I see. +[14:38:08] As I find problems, I like to go and have a try to solve them for modernizing our iOS code base, for example. +[14:38:10] Um, yeah, everywhere. +[14:38:17] I'm ending from Peaceful Creation. +[14:38:20] So I've been using AI pretty much. +[14:38:23] exclusively starting January. +[14:38:26] I would say like a first prank I've made. +[14:38:30] Fine.. +[14:38:33] that means I guess code. I couldn't find much success with it. I feel like once I switched to cloud code. +[14:38:43] things have gotten significantly better. Also, I've just read up a lot more on how to manage context, how to use planning, all these various tricks. +[14:38:46] and yeah, been able to land. +[14:38:49] like bigger stacks of divs, more complicated features. +[14:38:50] after that switch +[14:38:55] Yes +[14:39:05] Hi, I'm Sam. I'm on the Facebook metrics team and I have been using AI for all 2025 and +[14:39:08] basically trying to work on closing the loop. +[14:39:17] on business and product level planning to actual feature development and end-to-end testing and +[14:39:20] I try and use... +[14:39:26] as many different air tools as I can and push the limitations to understand kind of what's possible and what's coming next. +[14:39:29] Awesome. +[14:39:40] So we'll stop sharing so you can see our faces a bit and we'll jump straight to the questions. I think, you know, I'm seeing some of these and they're quite interesting. +[14:39:41] Um +[14:39:43] you know, rather than start with anything... +[14:39:50] to background. I'm gonna start with the top one that I think is a really good one, which is how do you think about when you create +[14:39:53] new custom agents or tasks you want to accomplish? +[14:39:56] versus using maybe default. +[14:39:59] skills or the default AI, right? +[14:40:00] How do you... +[14:40:06] think about going into a task if it's going to require something specialized or if it's something that it can do. +[14:40:07] you know, on its own. +[14:40:14] I can take a step at that. +[14:40:23] In my experience, I haven't had that many opportunities to have custom sub-agents. I think that there's room for sub-agents when we need agents to act. +[14:40:27] and have a thing that can reason on their own. +[14:40:32] So like there's reasons to have like an oncology, for example, where that agent is doing a thing for us and acting. +[14:40:33] with reasoning. +[14:40:38] As part of just the day-to-day life of being an engineer, writing code, or getting code written at least. +[14:40:43] The only times that I've actually had to write a sub agent or had Claude write a sub agent for me +[14:40:46] was when I wanted to avoid contacts. Uh. +[14:40:55] because if you're trying to, for example, you're in a loop where you have a clock do something and then you're also in the clock validated and go back and forth and try to iterate in the loop. +[14:41:01] Sometimes there's leakage where whatever reasoning qualities are off the bat to produce the artifact. +[14:41:08] then leaks into the validation so we'll validate based on the assumptions and information and how of the time that it... +[14:41:08] Get the thing? +[14:41:15] So you won't actually validate fully because club might get lazy and assume that they're directing the first place. +[14:41:17] That's the place for having a custom sub agent. +[14:41:21] is nice because the Customs.agent has its own contracts window that isn't polluted. +[14:41:23] with what you've been doing the rest of the session. +[14:41:24] That's the only place I've been. +[14:41:28] found it useful to have my own one data there anyways. +[14:41:33] To add to it, I feel like the new Asian team is +[14:41:34] pretty good at basically. +[14:41:36] Greenpoint, Josh said. +[14:41:41] You can have like an influencer agent, a validator agent, a researcher agent. +[14:41:47] and it kind of manages its own context. So you don't pollute your orchestrator, you don't pollute the implementer. +[14:41:49] Yeah. +[14:41:52] And I think just to add to that, like... +[14:41:53] that +[14:41:59] Next programming language is English. I think we all know that. So whether you write it as a custom subject or... +[14:42:03] markdown file scale in the end. +[14:42:08] just having a record of what that agent is supposed to do. +[14:42:12] I think is the most important so it can be reused by you in the future. +[14:42:15] Yes +[14:42:17] Yeah, and maybe the quick summary to that is like... +[14:42:20] Mostly you don't need custom agents for things and +[14:42:23] you should try it without one and then see if you need it. +[14:42:24] versus the other way around. +[14:42:27] maybe it's kind of a good default start for most people. +[14:42:31] And then obviously if you're reusing it for a specialized case then... +[14:42:35] Yeah, that might be something where you want something more specific. +[14:42:36] Um +[14:42:37] I think. +[14:42:42] There's another question about how we distinguish sort of research from planning. +[14:42:45] and how to think about the difference between the two. +[14:42:48] you know, particularly there are different like skills for +[14:42:54] researching and planning, but also just even conceptually how you think about... +[14:42:58] When do you transition from okay now I have the context to write the plan +[14:43:01] Okay, now the plane is good enough and moving on from there. +[14:43:08] Or do you think of those as more of a morphous kind of one step at the start before you get to execution? +[14:43:14] I think I think so that I figure research is like +[14:43:27] free technical implementation requirements and more business product level, what tools are available to actually solve this and what are the capabilities of the different tools. +[14:43:29] depending on the complexity of the task. +[14:43:35] Sometimes I don't do a research implementation or if I actually know what the solution is. +[14:43:40] the technical solution is then I just ask the agent to go directly towards planning. +[14:43:43] Yeah, a little slightly. +[14:43:48] different take, at least for more complicated tasks that +[14:43:49] I feel like multiple classes. +[14:43:52] I think research is almost mandatory. +[14:43:55] Like basically you need to hold. +[14:43:56] the whole +[14:43:57] cost that to be +[14:43:58] examined. +[14:44:00] by T. +[14:44:03] AI so that like it doesn't skip a step. +[14:44:04] I'm like basic PG. +[14:44:09] I didn't get your input or the plan and the outcome. It's just like a big path. +[14:44:14] you can miss any code in between because then I guess I feel like I'm hallucinating and jump steps +[14:44:16] So I find that like research. +[14:44:17] three-century light. +[14:44:20] having the entire flow was like very helpful. +[14:44:27] At least for me, I think it's more amorphous. +[14:44:30] Yeah, sometimes we'll just talk the clock and then that. +[14:44:34] talking to cloud or like explaining this is the problem that I'm looking at this is what I'm trying to deal with +[14:44:36] That's more of a research thing. +[14:44:37] and I'm at the end of that. +[14:44:48] there will be some point where I just say, you know, let's write a formal plan and that formal plan is something that I have committed into the code base. So it's like the plan of record that we'll actually use to walk you in track and do the thing. +[14:44:49] But even at that point... +[14:44:55] it's not done because often times the plan won't be perfect, it won't be complete, and you have to continue to iterate and discuss on it. +[14:44:57] So I don't necessarily think that there's. +[14:45:01] separate things except what you have a plan artifact that is ultimately produced as text. +[14:45:04] and everything else before that is just research. +[14:45:15] Yeah, and I think to some degree this is also different per person. It kind of depends how you do your work, right? We all do different kind of work and when we go into our work. +[14:45:18] you know, some people will have really detailed technical +[14:45:21] docs that help line exactly what they're doing, some people will have... +[14:45:22] you know higher level. +[14:45:26] And so to some degree, you want to start with maybe nearing the way that you're +[14:45:27] crimes. +[14:45:28] Um +[14:45:29] work. +[14:45:32] like how you how you do your current work. +[14:45:37] And then from there, you can sort of build up your own kind of, you want to, on how you want to do this. +[14:45:40] There was also a follow-up question on how do we... +[14:45:45] store these plans and like where we put them, maybe just a quick, you know. +[14:45:47] I can just say, all my plans are in Google Drive. +[14:45:51] that I'm mounting on all my dev servers and so. +[14:45:53] They all get stored with like a... +[14:45:55] These +[14:45:56] plus what I'm working on. +[14:45:59] I don't know if you guys have other ways of like storing. +[14:46:05] Yeah, using NullSrip.org, also seeing people just like put it in the code base. +[14:46:06] So I think. +[14:46:07] Both works. +[14:46:15] Thank you very much. +[14:46:17] We have another question on context. +[14:46:22] management and how you sort of manage the like state of knowledge on your claw. I'm curious for you guys like... +[14:46:24] You know, water. +[14:46:25] Maybe +[14:46:29] like the things you're starting with, right? The skills, the MCPs, your clot MD. +[14:46:35] And then also how you manage that as, you know, to optimize it over time, make sure it's not. +[14:46:37] missing stuff but also not too big. +[14:46:40] you know, what techniques do you guys use for... +[14:46:42] Making it work well for you. +[14:46:52] I frankly have no idea what's in my quad .md file. I'm a quad manager, is that fully on me? +[14:46:57] Usually the process will be if I ask it to do something and do something that I don't expect or... +[14:47:06] did something weird. I'll ask it to reflect on why it did that and then make some amendment to QuadDMD or whatever the right place to be. +[14:47:13] to address it going forward. As far as skills, and I think the skill marketplace is really helpful, there are some really helpful skills like there's a... +[14:47:20] hook, for example, to notify you whenever a quad session is done. I think that that is a very useful one. So I can go back and forth between like five different quads and I just. +[14:47:22] find out when one of them actually needs my attention. +[14:47:26] There's one for other skills for dealing with CI and dealing with data. +[14:47:28] Lots of different ones. +[14:47:29] Till next. +[14:47:34] Engineer is definitely a useful one. Highly recommend that. It just pops up in a lot of different places. +[14:47:39] But the skill is marketplace and just sort by what is useful and retrieve them. Alright, sort by. +[14:47:42] how often they're used and re-through them is a pretty good place to start. +[14:47:53] There are bootstrapping sets that I'm sure we could come up with that are useful for everyone and probably we should have a FBY one that is generally available or generally recommended. I know we haven't thought about that yet though. +[14:47:55] We have that coming tomorrow. +[14:47:56] Stop. +[14:48:00] Yeah, so at least for me. +[14:48:05] I started with deep research, like I usually would generate a +[14:48:10] if I have an architecture like marked down already like in social reports, I have to have to read that. +[14:48:12] And then kind of explain what I'm... +[14:48:17] trying to do within like this code base because what I found at least my personal experiences. +[14:48:21] without this and letting it write for its own code, sometimes you miss that. +[14:48:22] Uh... +[14:48:26] So after that, then I use 10x engineer to kind of like come up with a plan. +[14:48:28] And what was it? +[14:48:33] a lot. If anything, I have slight suspicion that it has to be explained further. +[14:48:35] So I do find sometimes I... +[14:48:36] with the eyes, kinda like... +[14:48:39] They really go engineer but kind of lazy at times +[14:48:44] So it'll get to something that it feels comfortable with and then just skim through the rest +[14:48:45] So. +[14:48:48] basically anything I've liked. +[14:48:52] has a 10% certainty I would just ask you to explain and then it'll +[14:48:54] kind of stuff check and then spot over what he missed. +[14:48:58] and then we'll just keep on reiterating that until I get to a good plan. +[14:49:01] Um. +[14:49:11] I got my thank you to I first of all think it's very interesting that all three of us seem to have different perspectives on this which I think it just means like a lot on +[14:49:18] out-personize this is and fitting into your workflow and building your own like context management system for me I think. +[14:49:23] kind of most important thing is my personal rules file. I wanna make sure that that's like. +[14:49:33] uh good and giving it the right directions and I do actually spend some amount of time reading through the thinking traces as I ask prompts and I see what +[14:49:43] Do I know that the agent is missing and wasting time on and then I try and either update my personal file or I try in. +[14:49:53] Insert skill files that give it the relevant context in the project that I'm working on, such that it automatically picks up that information and wastes less time. +[14:49:56] trying to. +[14:49:57] resolve. +[14:49:59] Whatever unknown. +[14:50:00] It happens. +[14:50:05] Yeah, just one quick thing to add so we're going to produce a output at you. +[14:50:06] Don't expect. +[14:50:07] you can sometimes just like. +[14:50:10] Do I retro with it? Do you see? +[14:50:14] Like this is the stuff I want but this is what you're getting, like what went wrong? +[14:50:18] and then can kind of like have cloud one like up a strong and the or like +[14:50:22] maybe you're missing some information, right? So next time you know to give it to them. +[14:50:25] Yeah +[14:50:33] Yeah, and I think to Sam's point, a lot of this is really personal for what work you're doing and what you're finding works for you. You know, I... +[14:50:39] find that my natural state is to install lots of things all the time and to add tons of new capabilities. +[14:50:43] I get really annoyed at Clawhead or Dev Made in life. +[14:50:45] helped to add a lot to my rules and so. +[14:50:48] I have to go periodically clean it up, but I'm also just like... +[14:50:50] somewhat of a... +[14:50:52] minimalist person and so I like. +[14:50:54] I am fluctuating. +[14:51:00] you know, set some rules and skills and then sometimes I just wipe the whole thing and say like okay what can I do from scratch now? +[14:51:01] Um +[14:51:07] But I think you, to some degree, have to reflect. And I think that's maybe the key for a lot of this stuff, is you have to. +[14:51:08] Try it. +[14:51:11] and reflect and see where it's working and not working and then +[14:51:12] put in some effort to... +[14:51:14] customize it to what? +[14:51:15] to what you need it to do. +[14:51:16] Um +[14:51:19] There's a lot of good questions in here. I will... +[14:51:19] You know. +[14:51:22] Keep, hold some of them. +[14:51:25] As I know we have later sessions on like orchestration and... +[14:51:29] you know, building, you know, things on the loop, etc. +[14:51:31] Like one thing that maybe would be helpful to talk about. +[14:51:34] is as we are getting +[14:51:41] lots of new skills and plugins and features and all of this. And there's like a proliferation of tools. +[14:51:42] you know, there were some questions around like... +[14:51:47] Why? Clock code or DevMate or DevMate with clock code or Confucius or other things. +[14:51:48] How are you guys? +[14:51:54] up to date on what is coming out. +[14:51:56] And then also, like, what are the things you are like... +[14:52:00] doing to try new things or to... +[14:52:03] maybe hold off and say like actually my stuff is working well. +[14:52:05] You know, I don't need to switch it. +[14:52:11] I think there's some people who are kind of overwhelmed by the number of posts coming out every day with new... +[14:52:13] New things to try. +[14:52:17] I declare bankruptcy. +[14:52:19] I can't keep track of it anymore, I'm giving up. +[14:52:25] You know, like workplace will show me the ones that get the most reactions from the product community or other places. +[14:52:29] Cloud Code versus dot-mate. +[14:52:31] I strongly prefer quality code than I have for a while. +[14:52:35] and I don't know that I will keep in mind about that at any point in time. +[14:52:47] Do you think that there's a lot of room for having a more standardized set of things that are generally available? But at least for me, my workflow is very much a, I have problems I'm trying to solve. I try to work with cloud to solve them. And if I can't, or if it's too slow or too inefficient. +[14:52:52] At that point, I start to work with quality to understand why and what parts of the loop we can speed up. +[14:52:56] And sometimes that will mean called do research and seeing what other skills are. +[14:52:59] hooks or plugins might be available that we can use too. +[14:53:05] Oh, so really just word of mouth and second brain has been very helpful and very interesting +[14:53:07] what pops up on workplace and beyond that. +[14:53:08] Fuck. +[14:53:12] I don't have a systematic way of doing it. I mean, there's too much stuff coming up. +[14:53:15] for anybody to practically keep it up. +[14:53:19] And honestly, I don't know that you should because not all of them are of the same quality or the same utility. +[14:53:23] to what you're actually trying to do and having new tools for us making new tools. +[14:53:26] might be interesting, but at least for me, that's not really what I... +[14:53:29] like to do, like to find a tool that will solve a particular problem. +[14:53:33] Yeah, pretty similar to one. +[14:53:35] Josh said, um, he was just like. +[14:53:37] If I feel like there's steps to that. +[14:53:41] can be automated or can be taken over by AI. +[14:53:44] I may look for that tool. If it's there, then like... +[14:53:44] Try it out. +[14:53:48] Like it basically just like app pieces. +[14:53:49] too late. +[14:53:55] current flow, right? Like I want to run over and I, oh, like I search for that post and like, you know, I can do overnight stuff. +[14:53:56] I want you on the link. +[14:54:00] have like better memory right for like I want to +[14:54:08] When I boot up a new OV, I can make sure everything is there. Like, that I saw open claw, I saw meta claw, and then I tried that. +[14:54:10] 。 +[14:54:13] whatever the gap is, see if there's a tool already in the marketplace or there's a post word. +[14:54:14] and just +[14:54:19] One tip is to always question whether this can be automated. +[14:54:21] or can you take over by AI? +[14:54:25] Typically the answer is yes. So it just kind of go from there and error it. +[14:54:29] Yeah, thank you. +[14:54:30] I +[14:54:40] would say to keep up. I spend some time on threads and Twitter. There's actually a lot of AI researchers and people. +[14:54:46] on there and there's some pretty good content. That's one way that I stay up to date. +[14:54:54] And also I know this isn't the most useful answer, but I do spend time outside of work on personal vibe coding projects. +[14:54:56] And there is a difference between using the tools. +[14:54:59] inside meta and outside meta. +[14:55:11] just, you know, we have a lot of proprietary systems and I think there's things that you can learn, you know, building whatever you want, even building, you know, like messing with Open Claw and seeing what's... +[14:55:13] what's out there and +[14:55:14] Um... +[14:55:20] Because everything is moving so fast, I think part of our job is becoming to be constantly learning. +[14:55:23] Sound +[14:55:25] Yeah, no, it's a lot. +[14:55:28] and certainly way too much for any person to keep up with. +[14:55:29] Um +[14:55:33] It's also where you can ask agents say, give me updates for like. +[14:55:35] stuff that's been interesting. +[14:55:39] you know, is worth looking into, but largely that's just going to look at likes anyway. +[14:55:41] So at least for me it's been... +[14:55:43] trending on workplace plus. +[14:55:46] one or two newsletters and then... +[14:55:48] You know, just people that I... +[14:55:50] talk to who are trying stuff and interested. +[14:55:52] you know, making sure that, you know, in our... +[14:55:54] meetings we're sharing. +[14:55:56] what we're using, what we're finding helpful. +[14:56:01] you know, I'm particularly looking for anything that's like somebody actually telling me I'm using this and it's helping me. +[14:56:03] is kind of the baseline starting point. +[14:56:05] But yeah, a lot of times it's just... +[14:56:06] You need a book. +[14:56:09] You need to have the problem to look for the solution, not the other way around. +[14:56:10] Um +[14:56:14] One other question, and this may be the last main question. +[14:56:15] that we have time for. +[14:56:20] And again, we'll follow up in the comments with a bunch more. +[14:56:22] answers as there are a lot of questions coming in. +[14:56:27] But how do you guys think about working in new code bases that you don't? +[14:56:28] have a lot of experience with. +[14:56:32] you know, that's kind of one of the unlocks from AI, but at the same time... +[14:56:36] you know, how do you validate its work? How do you get enough context to feel comfortable working there? +[14:56:41] you know, is this something that even in the future we should be doing or we should be +[14:56:44] you know, looking to stay more within our own sphere of work. +[14:56:48] the most important thing about the +[14:56:50] Awesome, Matt." Yes, the +[14:56:57] I think the should question the answer is certainly not any part of the upside of AI is that we're going to be able to solve bigger and bigger problems. +[14:57:00] and those problems necessarily are going to... +[14:57:03] expand the types of things that were comfortable looking at. +[14:57:07] I think if we only try to focus on the problems that are currently in our domain. +[14:57:13] we're going to max out what we can do pretty quickly when there's a whole unexplored area of problems just all out there. +[14:57:14] For the how? +[14:57:20] I think it's a place where we actually have to rely very heavily on CI and we have to aggressively +[14:57:23] Leverage AI to build up RCI capabilities. +[14:57:37] I think there's too many agents going on, there's too many changes happening and there's any changes being done by people who have no idea what they're really doing and what they're really working on or how to validate. They might try to validate based on how they would validate things before. +[14:57:40] But that's not necessarily how people that have been working on a code base want it to be validating. +[14:57:41] So. +[14:57:48] leveraging CI, you know, putting a lot of rules and context and skills in place for a different code basis so that it's clear. +[14:57:52] What to do? I mean, agents will follow it, right? If you have within your particular part of the universe. +[14:57:55] skill it says to submit a diff. +[14:57:56] in this curve base. +[14:58:03] the test plan is supposed to look like x, y, and z. Agents will tend to follow that guidance and use x, y, and z appropriately. +[14:58:07] So I rely heavily on the tooling. I think that's the answer. +[14:58:09] for learning about the Cubase. +[14:58:11] Yeah. +[14:58:11] Research. +[14:58:12] Just ask. +[14:58:14] I want to do X. +[14:58:16] Can you help me understand how X works? +[14:58:19] Yes +[14:58:21] I think one... +[14:58:23] it's a lot easier to do to uh +[14:58:26] kind of research a new code area. +[14:58:28] I'm not sure if... +[14:58:31] You'll have enough context soon +[14:58:34] like we work out a big feature but like the +[14:58:37] You can just still go there and give you a lot of good information. +[14:58:41] So even for debugging issues, that'll give you a huge head start. +[14:58:42] Thank you. +[14:58:43] I do think... +[14:58:44] Like. +[14:58:50] It used to be a company-wide effort where every team would basically produce these architectural docs like Markdown files. +[14:58:51] Uh +[14:58:52] that would help. +[14:58:52] Like. +[14:58:57] basically solve this case right where someone who has no contact sports on a new code base. +[14:58:59] But the agent has all the context of these. +[14:59:01] to understand the problem. +[14:59:05] And lastly, I think validation is going to be hugely important. +[14:59:06] So +[14:59:10] I think what I'm like, I just know this, when I'm trying to do the same later on. +[14:59:12] when we're this always gone out. +[14:59:16] So people are understanding this, but I think we are really the bottleneck. +[14:59:19] how many invaders there are to do the validation. So maybe like. +[14:59:21] more investments to go there where like +[14:59:22] Yeah, I was just like... +[14:59:25] Do their thing but have a really good validation set. +[14:59:35] Those are running on time. I'd yes, absolutely do that, but check with the SMEs before landing your diffs. +[14:59:41] Yeah, exactly. As we still work with people, there are still experts in the area, you know. +[14:59:42] You say I? +[14:59:43] Pretty good, dip up. +[14:59:47] and then go talk to people, right? They're happy if you come with a good diff that you really thought through. +[14:59:48] Yeah. +[14:59:50] you know, I think. +[14:59:53] We still value expertise here and I +[15:00:06] you know, happy to see that that being part of our best practices. So with that, thank you everyone for your questions. We're going to follow up in the comments. There's a lot of very interesting topics that we'll get into, some of which we'll get into later this week. +[15:00:09] So for some of those, you might point you to those sessions, but otherwise... +[15:00:14] Move on to much. +[15:02:07] Bye. +[15:02:27] Okay, this is me, Adolfo again. +[15:02:31] So this is your... +[15:02:32] request. +[15:02:34] summarize the +[15:02:35] made in so far. +[15:02:37] This is the first part of the day. +[15:02:41] Go for a Jarvis. +[15:04:53] Yes, as the learnings and the lessons +[15:04:54] on a file. +[15:04:55] for today. +[15:05:02] Yes, thor insan.. +[15:05:04] Yes, Jarvis. +[15:06:01] Sounds good, I'm gonna go have lunch and I'm gonna start whatever the new session starts. It's just gonna start recording, all right? +[15:06:03] Thanks. +[15:13:07] Yeah. +[15:19:03] the +[15:19:04] I don't understand because of what's going on. +[15:19:05] There's the back. +[15:19:23] the most amazing thing about the most amazing thing about the most amazing thing about the +[15:19:29] the +[15:19:43] And also... +[15:19:47] It's built and helps a lot to prevent contact fraud. +[15:19:48] Um +[15:19:49] the only load. +[15:19:50] That's me. +[15:19:52] And um... +[15:19:54] Yeah, so... +[15:19:56] Next slide please. +[15:20:04] There are essentially three types of skills based on where they are located. First one is skills and they are located in your +[15:20:12] home orders so you don't really have to come in with a repository. It's only business law to you. +[15:20:15] but also apply for all the projects. +[15:20:19] And there's projects that still are basically located in the... +[15:20:22] for that specific locations. +[15:20:25] anyone in that project when they're... +[15:20:30] working on certain files in that location. +[15:20:37] the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important thing the most important +[15:20:38] And also, the team shares feels which is based on the... +[15:20:40] and also like a top level off the front frame. +[15:20:43] where anyone and men will be able to... +[15:20:46] 인과학사업 스킬을 +[15:20:51] And we need to be kind of careful about the triggering logic when... +[15:20:54] für die Vaniralsdispir纳。 +[15:20:58] Next let these +[15:21:02] So in general, the scale and the hand-to-hand like this, we have +[15:21:04] at the very top. +[15:21:06] It's a yaml front matter. +[15:21:08] that he finds some... +[15:21:10] I'm going to try to outball the scales. +[15:21:14] That's where actually the Batman well will see. +[15:21:16] We're understanding about the skills. +[15:21:21] what it does and what's considered as relevant. +[15:21:25] And there's also some other field that we find in the truing logic. +[15:21:31] the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about +[15:21:33] for your prompt, also location based matching. +[15:21:34] Um... +[15:21:35] Yeah. +[15:21:38] The second part is the combat gathering for Wish. +[15:21:44] What information does that mean first? And that's where I want to find some... +[15:21:50] the most important thing about the most important skill the most important skill the most important skill the most important skill the most important skill the most important skill the most important skill +[15:21:51] And then, uh, if a skill can come paint... +[15:21:53] multiple workflows. +[15:21:55] you can have one. +[15:21:59] Over the fence the +[15:22:00] Useful flow is a step by step process. +[15:22:00] Um +[15:22:02] telling the +[15:22:04] That's when it works. +[15:22:07] how you actually will +[15:22:10] do it in front, do his task by himself. +[15:22:15] And you also want to specify the exact output. +[15:22:17] what the results would look like. +[15:22:20] Umm.. +[15:22:20] Yeah, next please. +[15:22:30] So here is what the YAML front matter was. Basically the description is something required. +[15:22:35] You have to have a clear definition of what videos are about. +[15:22:37] And... +[15:22:42] But it also recommend to have a following view like that. Bye, guys. +[15:22:43] Obviously. +[15:22:46] and you can kind of tell that. +[15:22:48] That's me too. +[15:22:53] future buy certain past patterns。 +[15:22:56] You can also tell it to tree grounds, let's say we use our keywords, for example. +[15:22:59] We're defining some marketplace search scales. +[15:23:02] and with a commerce search, turn the place search. +[15:23:06] the most important thing about the +[15:23:12] And you can also define tools that can be used in this. +[15:23:18] the most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most +[15:23:19] You can even add a GX8 to control the discoverability. +[15:23:23] the track speaks the most +[15:23:29] So, to develop the multi-state skills, we actually conduct... +[15:23:33] a thorough survey by asking my mate to analyze. +[15:23:39] that has like 1000 thieves that our team has submitted. +[15:23:41] And then we... +[15:23:46] This will divide that into different subcategories. +[15:23:49] where areas based on the public pass. +[15:23:50] Um +[15:23:54] From there we can define small smaller areas. +[15:23:57] and the architecture I'll follow is... +[15:24:02] We have a top level kind of general and it's our scale which +[15:24:05] helps to tackle. +[15:24:07] more general tasks. +[15:24:08] and some common local levels. +[15:24:11] the +[15:24:14] And it kind of referenced to some knowledge center. +[15:24:16] files which defines +[15:24:20] basically all the code has our teams that's been working on. +[15:24:24] some additional relevant knowledge. +[15:24:27] We're too long now, our teams... +[15:24:30] Add and tell us the cloud for example +[15:24:34] how the search for a way it flows into our backhand. +[15:24:37] And how was your life? +[15:24:40] Yes, let the heavens.. +[15:24:45] what the set the query in the the +[15:24:47] search request. And then we also have other team members helping to build this +[15:24:51] define the specific scales at high close. +[15:24:54] very detailed workflow or path. +[15:24:57] is something we are still +[15:24:59] I know. +[15:25:04] you right now。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 Yes。 +[15:25:09] the +[15:25:15] the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important +[15:25:17] So we've covered a couple of things, I'll just have to do the other news again and go through it quickly. +[15:25:25] showing the same in the escope. So the main object that we have is just past team specific with tribal knowledge. +[15:25:26] through the agent. +[15:25:28] the +[15:25:30] So this can also be seen as a main tool just +[15:25:31] on how we like it. +[15:25:34] Similarly, how we onboard a new B-number. +[15:25:38] Bitte Bitte +[15:25:39] coming to steps to basically add a musical. +[15:25:43] And it is the simplest way especially just to ask to create a niche in RAM. +[15:25:45] On +[15:25:48] Then last part mentioned, use the apply tool against the tool. +[15:25:51] to ensure that this skill is over here. +[15:25:54] added to context if it's relevant. +[15:25:59] Yes, the health should optimize the continent." +[15:26:05] and something very meta specific is to explain the workflow, different score production experiment on our users. +[15:26:06] this can be shown you say. +[15:26:07] Um +[15:26:08] 끝 +[15:26:11] Plastic table or disabled for different clothes. +[15:26:12] And... +[15:26:13] Um... +[15:26:17] Another important thing is to just give examples of +[15:26:19] Oh +[15:26:20] Oh +[15:26:27] the most trust done in the.. Go. +[15:26:27] in the near past and based on how they were like resolved. +[15:26:32] So there's some people. +[15:26:34] So, pitfall is off. +[15:26:35] which +[15:26:37] which agent. +[15:26:41] But they can be used overusing a skill, for example, if you want to know that. +[15:26:42] agents. +[15:26:45] a very plonk towards hallucination. +[15:26:47] We can use scales to reduce the amount of hands. +[15:26:50] For example, you can limit the scope. +[15:26:53] Bye. +[15:26:56] before the change to be the only inside a specific directory of mine. +[15:27:00] We can give profits like in show known you and get a preview unless actually +[15:27:03] the most important thing about the +[15:27:05] We can use similar problems to just use classic. +[15:27:07] We can also take care of. +[15:27:11] the +[15:27:15] on the privacy side, as the agent won't know that logging of user A line is about logs is +[15:27:17] The network also we can just add that. +[15:27:24] the afternoon let's present the food you lent to the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the food you lent the +[15:27:29] And similarly it's very important to have invariance up front on things which will not change. For example the basic overflow. +[15:27:37] the most variable names the most +[15:27:39] the most in the most in the most in the most in the most in the most in the most in the most +[15:27:40] on. +[15:27:44] Okay so we are quickly going to be a store and work a couple of things. +[15:27:46] and we will look at the... +[15:27:48] at the scale format. +[15:27:51] And then we are actually solving... +[15:27:53] a task which is +[15:27:55] which is basically a bug in the MPE search. +[15:27:57] Let's talk. +[15:27:58] So yeah, so the first one. +[15:27:59] I +[15:28:01] As you mentioned before, obviously... +[15:28:03] Uh... +[15:28:05] I'm sorry if it's not visible but +[15:28:12] Over the most simply prompted the most +[15:28:12] create a new skill file for a file which +[15:28:19] the dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[15:28:21] And it is able to read through similar skill files, obviously the file given over here. And. +[15:28:24] It took a lot of merit and it gave us the +[15:28:25] Peace. +[15:28:34] And it basically gave us the basic skill file. +[15:28:40] Over here, we start our description on God and apply to the X which is very helpful to God. +[15:28:41] Uh... +[15:28:46] to know if we're going to add this to the context window or not. +[15:28:51] After those are things which I guess the specific engineer who is like +[15:28:54] who knows a lot about this would be a good add. +[15:28:56] And all of this... +[15:28:57] Um +[15:28:58] I think I know it needs to be +[15:29:01] I have listed a nice time to be good by the... +[15:29:05] with a person paving the step and the reviews as well. +[15:29:09] the most important thing about the +[15:29:10] So basically +[15:29:17] We have a format which probably makes it easier for the agent to understand the code base. +[15:29:19] and format for us to play easy. +[15:29:25] the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important +[15:29:26] in comment asked, we can add a fast. +[15:29:30] which were done before we started using AI and short tips. +[15:29:33] on how we are going to solve it and include the... +[15:29:36] access to our area with a given task as well. +[15:29:38] And... +[15:29:40] Yeah, similarly, I guess. +[15:29:43] I guess the other sections are pretty. +[15:29:46] self-explanatory but it requires a lot of +[15:29:52] the dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[15:29:54] So now coming to the next slide, this is actually solving a problem. +[15:29:58] the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the most trust the +[15:29:59] So we will be using trowel over here. I actually... +[15:30:05] I'll leave the home page to you. That's taken three minutes for us to solve the issues. +[15:30:08] So look, the game over here is too... +[15:30:10] Firstly, mock. +[15:30:12] Wow, the system without... +[15:30:14] skills and try to solve it. +[15:30:17] and then use the skills and stop the same thing. +[15:30:18] そう +[15:30:21] Yeah, this is basically the problem with the unit. +[15:30:23] It's um... +[15:30:26] It's basically an NH case in the system. +[15:30:28] It's basically... +[15:30:30] a change provided at the final stage of our interview. +[15:30:31] Off I... +[15:30:33] and you're bending and rocking. +[15:30:36] so basically without the skill +[15:30:37] I'm about to end. +[15:30:40] that was kind of, it says things like... +[15:30:47] the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important +[15:30:48] find the correct... +[15:30:50] the right the right the right +[15:30:52] 4.1 will make changes. +[15:30:54] But.. +[15:30:55] and it's basically asking for prompt. +[15:30:59] I'm sure it will be able to solve the task given most problems but... +[15:31:05] I think that's what we are trying to optimize over the day. The reduced amount of time you will have to talk about. +[15:31:08] with the agent should have already been knowing these things. +[15:31:11] And the second one gets to the opposite of the simulation. +[15:31:19] Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. Yes you can use the skill fry the most possible way up. +[15:31:20] so we can see that it uses the skill file it +[15:31:24] Yeah, it found a port pointer. +[15:31:26] and it made a change with... +[15:31:31] which helps me to be dependent on the world, whereas it is in the golden age. +[15:31:32] But it doesn't matter, it's awesome. +[15:31:37] It also adds a test case which probably is the most +[15:31:38] இந்திடி. +[15:31:40] in the given +[15:31:43] till I will be torn into +[15:31:52] I mean, test case is a RU test case, if there is no coverage for it. So it basically added the test case as a prompting for it. +[15:31:55] I think the only thing left to do is... +[15:31:57] me is just to submit this as a test. +[15:32:02] Yeah. +[15:32:08] So similarly, I'll get them to the other screen. +[15:32:11] अगर देखान्दान्पल अपावुटीन +[15:32:15] the most possible nobies flow for +[15:32:15] production users +[15:32:21] And I guess for future directions, we simply need to create more scales for the files. +[15:32:24] of the specific look. I think that we are also looking into +[15:32:27] of your auto-breaking scales. +[15:32:32] So that scales for auto-evolve as more and more days go in. +[15:32:39] the lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils lentils +[15:32:41] They don't need to do better activities or they first need to do better way of using. +[15:32:50] آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آنو آ +[15:32:51] Do we have time for questions? No we don't. So let's take a look. +[15:32:52] Yeah. +[15:32:58] Thank you. I'll also keep it short. So here. +[15:33:03] We talked about skills and skills as kind of a tribal knowledge. +[15:33:03] Um +[15:33:13] One of the main other main use cases of skills that we are using in dating is basically to automate any kind of repetitive workflows that we do. +[15:33:14] Um +[15:33:17] as MLAs or software engineers. +[15:33:27] These skills can be really helpful for small to medium or even large stars workflows. For example, when... +[15:33:28] you +[15:33:50] training and you want to automate that so the skills can do that perfectly. +[15:33:53] If you go to the next slide. +[15:33:54] Yeah. +[15:34:03] Here I just wanted to share some of the high-level knowledge and kind of like best practices that I've learned in the past couple of months. +[15:34:04] I am building skills. +[15:34:13] One main thing is that when we want to build a skill, it's actually better to use the AI to build this skill, and actually there's another skill too. +[15:34:16] I hope you write this, it's called the Scale Creator. +[15:34:19] You can use this to skill creator. +[15:34:27] to create any kind of a skill and it has all of the proper information in terms of how you have to format your skill, how to format your files. +[15:34:36] in order for AI to be able to browse these skills and go to different folders. So definitely recommend that to use that. +[15:34:37] Um... +[15:34:41] make sure so one of one of the +[15:34:43] key important things is that +[15:34:45] We don't want to have... +[15:34:49] the +[15:35:03] It's called that empty +[15:35:17] And if you're a skill that needs too big, then that means that your context window is gonna be filled up really, very quickly. The proper way to do this is actually create different folders in your skill. +[15:35:35] especially for more complicated ones. +[15:35:40] skills can basically go and lock out the specific... +[15:35:41] part for that skill. +[15:35:43] Thank you. +[15:35:50] Yeah, and then the other thing is basically we need to have like a clear trigger condition. +[15:36:01] So the AI coding tool that you're using knows when to use that skill. Skills are meant to be more of a workflow automation. So we need to have a step by step. +[15:36:19] don't overload the context window at once. +[15:36:23] Ja, ik heb een quote die... +[15:36:29] Next slide here, just an example of how we use this code in dating. +[15:36:44] One of the main things that we do as part of like a dating and Signals team is we build features and one repetitive part of that is basically a doing feature engineering authoring the feature bringing it +[15:36:47] the +[15:36:56] to RAS. So we have already previous steps that we have done and we also have a lot of making page exactly explaining how to do this. +[15:37:05] So how we did this is basically we provided all of this information that we had before to the skill creator, the skill, so that +[15:37:09] It can build a skill that you can fully automate the entire process. +[15:37:15] And the result is basically a scope that now we use on a daily basis to trigger. +[15:37:34] increasing the execution quality but also the speed of offering features. +[15:37:36] Um... +[15:37:40] I think that is their next slide for me. +[15:37:41] Yeah. +[15:37:53] The one last thing here is that also skills can use tools, which is just one of the cool things that they do, which makes it more kind of like an agentic behavior. +[15:37:54] Um +[15:38:13] some of the material as config and then use some part of that. +[15:38:15] and add that to another file. +[15:38:21] So by just giving it the bash tool, it can automatically run our code for us. +[15:38:24] look at that specific file and follow the instruction. +[15:38:31] Yeah, I think that's all in terms of the skills that we wanted to talk about. +[15:38:33] I think we are right at time so. +[15:38:35] I think we can move on. +[15:38:44] I think in interest of time maybe we should just keep going and people can ask questions. +[15:38:48] to Ireland, Sapa and Ayush. +[15:38:48] Um +[15:38:52] Next, maybe we should move on. +[15:38:59] I was also impressed by the quality of the slides themselves. The content is generally always good but the... +[15:39:04] Right, like, you know, it is beautiful, right? Like, what I started is so... +[15:39:07] I don't know how much AI went into this, but... +[15:39:13] Definitely lent you the lent you liked the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent you lent the lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent lent +[15:39:22] the most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most +[15:39:22] I think we are going to present the short code usage. +[15:39:24] Yeah, thanks so. +[15:39:34] Hello everyone, this is Jan. We are from the growth team. We want to quickly present some learnings and +[15:39:39] best practices for using out code. +[15:39:47] By the way, I use both cloud code and the DevMate. They are both quite powerful. +[15:39:56] The reason is if one tool is dumb, you can always switch to the other ones, but you don't have to. +[15:39:59] wait for one scene to recover. +[15:40:00] Um... +[15:40:01] So... +[15:40:12] Yeah, let's get started. We will, some of us will present some demos on how we use Cloud Coach to +[15:40:13] uh... +[15:40:22] to help observe the marketplace growth work and sex Territory for healthy and sharing some learnings to do +[15:40:25] Okay, so let's get started. +[15:40:32] This is the agenda. We don't have time to cover everything, but feel free to take a look. +[15:40:33] Uh... +[15:40:45] because it has multiple tools that we could use by a code code, which can help us to accelerate data work. For example, +[15:40:48] how to use +[15:40:53] cloud codes, CLI, and other companies. +[15:41:02] the terminal command line. Also, we recently explored Maya Cloud, which is quite similar to the open cloud ideas. +[15:41:05] which also can integrate with code code. +[15:41:17] Another thing is how we use Google Doc to always mount our most recent contacts for each session so you don't have to worry about the session you love. +[15:41:19] from on demand. +[15:41:26] and let's see if we have time to cover everything. And we also have some use kits. +[15:41:28] cases from our team. +[15:41:29] We will work with you, sir. +[15:41:31] Um... +[15:41:33] Yeah, so +[15:41:37] To start Cloud Code, there are so many ways, different ways. +[15:41:43] But I listed some links here, feel free to explore by yourself. +[15:41:54] There are some popular ways. For example, you can always come to your VS code and use your command line to open the... +[15:42:06] cloud with just simply typing the cloud. And this is some other ways, for example, you can also use it in the +[15:42:11] the +[15:42:11] It's not super good today. +[15:42:16] It's not responding from the damage. I don't know why, but... +[15:42:20] But you can see if you use the terminals, there are... +[15:42:24] It's quite straightforward and you can +[15:42:25] You some... +[15:42:35] shortcuts or keywords to figure out what's happening on your cloud code. For example, what plugins, what skills have you... +[15:42:38] installed especially there are some talks +[15:42:44] very popular skills installed by the Alexa Justice by +[15:42:47] the +[15:42:51] a cloud template, which I think you can find it. +[15:42:57] So yeah, you can bunny all the +[15:43:03] cloud templates where you can find the most popular downloaded skills. +[15:43:08] and plug-ins, hooks, all these things. +[15:43:13] We can also contribute to add our own. +[15:43:17] marketplace skills and plugins to this. +[15:43:20] So, yeah. +[15:43:28] I think I added too many. Maybe I should delete some of the senses getting so slow. There are some... +[15:43:29] tips. +[15:43:36] to help you maintain the to help you keep as much free space as possible to help you +[15:43:41] on the conversation since every session we have 200k tokens. +[15:43:42] Um, yeah. +[15:43:47] So there are many good... +[15:43:48] Uh... +[15:43:56] and materials that shared by the company. Feel free to check out when you have time. Make sure you always. +[15:44:02] have the most recent version of Cloud added by using this command. +[15:44:09] There are some tips on using Cloud Code 2. For example, +[15:44:15] Always make sure that your cloud.md has the best context. +[15:44:18] and make sure you give very specific prompts. +[15:44:25] Make sure you use plan mode first. For example here you can +[15:44:28] the +[15:44:29] shift tab +[15:44:42] Yeah, I just change this to the plan mode and you can change it back to the execution mode. So it's always suggest to plan first and then... +[15:44:47] execute or allow it to run the code and create the code. +[15:44:54] and make sure that you also run the, you execute by yourself. +[15:44:56] not assume everything. +[15:45:05] work correctly and make sure you use the keywords of how hard they want to think about. +[15:45:08] So this will help to accelerate things. +[15:45:16] Some essential shortcuts you can also find from the links that I shared above. +[15:45:18] Okay, let's move on. +[15:45:21] Some more reasons. +[15:45:32] updates for cloud codes maybe worth the time from the recent exploration. Like my cloud, people probably heard of it. +[15:45:36] There was my cloud and also... +[15:45:39] Next I'll cut off the +[15:45:40] Publish then same day. +[15:45:48] which definitely shows the popularity of how people want to explore the open cloud which also is popular in the industry. +[15:45:50] So... +[15:45:55] I personally explore my cloud which is +[15:45:56] I think. +[15:46:01] Quite cool. So this is my club. You can see that I set it up +[15:46:04] the +[15:46:05] but quite... +[15:46:10] smoothly, actually if there are also issues with it. +[15:46:11] All right. +[15:46:20] Okay, sorry, I just kept it straightforward. So you have to use Dev Server for the MyCloud tool. There are some benefits of using MyCloud, which is... +[15:46:22] you can always +[15:46:26] have a persistent memory for your uh... +[15:46:27] a contest. +[15:46:32] However, people sometimes also worry about the ones they staff server. +[15:46:33] pause or +[15:46:41] about the reboots, right? So people also still use some Google Docs to keep mounting these contents. +[15:46:51] I will wish I would talk shortly, but my cloud actually which connected to the desktop will actually help you to handle it. +[15:46:51] So I... +[15:47:05] By the way, if you haven't read the dive server, you should try it from the dive server tool on the thumbnail. And the recent read the dive server are so popular that there are some... +[15:47:11] large depth servers are running out of probability. So. +[15:47:18] You can start from the small one and we can see if the company will solve this issue later on. +[15:47:23] So after the next day to the cloud server, you can talk with your cloud. +[15:47:25] on Google Chat. +[15:47:33] and give it a name and wake it up and help to set up some tasks. +[15:47:36] the +[15:47:47] I will, for this is one example that I was trying using my cloud to try one of the skills built by our team, it's Yue who are working on a very big stack of QE tasks. +[15:47:50] and it actually runs... +[15:47:58] Wow, but definitely need to talk with him and with your cloud to make it working with you +[15:48:04] My claw was powered by a claw code so if my claw is done you can use claw code itself to +[15:48:07] die Bang anfix." +[15:48:16] This is my cloud. You can definitely adjust the command to start it. +[15:48:18] fighting the +[15:48:23] My cloud and it's wrong at your Google chat +[15:48:30] But recently it also has some issues like too many people using it, it's sometimes breathing out. +[15:48:34] So, but I believe this will be some features that will be quite helpful. +[15:48:42] Next, we'll quickly go through how another popular way to go to use to +[15:48:45] keep your contacts always +[15:48:47] store safe. +[15:48:53] This is, we can always mount our contacts to the Google Doc. +[15:49:05] Thanks, Carrie, for sharing some of your workshop scripts on how to install. Basically, these are quite straightforward. You can use this comment on your... +[15:49:11] on your cloud CRI, like terminal here. +[15:49:22] So I just, I already installed this. So every time when I start up my code, I can just run the update context. +[15:49:23] All right. +[15:49:25] It will... +[15:49:31] 好。。 +[15:49:36] I need to set up this. But these are some previous running term notes that I learned this optics context. +[15:49:42] What it will do is it will go to my Google Drive that it's set up. +[15:49:46] of it created during the setup. +[15:49:49] For example, this is my Google Drive. +[15:49:51] Any folders? It will... +[15:49:56] In default, create this AI area where you can store your contacts in it. +[15:50:04] It will you can customize your products. I created two products for my data work if you can also +[15:50:07] Create Worklog Everyday. +[15:50:17] This was my yesterday. I'll be today. Yeah, work a lot. You can always store it and make sure you don't lose the context. +[15:50:23] and either the on-demand or the desktop. +[15:50:30] At the same time, your clock can also upset... +[15:50:32] that obsess to your... +[15:50:33] Google July. +[15:50:36] to ramp up on this contact. +[15:50:42] There are a few tasks, like steps, to do. For example, it will check your... +[15:50:44] Oh, I read a new morning. We'll. +[15:50:48] go to app, it's a context, it will go through the... +[15:50:50] what the fence poles the +[15:50:50] Reading about us. +[15:50:55] Google chat message about us deep stress about us +[15:51:04] and summarize what you need to do for today. And you can just keep on talking with your cloud. +[15:51:06] code and finish your task. +[15:51:18] Here it's a little bit hard to do the live demo using my flow to access both the Google Docs. +[15:51:21] and the 12-code, but essentially these three things are... +[15:51:26] can't access the same contact status you created. +[15:51:31] which can all start on your Dev server and your... +[15:51:33] Gólu kraw, eh, gólu, tak. +[15:51:37] It's just different than how many backups you want to have. +[15:51:41] and to keep your context. +[15:51:42] Yeah. +[15:51:45] Um, okay. +[15:51:46] Uh, yeah. +[15:51:48] definitely make sure you try this. +[15:51:57] in this way and it will be very interesting and see if it works for you and if you have any questions feel free to. +[15:52:01] asking how we're like a... +[15:52:05] It's your experts in the team or results in the team. +[15:52:05] for Terry so we can help. +[15:52:08] Um, yeah. +[15:52:13] Another thing is there is definitely a multi-agent concept. +[15:52:14] Um... +[15:52:18] So you can achieve these concepts from many ways. For example, you can use your... +[15:52:24] a cloud coast terminal here. For example, you can ask the... +[15:52:29] the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the +[15:52:42] Yes please。 +[15:52:47] There are also some fancy usage like in the micro cloud you can... +[15:52:50] create multiple clouds to cover each other. +[15:52:57] But the most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most most +[15:52:57] also take some time to make it work. +[15:52:59] I was +[15:53:03] based on my personal experience you should definitely try this but if it doesn't work +[15:53:05] I should put that this. +[15:53:11] These tools are changing day by day. The community is contributing to the tools. +[15:53:18] Like either other day, the API will be changed. So if it doesn't work, try it back the other day. +[15:53:20] It may work, just it will work well. +[15:53:22] Okay. +[15:53:23] Um... +[15:53:27] Some tips is that always make sure. +[15:53:31] You draw yourselves your own verification. +[15:53:38] Don't crave too very weak prompts. Be more specific as possible as in order. +[15:53:49] The data, as well as the previous demos, already shows that if you specifically cause some skills or plug-ins, it will have more specific good results. +[15:53:52] We can demo it shortly. +[15:53:56] But something in the future worth trying back worth trying +[15:53:59] after the tool become more mature. +[15:54:04] For example, we can build our 24-hour... +[15:54:07] AI assistant on our Dev Server. +[15:54:11] For example, you can already ask your cloud. +[15:54:15] 。 +[15:54:18] chat that, hey, I help me to make sure I make a diff every day. +[15:54:21] something like that. I know that sounds creepy, but... +[15:54:24] You can also always try these things. +[15:54:28] be proactive in mind-hunting. +[15:54:33] make sure that you give them the tasks that you want to build. +[15:54:43] Next, we will do some quick demos for the cloud code that scales the team views for marketplace. +[15:54:45] And I hope you want to take over. +[15:54:47] I'm going to take over. +[15:54:48] Um, yeah. +[15:54:51] And now, this is on my screen. +[15:54:52] Thank you so much for walking us through. +[15:54:55] basics of cloud code skills +[15:54:56] and I'll... +[15:54:58] Discuss this. +[15:55:01] quality specific skill that I built. +[15:55:05] to help streamline existing quality. +[15:55:09] the most important thing about the +[15:55:09] work stream processes and also since we're +[15:55:14] We've expedited exponentially the velocity at which we're making changes. +[15:55:17] It's essentially very crucial that we... +[15:55:21] consider the recommendation quality and implications. +[15:55:24] Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes +[15:55:24] alongside like development process. +[15:55:26] So this skill is done. +[15:55:27] Basically helps. +[15:55:33] validate all the quality considerations on any new change that we apply. +[15:55:39] So the skill design, we built it with having context about the existing workflows. +[15:55:39] So it's... +[15:55:40] as +[15:55:45] especially a lot of context about how we build quality filters. +[15:55:48] And with this, there's a lot of considerations about. +[15:55:53] the impression loss with filtering out low quality products but also +[15:55:54] How that. +[15:55:56] can negatively impact. +[15:55:57] Um +[15:56:01] user experience of showing low quality products. +[15:56:01] And then... +[15:56:07] It also involves a lot of data analysis, so we have a bunch of tables that store. +[15:56:09] the filter products. +[15:56:11] Impression laws, EDRs. +[15:56:23] All of that context is given to the skill. And then we built a five step process where anytime we want to add a new quality filter or a quality intervention, it will help identify the root cause. +[15:56:25] Do a deep research. +[15:56:26] Um... +[15:56:28] invoke the MCP and +[15:56:31] deploy a bunch of different agents and. +[15:56:34] do D-Pace switch about the root cause of the quality concern. +[15:56:37] Do some preliminary analytics to see whether it +[15:56:41] the filter that we're trying to apply or whatever change we're trying to apply doesn't necessarily. +[15:56:43] We did not do any products. +[15:56:46] And then it implements the solution and also. +[15:56:47] as it does then. +[15:56:50] So this is the first mode of using the skin and then the second one. +[15:56:55] which is probably more powerful is that for every new marketplace change. +[15:56:58] we can have a quick sanity check in terms of quality. +[15:57:01] and the skill will help track any quality concerns. +[15:57:08] So yeah, we can access this from the cloud templates in the plugin here. So you'd essentially just want to. +[15:57:10] Run this in your terminal in code code and um +[15:57:12] Listen then. +[15:57:16] at the marketplace quality skill to your skills context. +[15:57:18] Yes. +[15:57:22] I walk us through two specific +[15:57:28] examples of how we can use the skill and how it's benefited me personally in my daily tasks. +[15:57:34] So we're running this new test to expand marketplace notifications to the rest of the world. +[15:57:37] and right now we don't have any specific region. +[15:57:38] Um +[15:57:40] Wise quality filters. +[15:57:43] And this is really important to consider because... +[15:57:46] Yes, I.. +[15:57:47] lots of expected discrepancies. +[15:57:49] Струтф.. +[15:57:51] still essentially will help us identify some. +[15:57:56] Not all, of course it's not perfect yet, but it'll help us identify some of the... +[15:57:58] Major gaps. +[15:58:02] like we can see that the price range filter right now is very US. +[15:58:03] Die ist specifik. +[15:58:05] and then language especially. +[15:58:11] the most important thing about the most important thing about the most important thing about +[15:58:14] as default. And then yeah, it'll just give you a summary of all of this. It also helps create... +[15:58:15] Dance? +[15:58:17] about. +[15:58:17] Like. +[15:58:20] the gaps and the next steps. +[15:58:23] So this is something that would have taken a human. +[15:58:26] Multiple list, одей! +[15:58:28] to evaluate these gaps but... +[15:58:30] the cloud skill was able to achieve this. +[15:58:32] in less than 10 minutes. +[15:58:34] And then second thing was... +[15:58:39] Like Viran mentioned, we've been working aggressively on 5-2-A and +[15:58:41] We've been building a lot of filters. +[15:58:43] So an example is category filter. +[15:58:49] Here, the skill will help us analyze the CTR of each of the categories and then propose solutions. +[15:58:53] which categories are lower CTA guard and can be blocked. +[15:58:55] to improve. +[15:58:58] like quality and also +[15:58:59] Thanks. +[15:59:00] engagement metrics. +[15:59:01] Yeah, so... +[15:59:05] Right now it's definitely at the skill has a lot of. +[15:59:06] improvement. +[15:59:07] But... +[15:59:11] We do want to ensure that this can be used for any general changes. +[15:59:12] So. +[15:59:14] like if you're making a ranking model change. +[15:59:16] We want to make sure that... +[15:59:18] we're not shifting towards any. +[15:59:19] new content that has. +[15:59:20] BAM! +[15:59:21] And then... +[15:59:27] When we're having a new retrieval source, we want to make sure that we're applying the necessary quality filters to it. +[15:59:29] And similarly for any UI changes. +[15:59:33] Yeah, like marketplace team, please give this scale a try and +[15:59:36] I appreciate any feedback. +[15:59:39] so that we can continue iterating the skill. It's of course a process. +[15:59:43] Yep. Oh, thanks. +[15:59:45] the night." the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night the night +[15:59:55] That is one example from what our team done. I have another example which just like what I showed on my +[16:00:06] My class, so our team, Team Xie, has built a skill called Marketplace Faithway QP Etherread skill. +[16:00:13] It's a g5 state here which is just at the skills with helping team to +[16:00:18] the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important thing about the most important +[16:00:22] QE. This QE is really huge. It has 20... +[16:00:28] active test running and it's really kind of, and we are iterating. +[16:00:29] twice +[16:00:34] two times, kicking off two new batch of tests on average per week. +[16:00:37] So, wei zhen +[16:00:41] Come up with these skills and +[16:00:50] combine it with Google Docs. So for example, I think we was putting all the +[16:00:54] the +[16:00:54] QE setup in +[16:00:58] Have a seat my little traffic. +[16:01:00] My claw history, one second. +[16:01:05] the +[16:01:06] Yeah, yeah, yeah. So... +[16:01:11] So this is actually what we... +[16:01:16] So, this is where I cut this... +[16:01:17] Uh... +[16:01:23] cloud skills and this is the input where we just tell them the +[16:01:27] Can you set up of these so many arms? You can see. +[16:01:33] Wow, there are really many arms. I believe we cannot do this without the clock code. +[16:01:38] and asking that, hey, can you basically, this skill is that, hey, +[16:01:56] to give us give me Chris this +[16:02:05] Google Doc of this analyzed result. You can see that it is pretty good format of the outcome. +[16:02:08] even shows you that, hey, which arm is talking to me better? +[16:02:16] which we should keep on iterating. I think that skill step four is to propose new arms. +[16:02:25] and kicking off these new arms. I believe you can always use either my cloud or the cloud. +[16:02:29] cloud code here to +[16:02:32] execute on this +[16:02:37] You know, you walk these scales or plugins. +[16:02:40] So, um... +[16:02:41] Thank you. +[16:02:51] Yeah, so the last demo is how we use Cloud Code to do more deep dive. +[16:02:55] I believe that not only... +[16:02:56] Junior. +[16:03:07] and the senior engineers need to think about how to make our impact with the BAST. +[16:03:11] I'll compromise. I believe that's showing... +[16:03:22] showing impact you need to explore those new ideas from other code base or other products. So for one of the examples we are doing... +[16:03:24] But there's TVets in this. +[16:03:31] One is the CLI app that you run on your on-demand or your local Mac. That's like the service of the Tengling's client. +[16:03:33] Booted up, parked down bath. +[16:03:37] it boots this thing up and that's the thing that communicates to your simulator at the moment. +[16:03:41] Dev Infra has built tunneling so that if you were +[16:03:45] booting up from your on demand, you can talk to your Sim that's on your local Mac. +[16:03:47] And back. +[16:03:48] That's what +[16:03:49] I think Haku is. +[16:03:52] And eventually we're going to work on... +[16:03:56] Iowa's device support, the reason we don't have it at the moment. +[16:03:59] is because Apple does not expose... +[16:04:00] a +[16:04:01] You know. +[16:04:04] API for talking straight to an iOS device. +[16:04:08] for security reasons. There are lots of open source libraries who have +[16:04:10] reverse engineered. +[16:04:12] how this works. +[16:04:23] And it tends to break or it's a little shaky. So in the past, we've heard about IDB, which is what we used with Flipper to connect between devices and our lockup support for it. +[16:04:27] That was one of the reasons why it was abandoned. It was just people got tired of... +[16:04:28] packing it so. +[16:04:31] We're going to revisit that because a lot of tools need it. +[16:04:34] That's what the moment Dump App is only talking to. +[16:04:35] your steps. +[16:04:36] device work in progress. +[16:04:38] on the phone. +[16:04:40] You right. +[16:04:41] Swift. +[16:04:45] and that is where you kind of register what your command name is. +[16:04:46] and +[16:04:51] CLI parts are kind of like art parts or clap or whatever your language or choice. +[16:04:56] argument parser is and then you implement a function to return some JSON. +[16:04:58] Based on the arguments, pretty simple. +[16:05:07] And with the XPlugin architecture, you're able to design plugins just for your app, or you can make a generic one that everybody can use and every app can have their own. +[16:05:09] mix and match combo of plugins. +[16:05:17] This is kind of what the CLI looks like. You remember Dumbab. +[16:05:18] the Dash Dash app. +[16:05:24] is helpful if you are running multiple sessions. You know, you can have maybe uf facebook and anntario on your sim. +[16:05:28] If you only have one running, I think by default it just picks it up. +[16:05:32] on the defcs for command and then you can see you pass additional +[16:05:36] the most。 +[16:05:40] and define and get back to JSON. And the agents able to read it, they like the structured output. +[16:05:45] This is what the Swift would look like. +[16:05:56] I know that some folks over on the NAV team are working on deep links or navigating certain screens just by saying, you know, from dump app, you know, like, navigate to groups. +[16:05:57] just like take a shot. +[16:06:04] main screen or to an example screen from the intent library. So we use Swift's argument parser library. +[16:06:06] which is what that struct is in the middle. +[16:06:08] And... +[16:06:10] Yeah, it's basically like any CLI you would rate. +[16:06:12] anywhere else and then +[16:06:13] At the bottom. +[16:06:13] get the type. +[16:06:17] typed parse arguments and static func execute. +[16:06:21] and you return some kind of object that can be encoded to JSON. +[16:06:23] That's pretty much it. +[16:06:24] It's supposed to be simple. +[16:06:26] Notice it doesn't have state. +[16:06:28] You do that on the side. +[16:06:29] Thank you. +[16:06:32] we can kind of get into that name sample. +[16:06:35] Okay, we're ready to hear it. +[16:06:35] So. +[16:06:42] Let's build a plugin or let's just go see how it works. I'm going to switch off my screen share. I'm going to do my old desktop +[16:06:52] All right, so I got an on-demand boot it up. +[16:06:57] I've got a SIM. This is a test account, so don't worry. This is not private user data. +[16:07:01] If you want to just talk to your Sim, you know, I'm on my on-demand here. +[16:07:03] Let's get this over a little bit. +[16:07:06] I can just run arc. +[16:07:14] al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al insan al +[16:07:14] Let's do it again. So, you... +[16:07:16] Usually you'll have your agent. +[16:07:19] Keep doing this, but okay as you can see it's booting up. +[16:07:21] I'm on my on demand. +[16:07:28] in my phone you can see this is also an iPhone 16. Cool, so... +[16:07:31] Tape help for available commands. +[16:07:36] Great. You can see some plugins that people are even building. +[16:07:38] Um... +[16:07:39] Okay, let's just do +[16:07:40] you higher team. +[16:07:41] Let's see how it looks. +[16:07:48] Whoa, okay, that looks like a thing, but you can see you get accessibility identifiers. +[16:07:58] heights, views, all that kind of good stuff. So let's open up Claude and let's tell Claude to start interacting with it and I'll show you how you can kind of piece together. +[16:08:01] new interactions kind of thing. So... +[16:08:07] You just dump that. +[16:08:13] I should use whisper for this. Sorry. This is the AI talk. I shouldn't have done that. +[16:08:14] use dump app too. +[16:08:16] Get the view hierarchy. +[16:08:20] hierarchy. +[16:08:22] and render it in ASCII. +[16:08:28] So the idea here is that +[16:08:29] Um... +[16:08:32] It was boot up dumb bab. +[16:08:40] hierarchy, it gets the coordinates, and then it kind of draws something that looks similar to. +[16:08:42] Facebook app. +[16:08:45] And so while this is truning. +[16:08:47] Katie, are there any questions yet? +[16:08:52] Yes, we have a few questions. So the first question is... +[16:08:56] Is there an Android equivalent or is this only iOS? +[16:08:59] Yeah, there is an Android equivalent. +[16:09:04] I will defer to some of my Android cohort probably in the chat, Ron, Rick. +[16:09:06] That's a round. +[16:09:07] Um +[16:09:10] Amy Street dumpouts have been around for a while. +[16:09:12] on Android is. +[16:09:15] much more tied to the ecosystem. I think it uses ADV. +[16:09:20] we built IDB on iOS to kind of be like a replica of ADB. +[16:09:26] I was the landscape is just different. So any who yes, I believe it's on the Android. +[16:09:28] maybe dust it off, you know +[16:09:32] I remember when I searched for it the wiki I keep finding is for Steffo. +[16:09:35] for mobile config plugins for it but I didn't find the wiki itself. +[16:09:36] But yeah. +[16:09:42] Awesome. Thanks, Dustin. We have a few other questions. +[16:09:45] The next question is, is dump-up supported on Mac? +[16:09:50] Dump up! +[16:09:50] is supported on Mac. Somebody... +[16:09:54] uh put up a diff for that this weekend which was kind of cool +[16:09:54] So yeah. +[16:10:00] I will link to it after this chat is over, but or somebody wants to go do a commit search. +[16:10:02] Find the diff where people... +[16:10:04] Added the Mac OS X. +[16:10:08] SDK to the DumbApp libraries and FBFC libraries. +[16:10:09] Let it dumb bat. +[16:10:13] If you find that diff, you'll find whoever did it. And you can think of it. +[16:10:18] Alright, the next question in the comment section is... +[16:10:23] What is it recommended Sam to use with Dump App, especially for Android Dev? +[16:10:30] You can use whatever sin you want for an Android, nail club, I'm an iOS guy. +[16:10:39] The next question is, does the dumpout have to be running previous to prompting quad? +[16:10:44] I was gonna say no, but this doesn't look like it's doing too well. +[16:10:47] How about this? I had it open on another tab. Let me... +[16:10:48] I mean, it should. +[16:10:51] something to be a hierarchy. +[16:10:53] Let's try here. +[16:11:05] This looks like it's gonna go better. Okay, there we go. +[16:11:20] Okay, so it's getting the accessibility tree, which I know somebody added recently. So one of the things we were finding is that like the payload was huge. +[16:11:27] Especially if you have a pretty complex view hierarchy. So the successability mode, I think is 80% or 90% smaller. +[16:11:28] Bye. +[16:11:30] Okay, cool. Look, there we go. +[16:11:40] And then you can do things like... +[16:11:49] Tell me about that. Memory usage. +[16:11:55] I'm assuming there's a plug-in for this at this point. I think it's called status. Let's see if I figure it up. There we go for +[16:12:07] neat +[16:12:18] Okay, that looks like a thing that I would trust. Okay. +[16:12:25] I'm probably not going to build a plug-in right now just with how my luck is going with the prompt. I want to sit here making y'all watch me +[16:12:30] I'm going to bounce back into my slides really quick. +[16:12:33] I'll share this again now. +[16:12:42] So. +[16:12:45] Here are some success stories about people who had been using .app. +[16:12:46] iOS. +[16:12:48] and to hold a friend. +[16:12:50] basically got +[16:12:53] their entire flow for like editing a video. +[16:12:59] You know, they can like crop and do all this stuff. You can go read the post. There's a dumpout for iOS group. +[16:13:01] I shared his post and there is +[16:13:02] Awesome. +[16:13:04] Um... +[16:13:07] screen recording from WhatsApp. +[16:13:09] screenshot testing. +[16:13:12] It ships in threads. +[16:13:14] Facebook Messenger Instagram. +[16:13:17] Messenger, or I said Messenger. Basically everybody's got it. +[16:13:21] It's really exciting to see the new plugins that people are making. +[16:13:22] Yeah, I done. +[16:13:23] and the vector is on Mac. +[16:13:25] So... +[16:13:27] So what's next? +[16:13:28] as I mentioned before. +[16:13:32] We got to solve the connecting to an iOS device. +[16:13:36] I know the Blocks Lab wants this, the UIE Companion wants this. +[16:13:41] We want this, so I wrote up an RSC about it and um... +[16:13:49] Basically like right after I wrote it and I passed it to a different agent it was just like yeah this is like a security nightmare and I was like you're right. +[16:13:57] But my latest shower thought is that the new Happy CLI has a really cool flow, you know, where you like generate a QR code and then you scan it from... +[16:13:59] your device and then you're kind of like off. +[16:14:06] encrypted definitely needs to be encrypted if you're going to be passing any kind of like user data +[16:14:07] that seems like a good flow to start. +[16:14:10] So I'm going to go investigate that. +[16:14:15] We talked about integrating it with Metastee LI so that makes the commands more discoverable. +[16:14:17] Um +[16:14:20] We've got lots of plans for plugins. +[16:14:21] And, uh... +[16:14:24] I think that's enough to keep us busy. +[16:14:25] Yeah, that's pretty much it. +[16:14:31] Last thing, as I mentioned, that FBFC library meta dump app is where you can go. +[16:14:33] to get +[16:14:34] Start it. +[16:14:37] That's also where you can go find that death about the Mac support. +[16:14:40] Our dump app is your one stop call. +[16:14:42] that any tool can use to connect. +[16:14:47] Thanks, Manny, for refactoring that last week. We've got a workplace group. +[16:14:48] There's my little M skill. +[16:14:49] Got an on call. +[16:14:51] If you want to help out. +[16:14:53] Dips and PRs are welcome. +[16:14:55] And uh... +[16:14:57] That's it, so... +[16:15:03] I mean, we've got extra time, so, Kate, if there's any questions, I'm happy to answer them. Otherwise, I'll just... +[16:15:05] Y'all got 15 minutes back after lunch. +[16:15:12] Thanks Dustin. Yeah we actually have quite a few questions in the comment section. So the next question is +[16:15:17] When should we use flipper versus dump app versus agentic UI? +[16:15:22] That's a great question. I am going to steal this straight from Mani Deep. +[16:15:23] He has a great... +[16:15:25] talk about articulating this. +[16:15:27] The way I like to think about it +[16:15:28] is. +[16:15:30] Dump app is good. +[16:15:40] because it basically communicates like a snapshot. So there are some things that are suited to like streaming data. So for instance, if you were saying, +[16:15:47] You know, like there are things in Flipper that are really useful for like recording all the GraphQL queries that are coming in and out of the app. +[16:15:48] Okay. +[16:15:54] It's really nice to have the visualization. It's really nice to have the bars. It's really nice to be able to click into the periods and see things. +[16:15:59] Dump app is more about snapshotting and so if you wanted to replicate that in dump app you would have to like +[16:16:04] Create a single tenor session-sculpt object. You recording. +[16:16:09] preeminently or you would have to send a command to start recording. +[16:16:12] and then send another community to say stop reporting and then dump the stats. +[16:16:16] You could theoretically do that, but I think that... +[16:16:21] you know, from my experience, like that's a little bit harder to do with dump app or like another example would be +[16:16:28] Dumbab can tell you about like QPL events that might have fired, but if you want to see like the exact time stamp and +[16:16:29] Things like that. +[16:16:40] I think some of the visualization things in Flipper are better. The downside to Flipper is that it's just a spew of data. I don't know if you were all remembering the tools. +[16:16:46] some of the plugins like view hierarchy you know it it says a snapshot and you can see the view hierarchy +[16:16:51] but for GraphQL or Mobile Config or Analytics or the QPL or Lume. +[16:16:58] Those things are meant to show you kind of streaming data. And so I would say those are a better fit for. +[16:17:02] Flipper and its UI versus... +[16:17:11] the text UI in Dump App. The other thing to remember is that Dump App is meant to talk to agents. So the other thing too is like, you don't want to flood it with like. +[16:17:17] a million bajillion JSON files that take up 80 megs. You know, that's probably not good for the... +[16:17:19] But there are probably other people. +[16:17:21] who can get better comments about that. +[16:17:27] Thanks, Justin. The next question is, does this work for blocks as well? +[16:17:32] Oops, um, I don't think... +[16:17:34] So if you want to inspect the blocks. +[16:17:35] few heartbeats. +[16:17:35] Yep. +[16:17:36] That works. +[16:17:39] if you want to inspect objects at runtime. +[16:17:41] I'd love to hear what. +[16:17:43] features you would want from it because, you know. +[16:17:47] boxes in my org so we could build something for you. +[16:17:52] Alright, the next question is... +[16:18:02] Does this work for the WWW repository as well to test block screens or server side changes that change the Facebook app on iOS? +[16:18:04] Interesting. +[16:18:06] So we could make that work. +[16:18:09] So if you're using a pre-build... +[16:18:10] I was out. +[16:18:20] We would have to make sure that you have a variant on your phone where the dumpout code is compiled into it. Right now, a dumpout is not compiled into like prod builds. +[16:18:23] So, one, you have to make sure that you have... +[16:18:24] I'll do it. +[16:18:26] basically like debug stuff in it. +[16:18:33] and then we would just need to enable the arc dump back command or maybe we do this with metacli. +[16:18:35] to just boom it up. +[16:18:37] There's nothing. +[16:18:38] IOS. +[16:18:43] specific about it. You know, it's just like a silly Python web server. So +[16:18:44] theoretically. +[16:18:48] with a couple of dips like yes, you could totally do this with blocks to test. +[16:18:53] All right, we have quite a few more questions, so we'll try to get... +[16:19:01] to as many as possible. The next one is writing a dumpout plugin for something like say, inspecting QPLs logged. +[16:19:03] feels redundant. +[16:19:09] what Flipper already offers. Is there maybe a quick dump-up, shim, wrapper for existing Flipper plugins? +[16:19:11] Yeah, we've talked about so... +[16:19:12] There isn't yet. +[16:19:14] But we had a... +[16:19:20] giant meeting with UI and flipper folks because again flipper is still technically in our org. +[16:19:24] the +[16:19:27] about this and I've also heard the feedback of like, it's lame if I have to like write one JSON. +[16:19:31] payload thing for flipper plugins and then another one for dump apps. +[16:19:36] So one works you more considering or somebody wants to try it out for the hackathon of... +[16:19:38] just standing, you know. +[16:19:41] Jason from your flipper plug and everything to dump that. +[16:19:44] It might be tricky because... +[16:19:45] They're different. +[16:19:47] hand shakes and whatever. +[16:19:49] what we have talked about. +[16:19:52] The people who work on Dumbass is just making a... +[16:19:54] blocks or sorry not blocks A. +[16:19:56] flipper command. +[16:20:05] that you can basically go, that's like a one stop shim, as this person mentioned, where it kind of just goes and fetches all the JSON from a flipper plugin. +[16:20:05] And then just... +[16:20:07] Brace it back through jump pad. +[16:20:08] So. +[16:20:15] Yes, duplication is a questionable thing. When to build one versus the other is a thing. I kind of think as... +[16:20:20] more and more plugins get built will get a better definition of this. I think people will figure out what's useful. +[16:20:25] Alright, the next question is... +[16:20:29] How does this compare to the existing Claude skills for controlling simulators? +[16:20:39] Yeah, I haven't actually, so I've seen the Claude skill for Iowa's Sim. As far as I understand that was more about like booting. +[16:20:41] connecting and listing. +[16:20:42] Sims? +[16:20:49] I don't know what their capabilities are. I said, I know UIE companion can do some of these things. +[16:20:51] But this is more like... +[16:20:53] The main thing with... +[16:20:54] dump app is that +[16:21:01] Since you write a plugin in Swift that gets compiled into your code, you get access to runtime information that. +[16:21:02] is not. +[16:21:08] So, so, you know, if you're using UI, it has access to tapping and touching and what it can see +[16:21:09] dump app has access to light. +[16:21:17] the objective C runtime, the Swift runtime, you could write a plugin to introspect the cache, you know? +[16:21:21] So there's capabilities below basically like site. +[16:21:26] And so I think that that is definitely an area where I see this being different than. +[16:21:29] the UI companion or the SIM skills. +[16:21:41] Alright, the next question is, is the main difference between dump app and slipper is that it's a text interface that we try not to flood. +[16:21:46] That is definitely one and again I will try to link to Manny's +[16:21:49] Document if he's not in the chat link in himself. +[16:21:53] Another one I want to bring up is Dump Hat. That's more about snapshots. +[16:21:54] Or, you know, like... +[16:21:57] bi-directional fans communication. +[16:22:05] Flipper is like a fire hose of data, you know, and often used to do like rich visualization. And so I think that. +[16:22:10] As I pointed out, like recording all the GraphQL queries that are flying through. +[16:22:15] and having a nice UI where you can kind of click and see the field and the query and all the stuff that went with it. +[16:22:17] That makes a lot of sense for like a rich... +[16:22:23] Flipper UI that doesn't make a total lot of sense to me in a text UI or what an agent might necessarily do with that +[16:22:27] I could see a world where you might want to like collect a query. +[16:22:33] and maybe like it's persist ID or you know the variables that you passed to it or something like that but I you know +[16:22:38] I think a good example of something that I would do in dump app that I wouldn't do in flipper is like +[16:22:43] reading NS user defaults, like you could trivially write a plugin to be like. +[16:22:45] Go get me the value for this key. +[16:22:49] Okay, well like what are you gonna do? Are you gonna like write a flipper plug and then just like +[16:22:54] jumps every key value pair inside of indices or defaults, you could. +[16:23:02] But like, again, for an agent, like that's way more information than it needs. Like it's way more useful to it if you can just be like, hey, what's the, what's the value for food? +[16:23:05] and it just gets you back a payload that's like it's this and it's tiny. +[16:23:06] So +[16:23:10] Again, like we're still nuancing it, but I do see differences. +[16:23:13] Thanks, Dustin. +[16:23:23] The next question is, we have an idea about building a bug bash agent that automatically executes instructions from a bug bash doc. For example, pass. +[16:23:25] GK MC +[16:23:27] inject ads, check for screenshots. +[16:23:30] Is this something Dump-Up could help with? +[16:23:31] Yes. +[16:23:37] Totally. I already know, I think IG has a feed injection plugin already. +[16:23:37] So. +[16:23:45] add injection, I don't, depends what app you're using, you know, but it seems pretty feasible. I mean another thing theoretically is like you could pass in a +[16:23:50] The way we built it is that the dump app services are session scoped, so... +[16:23:52] Theoretically, you could pass it in it. +[16:23:54] an ad ID. +[16:23:55] and then you +[16:24:01] inside of your dump that plug in could use GraphQL to go do like an actual +[16:24:03] And like process it async. +[16:24:06] Um, we're the same I guess, but um. +[16:24:09] Yes, I think you can do this. The one caveat is... +[16:24:18] For bug bashing or whatever else, at the moment the limiting thing is that you have to have a simulator. So a lot of times it is limited to like local. +[16:24:19] You know, uh... +[16:24:20] Purification. +[16:24:28] Alright, is there any prerequisite to use the ARC dump app? +[16:24:30] see a lot of command for iOS app. +[16:24:33] Nothing but a good attitude. +[16:24:36] Thanks, Susan. +[16:24:40] How does jump app compare to SB iOS debug loop? +[16:24:42] I don't know. +[16:24:44] I never used the bug loop. I'll have to look it up. +[16:24:53] I think we are. Let me just make sure we got all of our questions here. +[16:25:03] Okay, we have another one. Can you make this work on your OD? +[16:25:10] I find useful to see observe what exactly people that know how to use this tool and see things that don't work. +[16:25:15] Yeah, you can use it on your ODE. I did earlier. +[16:25:18] There are lots of videos in the Dump App Privates group of mannys doing it. +[16:25:20] Like I said, it's... +[16:25:27] If you've never done it before, what will likely happen is when you run arc dump, that will try to connect your sim on your local machine. +[16:25:33] And then it will tell you about a permission error where you need to go into your local Mac and one time run EK. +[16:25:33] connects. +[16:25:36] to the name of your... +[16:25:41] On demand server, so you know, it'll be like one two three four five dot OD +[16:25:44] You pass me a UB key so that it's... +[16:25:45] authenticated communicator. +[16:25:48] and then that one type setup is done and as you could have seen on my... +[16:25:53] ODI didn't have to do that this time. So I did it earlier today. +[16:25:54] So again, +[16:26:00] having the the authenticated flow is is necessary you know for communication because it's passing. +[16:26:06] you know, data around. I don't actually know why Dev interested. I'm assuming because you're going from your your death server to a local. +[16:26:07] You got them crepe that are... +[16:26:08] you know. +[16:26:12] You have to do when you do an OD connect. So it's kind of the same thing +[16:26:17] Okay, let's go ahead and take like two more questions. We have a few more coming in. +[16:26:24] Are you able to use this on an iOS on demand? I seem to be getting a no EK peer connector. +[16:26:29] Yeah, that's what I just brought up. So what you'll have to do if you're seeing the no EK thing +[16:26:31] You need a coat to wear. +[16:26:36] For one time you need to open up on your local Mac your terminal. +[16:26:42] and then it should give you the command you have to run. It's like EK Connect, and then it's the name of your OD. +[16:26:44] and it will make you +[16:26:47] get your UV key or whatever your security thing is. +[16:26:48] and then it should connect just fine. +[16:26:49] and you should be good to go. +[16:26:52] All right. +[16:26:59] The next question is, can we build memory inspector plugins similar to XCodes 1? +[16:27:03] Yeah, you just need somewhere to capture the state. +[16:27:04] So, yeah. +[16:27:08] like if it's in your device you have to persist it or you have to... +[16:27:12] take the JSON that it passes back to you and store it somewhere on, you know. +[16:27:17] You're on demand or paced or I don't know wherever it's safe. +[16:27:18] Yeah, good job. +[16:27:27] All right, let's do one last question. And then if anyone has any other questions, feel free to leave them in the comment section. +[16:27:30] and Dustin will get to them after the session. +[16:27:39] So the last question is, how should we think about introducing plugins? Should we start introducing plugins that are specific to particular products? +[16:27:40] For example, Facebook. +[16:27:44] the food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food food +[16:27:46] I would say just let it fly. +[16:27:53] If you end up building something, I mean, it's so trivial to build these things and refactor them that if you build one. +[16:27:53] for groups. +[16:27:56] and it turns out to be super useful for pages. +[16:27:59] Cool, we'll fix it later. +[16:28:02] especially in this age of AI where there's like a... +[16:28:04] 16 open claws like I +[16:28:05] If there is, get there. +[16:28:08] 15 different plugins for view hierarchy. Like I'm not. +[16:28:10] Without worried about it +[16:28:10] yet. +[16:28:15] I think you just let a thousand flowers bloom and we can refactor after the fact. +[16:28:24] Perfect. Thank you so much, everyone, for joining our session today. And thank you, Dustin, for presenting. +[16:28:29] And like I said, we will get to the comments if there's any additional one section. +[16:28:31] and reply to those. +[16:28:32] today +[16:28:36] Thanks everybody. +[16:31:24] by just typing A into your search bar. We also have an iOS app that is released, so you should see that auto downloaded on your phone. +[16:31:29] So you might be wondering what exactly is HM? +[16:31:41] So essentially it is a command center to view and manage your coding agents. You can access them from anywhere. So while you're on the go or if you just want to open up this website in your laptop. +[16:31:48] And essentially you can start and continue agents from anywhere. So you can kick off a coding agent in your Quad CLI. +[16:31:55] you can resume it on Agent Home. So if you ever lose a conversation, want to continue it when you're at lunch or anything like that. +[16:31:56] you can do that through H-Comb. +[16:32:05] You can also always stay in the loop. You have gchat notifications when an agent is finished with something, needs your review, needs your input, anything like that. +[16:32:10] So you can really just think of it as your low version way to stay up to date. +[16:32:13] with all of your coding agents and kick off new tasks from anywhere. +[16:32:25] Like I mentioned, it's available on web and mobile. Highly recommend going to check it out. We currently have it launched on iOS and we are actively developing our Android app right now. +[16:32:34] Some technical information. So, in Agent Home, your conversations currently run on sandcastles. +[16:32:46] instead of ODs. And so what this means is that it's really easy to kick off a session. Whether on Web or mobile, there's like no setup needed on your end other than to type into a prompt. +[16:32:49] You can also resume a session. +[16:32:53] even much later because we can just get you another sandcastle server. +[16:33:01] You can work in any repo. You can turn it into any mode. We support code, plan, and ask modes. +[16:33:04] And you can also attach images, diffs. +[16:33:05] Thanks. +[16:33:07] any context that you want for an agent. +[16:33:17] Lastly, I know that a really big use case is probably to start a conversation on quad and then resume it later on mobile or pick off. +[16:33:20] a new quad conversation on Agent Home. +[16:33:27] We sync all of your Cloud MD files so if you have plugins installed and stuff in your Cloud MD. +[16:33:29] that should carry over to Agent Home. +[16:33:35] You can also set your working directory. We have a ton of settings that you can configure yourself with. +[16:33:37] Again, we'll show in a minute. +[16:33:47] So just going a little bit deeper into our capabilities. We currently support both DevMe and Claude. +[16:33:54] and are trying to update these to like the most recent models whenever they come out. So it should be pretty up to date with what you're seeing anywhere else. +[16:33:59] You can make continue sessions from anywhere and take them to anywhere. +[16:34:05] So for example, you can start a conversation in VS Code on your ODE. +[16:34:09] You can later resume it in Agents Home on a sandcastle server. +[16:34:11] And then later you can actually take it. +[16:34:14] to Claude Code CLI and resume it there. +[16:34:20] You can do any combination of these. We also support integration with fabricator sessions, tasks, file a diff. +[16:34:21] Everything. +[16:34:26] Like I mentioned, lots of configurations and attachments and you can do +[16:34:30] anything you want to give the agent as much context as you want. +[16:34:33] and already talked about cloud compatibility. So. +[16:34:34] I won't repeat it. +[16:34:47] being said, the team is moving super fast. We launched the first version of this in February, and we are currently working very hard to +[16:35:04] on BS code or cloud CLI. +[16:35:10] in an OD and then have that OD continue running at age at home. So it essentially lets you remote control your procession. +[16:35:11] from Egypt Hall itself. +[16:35:18] We also currently support live previews if you want to make changes on web. +[16:35:24] We are trying to build mobile and other surfaces as well, so you can basically do your end-to-end development in Agent Home. +[16:35:28] And lastly, we're always working on having a better UX. +[16:35:29] on. +[16:35:35] and giving you better controls to sort your sessions, rename them. Pin them. +[16:35:36] organize them in any way. +[16:35:44] I think that's it for now. I will pass it off to Dan to just really show you what Agent Home can do. +[16:35:52] Hi, everyone. Really excited to be sharing this with you guys and the Facebook engineering team. +[16:35:53] Um +[16:35:59] So I'm going to show a little bit of Agent Home on the web and then spend some time on the mobile app as well. +[16:36:08] So this is like the landing page for the web. I mean, you've seen this before. You can basically, you know, prompt something, add your images attachments. +[16:36:14] Select from your agent type, what mode you want, what plugins you want from the... +[16:36:17] cloud template plugins that's internal right now. +[16:36:25] So I had already picked one off here and because we're demoing to Facebook engineers, I chose to do a Facebook +[16:36:29] themed simple change that we could just try out. +[16:36:32] So if I go up to the top of this chat, I basically said... +[16:36:41] On facebook.com there's a column on the right side of the landing page that says contacts and I want to make that say my contacts and have it be green and double the size so +[16:36:43] Super ugly, nothing you'd ever want to do. +[16:36:44] But... +[16:36:49] You can see here that the agent basically ran and +[16:36:50] completed the task. +[16:36:56] And then here, I can actually test my changes in the preview mode. So I already have this open over here. So I'm going to go here. +[16:36:57] And, um... +[16:37:00] You can see on the right side of my screen. +[16:37:03] This my contact's text is like super ugly. +[16:37:09] So I've never reserved an OD for this. This is all just using sandcastle. +[16:37:13] and it's super powerful for just getting quick things done. +[16:37:16] or kicking off work that you then want to resume later. +[16:37:22] And just to show you, I can say, how about changing this to orange? +[16:37:24] and hit send. +[16:37:28] And that's going to work for a little bit. +[16:37:29] Um +[16:37:38] But I can also show you while that's working. We have some buttons up here. Like for example, you could submit a diff of the code changes right now. +[16:37:41] You could look at the code changes that it actually made. +[16:37:41] Um +[16:37:42] I could... +[16:37:48] continue in VS code so I can open this up in my OD that I'm currently running. +[16:37:50] are a continuant in cloud code as well. +[16:37:55] So let me go over here to see if this actually worked and it actually updated them live. +[16:37:57] So you can see that. +[16:37:58] updated live there. +[16:37:59] So +[16:38:06] All of this functionality is also available on the mobile app. So I'm going to show that to you now. +[16:38:07] Um +[16:38:09] running in my simulator here. +[16:38:10] Okay. +[16:38:14] So here we can see this is the mobile app so +[16:38:16] It's really... +[16:38:18] Easy to use. +[16:38:23] There's all kinds of settings in here for what plugins you want installed. +[16:38:27] You can see the entire plug-in repo that we have internally. +[16:38:32] You can turn certain plugins on and off. You can set your working directory. +[16:38:36] So for example, if you always work inside of like a certain app. +[16:38:40] You can set that and then when you start new agents, they'll have that context. +[16:38:43] and you can save time on your prompts. +[16:38:44] Um +[16:38:49] Here I'm going to kick off a new prompt. Let's say +[16:38:58] Let's say we want to make a change to Agent Home app itself, which I've actually been doing quite a lot. I've been using the Agent Home app to change the Agent Home app. +[16:39:00] which is very meta. +[16:39:03] So here it says good afternoon, Dan. +[16:39:05] And then I'm just going to give it a screenshot. +[16:39:08] because I actually took a screenshot earlier. +[16:39:11] from dark mode where it says good afternoon, Dan. And I'll say. +[16:39:13] Instead of +[16:39:15] Good afternoon. +[16:39:18] Thank you. +[16:39:18] users +[16:39:20] First name. +[16:39:22] Make the full name +[16:39:24] So I'll do that. +[16:39:26] So I'm going to kick that off. +[16:39:28] It's going to start that conversation. +[16:39:37] And again, like you could do this from your couch, you know, like you, or wherever you are, like the cafeteria, you don't have to reserve anything. It's like instant setup. +[16:39:40] So this is going to start, and while that's starting. +[16:39:43] I'm going to go here and show you the session list here. +[16:39:44] Hold on one sec. +[16:39:46] And that worked. +[16:39:48] Perfect job for a moment. +[16:40:13] So what I have in the background here is I've got +[16:40:16] Just to show you, I do have an OD checked out and I've got... +[16:40:24] A deadmate session running here that's just like listing a bunch of images that are used in the in the facebook.com website +[16:40:27] And then I've got two clock sessions actually. So one. +[16:40:32] where I asked it how notifications work in Facebook.com and another one. +[16:40:37] where I asked it to find the largest React file and just refactor it. +[16:40:39] So, oh man. +[16:40:42] I don't want to touch the function yet. +[16:40:47] Hold on a sec. +[16:41:04] right let me show the heavens +[16:41:13] apologies for that. +[16:41:15] live demo here +[16:41:16] Um +[16:41:19] I can find these anyway, so let's see. +[16:41:23] If I go to my agent home on the web. +[16:41:25] Um... +[16:41:30] So this is the exact chat that I was just having. +[16:41:32] with Deaf knit。 +[16:41:34] And now I can see it. +[16:41:35] in Agent Home. +[16:41:43] So this is really handy for, you know, if you've started some work on your OD and you want to like just keep track of what's going on. +[16:41:44] You can do that. +[16:41:58] Hey Dan, by the way, it sounds like from the comments section that there's a sev out, so that's the reason why you're having issues with it. +[16:41:59] Thank you. +[16:42:00] Thanks, fam. +[16:42:01] Thank you. +[16:42:02] Okay. +[16:42:07] Yeah, this was like just working, so I'm sorry. +[16:42:12] that it's not loading the sessions. It looks like it's actually not loading them on the web either. +[16:42:13] Um +[16:42:18] Yeah. +[16:42:19] Well +[16:42:25] What I was going to show and what I encourage you all to use later on. +[16:42:28] is that you'll be able to see your Cloud Code CLI sessions. +[16:42:29] Um +[16:42:30] arm. +[16:42:35] your age-at-home mobile device and even take over those sessions as well. +[16:42:37] You can take over those sessions. +[16:42:43] and then you'll see the Cod code session on your OD stop with a command line interface to resume it later. +[16:42:45] command line. +[16:42:46] All right +[16:42:48] chat to resume it later. +[16:42:49] and you can also take over. +[16:42:54] the Visual Studio VS Code sessions and take those over as well. +[16:42:58] Yeah. +[16:43:01] with that. +[16:43:05] Let me, I guess we can go to questions and hopefully... +[16:43:10] in the next few minutes. This will work again and I can show it off. Or you can just try it. +[16:43:12] once this resolves on your own device. +[16:43:23] Hey Dan, I have been monitoring the chat and it sounds like there's a few questions here and apologies in advance if I don't share any of this. +[16:43:29] Will Agent Home or some other tools support continuity of agents for scenarios where you want to deliberately +[16:43:31] reset context to avoid context. +[16:43:32] Rocks? +[16:43:38] but still want the agent to retain the task context and perhaps link them together in history. +[16:43:40] Um... +[16:43:44] Yes you are.. +[16:43:45] So this question actually touches on something that our team's working on. +[16:43:56] which is like the ability to essentially group projects together or group sessions together so that you can start a new agent with the context of some of your previous work. +[16:43:58] So this is something we're actively working on. +[16:44:03] I don't know specifically what that question was. +[16:44:07] was hitting at, but I think that's the general theme and that's definitely something that +[16:44:08] like we're working on. +[16:44:17] Awesome, thanks for that. Can we view and modify all context that is saved in Agent Home for a particular session? +[16:44:24] Yeah, I mean, the sessions are... +[16:44:30] Like you can you could the sessions are all public you can share a session that you've created with someone else +[16:44:32] They can look at that on Agent Home. +[16:44:36] There's also like a Devman trajectory viewer that you can look at. +[16:44:37] Um +[16:44:40] I guess, can you say the question again actually? +[16:44:47] Yeah, can we view and modify all context that is saved in Agent Home for a particular session? +[16:44:53] Yes, yeah, it's all there for you. You can go back to it at any time and you could even... +[16:44:59] For example, grab the context or the output of the agent and start a new agent with that output as well. +[16:45:03] Awesome, thank you. +[16:45:11] And here's another one. What is the difference between Agent Home and the Happy Engineering app? They sound like a similar idea. +[16:45:13] That's a great question. +[16:45:17] So the happy engineering app allows you to essentially remote control. +[16:45:19] a Claw of CLI session. +[16:45:22] that's already running on your OD. +[16:45:28] The difference is with agent homes, it's not clogged specific, so it will work on. +[16:45:35] Any agent type that the DevMate VS Code team has incorporated into +[16:45:38] into the main core system. +[16:45:42] It also doesn't require any setup, so you don't need +[16:45:50] to have a Dev Mate, or you don't need to have like an OD with something running already to start off an agent home. +[16:45:51] You can just... +[16:45:52] do it. +[16:45:56] like any from nothing because it's just essentially +[16:45:58] grabbing a new sandcastle instance. +[16:45:59] when you need it. +[16:46:02] Awesome, thanks for that. +[16:46:05] I have a, the next one is a tube. +[16:46:06] two questions. +[16:46:07] from the same person. +[16:46:15] If I want to do automated delegation to cloud code, how can I do that? And does the agent have any API to support that? +[16:46:17] And there's a second question. +[16:46:19] Also, how can I make +[16:46:23] it download custom plugins which we are going to use by the agents. +[16:46:31] So yeah, these both sound like they're just like deeper integrations with block code. +[16:46:34] We do have the plugins that are available on +[16:46:39] Like if you do bunny CT for quad templates, like all of those plugins are available. +[16:46:40] Um +[16:46:41] We also +[16:46:46] automatically sync your .files. So like if you have a cloud.md. +[16:46:48] you'll be able to just use that an edge at home. +[16:46:49] There are some... +[16:46:57] like more detailed specific plug-in requests that we do not support yet, but like the reason that we're basically launching. +[16:47:03] and starting to grow. So we want to hear all your feedback about like what's really important. +[16:47:04] Nina Diabini. +[16:47:06] than you wanted to have about that. +[16:47:10] No, that's how we, yeah, we already support plugins we're working on. +[16:47:17] ask close integrations with quad-code as possible. But yeah, feel free to also comment them and the questions, leave them in our feedback group. +[16:47:19] which I'll share my screen to show at the end. +[16:47:20] I'll be right back. +[16:47:24] Awesome, thanks both. +[16:47:26] I have another question for you guys. +[16:47:31] What use cases do you recommend us using the agent home over? +[16:47:38] the dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[16:47:38] code for overall. Over VS, sorry, over versus code overall. +[16:47:40] I think this +[16:47:53] Definitely depends on your personal preferences. I would suggest, I think Agent Home is really great for stuff when you're on the go, where you have a quick thought and you wanna share it with a coding agent. So if you're on your commute or anything like that, anywhere where you can't pull out a laptop. +[16:47:55] The Agent Home Mobile app is fantastic. +[16:48:03] I think maybe if you want to manually write code, have deeper access to all the specific files that you want to edit, VS Code might be a better fit for that. +[16:48:15] And then for the Agent Home desktop website, I would use that if you wanna work with multiple agents at once. I feel like the UI is really good for managing, seeing what's going on with all of your individual agents. +[16:48:15] on. +[16:48:17] Yeah, anything else to add, Jim? +[16:48:21] Yeah, I really like to use it when I have like... +[16:48:25] small to medium sized tasks, which are... +[16:48:30] like orthogonal from each other and I just want to kick off like six or seven of those. +[16:48:33] and then kind of get back to them later. +[16:48:33] Um +[16:48:37] I've also found it really useful for working on the mobile app. +[16:48:40] So like you could just be on your couch +[16:48:43] looking at the yellow Facebook app. +[16:48:49] see a bug there, take a screenshot of it, circle it, send it to Agent Home. Like all without... +[16:48:50] getting on your computer. +[16:48:51] And then +[16:48:58] you know, look at the results of that later on. So those are like the two main use cases that I really, really love about Agent Home. +[16:49:02] Awesome, thanks both. +[16:49:07] When switching between sessions on devices, will the agent of choice retain context? +[16:49:10] In other words, using your example of cloud code in Agent Home. +[16:49:14] and moving to CLI, will the CLI instance have the same context? +[16:49:15] Yeah. +[16:49:24] They will have all the context from what you've typed before an agent home. And vice versa, if you start a session on another service and then bring it into agent home, it'll have all the previous context. +[16:49:30] This next one here. +[16:49:36] Does the Dev Server Cloud Code sessions show up in Agent Home, or can we only do this in sessions? +[16:49:37] started on ODS. +[16:49:46] They should show up in Agent Hall. The ones that I've started on my Dev Server do. If you're not seeing them show up, let us know. +[16:49:49] Awesome. +[16:49:51] Can we do schedule tasks? +[16:49:54] on like using agent home. +[16:49:58] I don't +[16:50:08] I think we currently support that at the moment. I think what is supported is if you kick off, like if you delegate sessions from other places that happen to use quad or deafening. So if you like. +[16:50:09] Auto delegate. +[16:50:12] they should show up in Agent Home. +[16:50:18] Thank you. Yeah, I just added that like... +[16:50:21] We love getting these questions because +[16:50:28] There's so many different uses for Agent Home and there's also so many different workflows that each engineer goes through. +[16:50:30] So like figuring out... +[16:50:37] which are the most common and which would be the highest value for the team to focus on is really important. So we really want to hear your feedback inside. +[16:50:40] the Agent Home Feedback Workplace Group. +[16:50:44] Perfect, thank you. +[16:50:50] Does it keep track of things done in analytics, agents, etc.? I keep losing that data. +[16:50:54] Not today, to the best of my knowledge, but... +[16:50:56] This is helpful feedback as a feature request. +[16:50:59] Alright. +[16:51:02] This one looks like it's specifically to Dan. +[16:51:06] Dan, when I would use DevMate versus Cloud Code versus Agent Home. +[16:51:11] Oh, sorry. When would I use Tev Mate versus Cloud Code versus Agent Home versus... +[16:51:13] Conditions. +[16:51:18] Bitte Bitte Bitte +[16:51:20] Yeah, kind of what I answered before. +[16:51:22] Um, I mean +[16:51:27] I work on this mobile app a lot, so I'm in VS Code a lot. +[16:51:27] Um +[16:51:29] doing work there. +[16:51:31] like when I'm at my computer. +[16:51:32] Um +[16:51:33] But sometimes... +[16:51:36] if I just have some thought that's like. +[16:51:43] I want to get this done, but it's completely orthogonal to my current workflow. It's not going to have any merge conflicts or anything. +[16:51:44] Then I'll just kick off. +[16:51:45] work on agent home. +[16:51:48] because it's so low friction to do that. +[16:51:54] Additionally, like if I'm on the go or on my phone or I see a bug on my phone, I will also kick off work. +[16:51:56] on Agent Home to do that. +[16:52:02] And then later on, I can go and pull that work onto my depth server and continue it. +[16:52:09] But we're also working on ways to, like I showed in the demo, with the color and text change on Facebook.com. +[16:52:12] We're working on new ways to be able to just... +[16:52:17] Really you validate it." Yes you are the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most important part of the most most important part of the most important part of the most important part of the most most important part of the most important part of the most important part of the most most most most most most most most most most most most most most most most most most most +[16:52:18] your work without even having a web server. So that's kind of the direction that. +[16:52:23] things are headed, but we're kind of in this in between state right now where you really do need to... +[16:52:26] to check your work on a lot of dips and changes. +[16:52:29] Thanks for that. +[16:52:31] I have another question. +[16:52:37] Can Agent Home Reserve and Agent Running for a very long time save 48 hours? +[16:52:41] um +[16:52:42] So... +[16:52:46] The way it works right now is when you get your sandcastle server. +[16:52:47] Um +[16:52:53] The agent runs on that until it finishes and it's been idle for a certain amount of time. That could be like... +[16:52:55] 30 minutes to an hour. +[16:52:55] Um +[16:52:58] Once it's been idle for that long, the machine is given back. +[16:53:01] But if you want to reserve... +[16:53:04] And that's only when the agent's like completely stopped. +[16:53:07] If you want to then resume that session. +[16:53:11] you can just like send another message to it and it'll essentially +[16:53:16] pull a new sandcastle machine, rehydrate your session, and continue going. +[16:53:21] So we don't have the ability to just have your own machine that's... +[16:53:25] dedicated to your agent for 48 hours because of load balancing requirements. +[16:53:26] Um +[16:53:32] But that's currently, like, but you do have these longstanding sessions that you're able to resume at any time. +[16:53:36] Awesome, thank you. +[16:53:45] From somebody else, one problem I have with agent team is the spawn agent's context is gone for good when the CC session is dead. +[16:53:47] as your team looked into persisting. +[16:53:48] context. +[16:53:50] for spawned agents. +[16:53:54] That's for subagents. +[16:53:56] I assume you're not the other one. +[16:54:02] Yeah, I think this is I assume this is for agent teams like dedicated sub agents +[16:54:06] We are currently looking into this generally so persisting. +[16:54:11] context across different sub-agents in a session and across related sessions. +[16:54:16] Thank you. +[16:54:18] Next one. +[16:54:22] Can the working directory be set to a custom directory in home? +[16:54:27] the most important thing about the +[16:54:28] in home like G Drive to work with workflows like Second Brain. +[16:54:31] Um, I +[16:54:37] I don't know about second brain. You can set your working directory to be. +[16:54:39] anything in the repo. +[16:54:45] I don't think that includes like if it was a personal directory in your home directory. +[16:54:46] but again +[16:54:52] If this is like a really important workflow for you, then let us know in our Agent Home Feedback Group. +[16:54:55] Thank you. +[16:55:01] When would I use DevMate versus Cloud Code versus Agent Home versus Confucius? +[16:55:04] I think we got that one already. Oh. +[16:55:04] Okay +[16:55:05] No worries. +[16:55:06] You're right +[16:55:11] Can you talk specifically about use cases when one should be preferred over the other? +[16:55:14] Um +[16:55:20] I mean, I think I touched on some of this. Like, I really like Agent Home for these orthogonal... +[16:55:25] like non conflicting tasks that I want to get done that are clear. +[16:55:26] Um +[16:55:29] and also for when you're on the go using the mobile app. +[16:55:30] Um +[16:55:32] versus if you were doing some... +[16:55:37] in-depth work where you had to test constantly on your OD. +[16:55:38] That's +[16:55:41] maybe a time when I wouldn't use agent home for that. +[16:55:42] Um +[16:55:47] We also find Agent Home to be really, really useful for like PMs and designers as well. +[16:55:50] That seems like a pretty popular use case. +[16:55:53] especially with that preview mode that I showed you. +[16:55:55] like if you're +[16:55:58] if your PM or designer wants to just like send a link to +[16:56:02] a working prototype that they made using Agent Home. +[16:56:05] and a depth server link that it can send you. +[16:56:08] That's also a really good use case too. +[16:56:11] Perfect, thank you. +[16:56:24] When I want to open VS Code for the code change made in an agent home conversation, is it possible that I select which server it will be opened on like local OD or dev server? +[16:56:29] If it was a... +[16:56:33] Actually, this is working again, so I'm going to just share it with you. +[16:56:35] So here... +[16:56:38] Here's my list of sessions that I was trying to show earlier. +[16:56:39] So +[16:56:44] I actually have this OD reserve right now. It's like 145231.OD. +[16:56:48] So I can actually see the cloud code sessions and. +[16:56:49] the +[16:56:50] Um +[16:56:55] and the VS code sessions that are running on that OD. So I can see them right here. +[16:56:56] Um +[16:56:58] So for example... +[16:57:01] If I open this one... +[16:57:04] Okay. +[16:57:05] So this is... +[16:57:11] This is basically a list of like all the PNG files that are in... +[16:57:12] please book.com. +[16:57:17] And here you can see like this is the same exact text and now I'm viewing it on here. +[16:57:21] And if I want, I can start typing and just take over that VSCode session. +[16:57:22] Same thing for Claude. +[16:57:23] um +[16:57:26] Hopefully the session list will load again. +[16:57:29] um +[16:57:31] Yeah, so like if I'm on cloud right here... +[16:57:33] Um +[16:57:36] Actually, let me open this one instead. +[16:57:39] I know we're about to be out of time as well. +[16:57:49] Sorry. +[16:57:53] Yeah. +[16:57:58] But I was going to show you basically like kicking over a plot session and then. +[16:58:00] And then how that works. +[16:58:01] But yeah, so you're like +[16:58:03] It essentially tells you which... +[16:58:07] ODs your sessions were from if they were from ODs. +[16:58:10] Otherwise, if they were just on a sandcastle. +[16:58:13] instance then it wouldn't show anything. +[16:58:16] Perfect, thank you. +[16:58:25] Next question is, what is the ideal status in future for using agent home plus clock code? +[16:58:27] It seems we can run it overnight no matter. +[16:58:31] we are on a laptop or not, should we shift from ODs to dedicated? +[16:58:32] Dev servers. +[16:58:35] I think it's up to... +[16:58:37] personal preference like. +[16:58:39] how you want to work. +[16:58:40] I would also say that +[16:58:41] Like. +[16:58:46] Everything is shifting right now, you know, like the whole industry and also our company. +[16:58:55] in terms of how people get worked on effectively. It's all changing, so we're all trying to find the most efficient way to spend our time. +[16:59:02] And we are going to keep growing Agent Home to be a tool that is like incredibly useful for... +[16:59:11] for all engineers at the company and that's our goal. So you know again like if you have use cases or feedback. +[16:59:11] Bye. +[16:59:12] Bye. +[16:59:16] would make it better for your specific workflow. +[16:59:21] then please let us know and we're going to be prioritizing and organizing all that feedback soon. +[16:59:24] Perfect. Thanks, Dan. +[16:59:32] We probably only have time for one or two questions. I have a, I hope a short one. Can agent home take screenshots for Android? +[16:59:36] It cannot do that yet. +[16:59:37] but +[16:59:41] Just like we have the preview mode for web, we +[16:59:45] We are planning to work on mobile preview. +[16:59:46] soon. So that's just... +[16:59:50] We're just looking at testing the ability to automatically test. +[16:59:53] and generate screenshots and test plans. +[16:59:54] on different surfaces. +[16:59:57] But right now, web is the only one that works. +[17:00:04] Perfect. Thanks. Let's do a last one here. Can we integrate Agent Home with My Daily Podcast Digest? +[17:00:10] Example, my daily digest tells me about a task and I can trigger Agent HOME to start fixing that task. +[17:00:20] We are working on something like this, um, with having an agent home assistant that can kind of see what else you have going on and help you prioritize or start working on things. +[17:00:20] So stay tuned. +[17:00:25] Thank you guys. I think that's all we have time for. +[17:00:27] Awesome. Sharing our... +[17:00:32] feedback and beta groups. Feel free to join them or you can just search these names in workplace in case this. +[17:00:35] slideshow is a jerk, but thank you so much, please. +[17:00:39] share all your questions and feedback in our feedback groups too so we can prioritize it. +[17:00:42] Thanks, Tom. +[17:02:39] Pretty much all for me as well. +[17:02:48] See the sound of the team saying There's so many new things I didn't know what Dumpap was So Rachel, I think the last two ideas Had no clue +[17:02:48] Yeah +[17:02:51] Yeah, yeah, the A.G. Holmes and he's saying. +[17:02:56] Yes。 +[17:03:11] maybe just wait for more minutes, see for more folks +[17:03:29] Thank you. +[17:03:55] Hello, team. +[17:03:59] lot of things that I've never heard about. +[17:04:06] And I saw that out. +[17:04:10] Some of team members actually developed some really cool stuff during the weekends. +[17:04:14] I saw that on TOFU +[17:04:17] the class the class +[17:04:18] to write the entire notification wiki page. +[17:04:20] All right. +[17:04:24] and you develop the FIFU agent. +[17:04:25] that we've been playing with. +[17:04:39] Yeah, I think also an onion work on quality like I think our presentation was really informative for me But I didn't know about both of those +[17:04:40] initiatives. +[17:04:41] Yeah. +[17:04:54] Nice. Yeah, one of the major topics we have for Hackathon is probably to develop those things for almost all the work streams. Yeah, exactly. +[17:04:59] Oh, I have a new pool of drawing on Seattle. +[17:05:00] That's nice. +[17:05:03] Cool, maybe we can get started. +[17:05:06] I think we're primarily just using... +[17:05:08] The sheet. +[17:05:11] to talk about, but let me... +[17:05:13] Make sure you're not screaming. +[17:05:36] And +[17:05:39] She told us supposed to use a gross copy version. +[17:05:42] Yeah, the reason is because the mouse version. +[17:05:44] They're really low actually. +[17:05:47] I was trying to load a couple minutes back, but it really worked. +[17:05:53] Cool. So yeah, I think you guys all attend this morning session, you know, what we're trying to do. +[17:05:56] I think TLDR is that we try to convert every +[17:06:01] the work we're doing now to be some sort of like this. +[17:06:02] automated flow. +[17:06:04] So, uh... +[17:06:06] Yeah, so I think there are a couple of... +[17:06:07] Um +[17:06:10] areas we can try to tackle in this session. +[17:06:12] I try to give you some of +[17:06:13] my uh +[17:06:16] POV on the B and... +[17:06:19] the feed products and then I'll pass it over to Adolf when you are. +[17:06:21] for now if at the in-frog perspectives. +[17:06:26] So I think for BE we can potentially to because +[17:06:28] We already have the BE a play out. +[17:06:31] which covers like all operational stuff. +[17:06:33] and then plus ETS. +[17:06:34] the D projects. +[17:06:35] And... +[17:06:37] We also have the same topics. +[17:06:38] So I think... +[17:06:42] folks can sort of like bring storm. +[17:06:44] How can we... +[17:06:49] automate those things in a sense like creating a scale list or creating plugins, etc. +[17:06:52] especially for the ETS, I think, with the... +[17:06:54] that the workflow should be. +[17:06:56] you meant easier to achieve. +[17:06:59] So that's for B and then. +[17:07:03] for the feed products. +[17:07:06] I think we have observed a repeat pattern that +[17:07:09] We are building to a new surface. +[17:07:11] We have a certain... +[17:07:15] pipelines going on such as retrieval ranking filtering +[17:07:16] Um +[17:07:19] 。 +[17:07:20] tuning, et cetera. So. +[17:07:21] Right now it depends on... +[17:07:23] BeachwalkStream 2 +[17:07:24] basically spin out. +[17:07:25] the template. +[17:07:27] built from scratch. +[17:07:29] And once you want to see if there's a way that... +[17:07:33] from a system perspective is already a template. +[17:07:37] which is calling a scale of feeling in the various contexts that automatically +[17:07:38] crazed all those +[17:07:40] the most important thing about the +[17:07:42] template for you. So that's sort of what we're trying to achieve. +[17:07:46] Okay, with that I will... +[17:07:49] Up, come to Adolfo to see. +[17:07:51] when he shows up after. +[17:07:53] going sides from like not if perspective +[17:07:56] Yeah, I think... +[17:08:01] Not only on notice, but... +[17:08:04] something that could be interesting is the... +[17:08:08] automation of a lot of the areas that we do. +[17:08:12] Today I was experimenting with... +[17:08:18] I think it's a skill that let you create experiments and also query experiments that are active. +[17:08:22] So it will be interesting to have. +[17:08:25] some sort of framework that +[17:08:28] We provide the diff with the parameters I need. +[17:08:30] Can I just say, um... +[17:08:32] experiment and it has like a +[17:08:33] Esqueço. +[17:08:39] but that checks the experiment is running correctly and provides the metrics. +[17:08:40] Um +[17:08:44] I think those things are quite interesting. So I'm not sure if... +[17:08:47] Those will be part of the, I will think so. +[17:08:51] And those are kind of the things that are... +[17:08:53] more related to +[17:08:59] having agents that are like a schedule or like are constantly active and have +[17:09:02] persistent memory. +[17:09:03] Um +[17:09:04] So I'm still +[17:09:12] trying to say how can that be implemented in that area, but that's something that I think is very interesting for us. +[17:09:14] formada en general. +[17:09:15] Um +[17:09:20] Yeah, for notifications I think there are already some... +[17:09:26] ideas.. that.. the.. the.. the.. the.. ideas.. the.. the.. the.. the.. the.. ideas.. the.. the.. the.. the.. ideas.. the.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. ideas.. the.. the.. the.. +[17:09:30] put on the document. I added a couple of them mostly on testing. +[17:09:33] We usually don't test. +[17:09:39] Or these that great like we rely on the +[17:09:41] the changes. +[17:09:43] Probably small enough. +[17:09:47] for not having to fully test but if we have an automatic way +[17:09:49] for... +[17:09:52] the agent to test the end to end of the system. +[17:09:53] I think that will be... +[17:09:54] Yeah. +[17:09:59] that will reduce the amount of time that we have to do at the bogey as well. +[17:10:02] It will be interesting that. +[17:10:10] It should have context of QE and GKs. So it knows what is the paths that are actually supposed to be executed. +[17:10:14] This is a little bit different than that automatic +[17:10:16] tests +[17:10:17] since um +[17:10:22] are needed to be written for specific purposes on the logic. +[17:10:27] And the other tool that I have proposing here is... +[17:10:33] a tool that allows to, for example, get a notification of any type. +[17:10:38] So we said I need to get a notification such type such medium +[17:10:43] and then it figured out what are the parameters that need to provide... +[17:10:48] product ideas from recent products that you watch or something like that, so it's not just a random +[17:10:50] Oculus. +[17:10:54] product that is usually used, but something that... +[17:10:59] It gives you an idea of what the actual user experience will receive in the notification. +[17:11:05] Oh, thanks a lot, Ful. +[17:11:10] the +[17:11:11] Now, pass over to Yvonne if you have. +[17:11:12] Oh, hi. +[17:11:15] Thanks. So... +[17:11:16] First of all, this is... +[17:11:17] Well +[17:11:23] Quite exciting. All the skills that I've been learning today and everything that we see we can do with. +[17:11:24] All these. +[17:11:26] AI plugins and agents. +[17:11:29] from the infrasight. +[17:11:31] We have several projects that +[17:11:33] that we're thinking of. +[17:11:37] and potentially more when Jekyll comes back and... +[17:11:39] He's gonna be back next week. +[17:11:41] from his one month leave. +[17:11:43] Um... +[17:11:49] So basically, some of the projects from the infrasight that I wrote down were +[17:11:54] the feature recovery, so an agent for automatic feature recovery. +[17:11:56] Um, for all. +[17:12:01] in for use cases not just for fightfu but I'm envisioning we can +[17:12:02] use it first for fight. +[17:12:03] food. +[17:12:06] For this one, I want to just add our node here. +[17:12:07] that. +[17:12:09] I don't believe +[17:12:14] We're still at a place where we can let the AI go wild around our code base. +[17:12:22] especially since we are, I mean we've all witnessed multiple times that you know sometimes it hallucinates to get some things. +[17:12:24] It produces some errors. +[17:12:30] And this is something that our infrastructure can't really handle at this point. We have, because it's a multi-tenant. +[17:12:32] Our services are multi-tenant. +[17:12:36] So, for example, if we let it, you know, just auto. +[17:12:39] autonomously work on. +[17:12:42] improving the feature coverage for fight food it might +[17:12:47] cause a sev or mess up other use cases across the market place. +[17:12:54] So for that purpose, I kind of created a smaller environment where I have some custom files. +[17:12:55] that we will let it. +[17:12:57] kind of go wilding. +[17:12:59] Let's see what it will do. It'll try to... +[17:13:01] maximize the feature coverage. +[17:13:04] for 5-foot on its own. +[17:13:07] This includes some custom. +[17:13:08] thrift structs. +[17:13:10] that it can alter so it doesn't mess up. +[17:13:13] the existing ones, etc. So that's one. +[17:13:15] All right, the other one. +[17:13:20] would be we have in our roadmap a ranker feature a ranker debugger +[17:13:27] So the ranker debugger will, that will also be an agent, which will basically. +[17:13:30] be capable of +[17:13:32] doing an end to end. +[17:13:37] a comprehensive end to end testing of a given ranking use case. +[17:13:42] and provide fixes and next steps and recommendations. +[17:13:49] And then we have also a new use case set up. +[17:13:51] which will not... +[17:13:54] only fix existing +[17:14:00] issues with existing use cases, but it will also be able to create child classes. +[17:14:04] Both on the double-dubb side and on the C++ side. +[17:14:06] to support new +[17:14:07] um +[17:14:12] new new rancor use cases be it growth or any other +[17:14:21] So it will be capable of analyzing the code and create extending certain classes as necessary and maintaining a proper architecture. +[17:14:28] and doing a quick setup and figuring out what are the requirements for this. For example, for a logged out ranker. +[17:14:33] logged out use cases we're using with 5FuNOW and several others that are coming in the future. +[17:14:36] it would be able to create the... +[17:14:38] the boilerplate code. +[17:14:46] necessary to run an end-to-end test and re-verify itself and iterate on it until it maximizes. +[17:14:50] So I think all those three that I just mentioned. +[17:14:52] kind of fall within the raker. +[17:14:54] debugger, right? And +[17:14:55] um +[17:14:57] because I assume this would... +[17:15:00] potentially be like several agents working together. +[17:15:05] where they would be communicating with each other. And I assume from the growth side. +[17:15:07] when setting up a new... +[17:15:08] use case. +[17:15:09] that +[17:15:12] you would be able to directly... +[17:15:13] Um +[17:15:14] ask +[17:15:17] Let's say cloud code to +[17:15:20] to set up a new use case and that new use case that +[17:15:22] that instruction would then... +[17:15:24] use some of our agents on the infraside. +[17:15:25] uh to +[17:15:27] to set it up quickly. +[17:15:34] and someone on your side to set up some of. +[17:15:36] some of the boilerplate code on your side. +[17:15:38] you know, before even calling Ranker. +[17:15:40] Oh +[17:15:42] Sounds good. Thanks Eva +[17:15:44] So, yeah, so +[17:15:45] just share a one. +[17:15:47] ว่า ว่า มายสอด อ่อง ดี +[17:15:50] how the future will be look like especially with buildings tough +[17:15:53] I think from like the season +[17:15:54] perspective +[17:15:56] we're trying to set up the... +[17:15:57] uh, the... +[17:15:59] the template. +[17:16:00] skills. +[17:16:03] that can be reused with different products. +[17:16:04] and for each product. +[17:16:06] I would anticipate. +[17:16:08] I try to can build the customer skills. +[17:16:09] feeling there. +[17:16:12] customer base context. +[17:16:14] clicking on the actual... +[17:16:15] like the product, the code. +[17:16:19] So it's more like you have the customer skills. +[17:16:23] and it is calling another underhood custom skills to building all the blocks. +[17:16:24] all together. +[17:16:26] So that's what I can imagine for the... +[17:16:29] how the future will grow for what we look like. +[17:16:35] I don't know if the heat hentai wants to add. +[17:16:36] any comments. +[17:16:43] I guess now we can go to the the brainstorm session. +[17:16:50] I just share that this stock, this sheet in the growth channel, I think the original one. +[17:16:53] It's very overloaded. I don't think it's +[17:16:54] usable at the moment. +[17:16:57] So I think we can just add in ideas over here. +[17:17:01] For these sessions, I think we can all focus on the. +[17:17:04] the B part like the team operations and then +[17:17:07] the system ideas that would get which can apply to +[17:17:08] all growth channels. +[17:17:15] but feel free to either at the top or at the bottom. +[17:17:16] whichever +[17:17:17] you want. +[17:17:21] I guess we can stay at roughly like 10 minutes. +[17:17:23] for all the folks that are in our days. +[17:19:22] By the way, sorry for interrupting, just a quick question. +[17:19:25] Is it possible to create an agent? +[17:19:27] And if so, someone can add it into the sheet. +[17:19:29] since I'm not. +[17:19:31] part of the growth. +[17:19:32] Anymore, but... +[17:19:36] I think it's potentially a good idea. So if... +[17:19:40] Is it possible to create an agent that can... +[17:19:44] that can kind of go through Facebook, Newsfeed or others. +[17:19:46] channels and then +[17:19:48] kind of analyze +[17:19:49] the +[17:19:51] the units that are existing. +[17:19:53] there and kind of +[17:19:55] create ideas on its own for +[17:19:59] potential new units that could be tested or added. +[17:20:01] That might be interesting. +[17:20:09] I think it can. +[17:20:12] search the codebase for the unit. +[17:20:15] but you had to give a very specific prompt. +[17:20:17] Like when you see unit, what does it mean? +[17:20:18] You have to. +[17:20:21] You have to be very specific. +[17:20:25] Well, like explore new directions that the team could take. +[17:20:27] and kind of recommendations. +[17:20:36] Like, let's say Fifu was one idea, right? So can it come up with new ideas on its own? +[17:20:38] Thank you. +[17:20:41] So if I were saying, for example, like that we +[17:20:42] services on. +[17:20:48] and if there is other use cases for example like on thread or on like a +[17:20:50] that's the only person you're always had. +[17:20:51] that's what you mean. +[17:20:54] Oh, yeah, or let's say +[17:21:00] Yeah, exactly. Either that or maybe, let's say, we give it a prompt and say, analyze. +[17:21:03] best performing. +[17:21:09] Units on like it you can go into hivetables you can search for CTR for other metrics. So, you know it +[17:21:17] And then based on the knowledge that it collects, it can potentially see gaps, like what you mentioned, maybe. +[17:21:20] in WhatsApp we can do this. Like what is, what can we... +[17:21:23] Um, like, uh... +[17:21:27] How can we utilize marketplace recommendation or marketplace to products? +[17:21:35] on other channels like what would be new directions, new things to try, what would be interesting things based on data analysis that it does on its own. +[17:21:37] Right on things that you know already work +[17:21:45] Right in other channels like what are we potentially missing can it identify gaps in other channels or? +[17:21:46] you know, find recommendations. +[17:21:56] I don't know, it's just a thought. +[17:22:05] This is all new to me, so I have a feeling that it can do it, because it's, yeah, that's my experience so far, but I just... +[17:22:08] I have a feeling we can potentially utilize it for even. +[17:22:10] moving beyond like +[17:22:13] creating the idea ourselves. We can probably... +[17:22:16] have it do an analysis. +[17:22:18] and then generate ideas on its own. +[17:22:31] I think an ai.. an ai.. +[17:22:31] the marketplace quality is already +[17:22:36] 以上。 +[17:22:36] It's basically analyzing the course. +[17:22:38] different channels. +[17:22:40] بوارنن +[17:22:41] quality field turning applying to and they can +[17:22:43] compare the difference. +[17:22:46] Let's get to this possible tool. +[17:22:49] so apply one treatment to further channels. +[17:22:50] if that's what you. +[17:22:56] Yeah, for example, let's say if we ask, identify... +[17:23:04] channels or places where we could, where there's a gap, right? Where we're not recommending marketplace products but you think that... +[17:23:06] You know, it could be useful. And then basically... +[17:23:13] they can basically go into its own data analysis. And for example, on threads, they can find that, I don't know, 60% of the threads. +[17:23:25] Population are also users of marketplace so they have high commerce intent or whatever whatever and then based on just you know the internal analysis they can maybe find out gaps like what +[17:23:29] What would be an interesting area to explore? What would be an interesting product to try? +[17:23:30] Um +[17:23:31] Yeah +[17:23:40] And then maybe even connect it with like other agents that can automatically set up a QE automatically, you know, write the code automatically do these. +[17:23:41] potentially even be like us. +[17:23:45] self-iterating loop of finding out. +[17:23:48] new areas and kind of self +[17:23:49] exploring. +[17:23:51] I don't know. +[17:24:00] 這是上星的Asian Contra +[17:24:06] I just want to know if you give it a top guess to open it will start to hallucinate +[17:24:13] So if you have some very concrete table that you can provide to the agent, tell agent, hey, this is a table for... ...more. +[17:24:14] on QP units. +[17:24:15] on your speed. +[17:24:19] I'll understand what they are, what type of unit it is. +[17:24:21] with the performance and uh... +[17:24:25] make suggestions. It will try to give you some suggestions. +[17:24:26] I just uh +[17:24:30] al +[17:24:31] When the problem is to open it, you cannot fully trust it. +[17:25:15] 以上。以上。 +[17:25:17] TRs are adding a lot of ideas, but if you're just adding ideas, you're not pointing to work on it. +[17:25:20] or you're helping inviting other engineers to join. +[17:25:23] Maybe in the oner section you can add a plus sign. +[17:25:24] seeing it up +[17:25:27] So that other people can enjoy and enjoy. +[17:25:29] Because otherwise I see that up. +[17:25:32] She probably won't have more than 10 projects. +[17:25:43] Yeah, we know that AI will turn on your efficiency, so we don't we don't know whether we'll be able to finish our 10 +[17:25:48] I think it intended to. +[17:25:49] working together. +[17:25:52] We've got a folksy sound for life. +[17:25:56] all of that information I probably just will quiz you about together. +[17:26:10] And then I got to know that there is a... +[17:26:13] call the scale that's literally called a 10x engineer. +[17:26:19] We might be using 54. +[17:26:22] really interesting. +[17:26:26] the most important thing about the +[17:26:29] I use it all the time. A run introduced it to me I think two weeks ago. +[17:26:40] That's about naming your skill. Next time we will create a one called 100 Axis. +[17:26:46] I think it may be a lot. +[17:26:49] search for marketplace I think it's already had like 10. +[17:26:56] the dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[17:26:56] from marketplace teams. We'll ever be open to that. +[17:27:06] We should have a screening, maybe. +[17:27:07] the same way we are. +[17:27:08] wrong the guy +[17:27:12] I mean, I don't know if anyone want to... +[17:27:14] Share some thoughts. +[17:27:18] Maybe I can go and then whoever wants to share. +[17:27:20] I think to me I feel like +[17:27:22] we do have a lot of... +[17:27:27] Repeat the pattern, especially when we're on a new service. +[17:27:30] from an infrared perspective, from a new channel perspective. +[17:27:32] So I would say. +[17:27:36] Previously we really depend on engineers, like understanding of the code base. +[17:27:40] the +[17:27:43] and they're not there to practice to apply or to implement those solutions. +[17:27:44] But now we can. +[17:27:48] just creating a skills or a bunch of skills and then allowed them to run skills. +[17:27:50] that the skills will actually... +[17:27:52] just creating all those. +[17:27:56] code for them, which is more maintainable and more scalable. +[17:27:57] in that sense. +[17:28:03] Especially for like infrastructure services, it's like a completely different code base. +[17:28:05] So it's harder to find a restaurant. +[17:28:09] I think with skills, I think we're opening up very easy. +[17:28:11] practice for them to true. +[17:28:11] too fast. +[17:28:12] Bye. +[17:28:15] really fast to spin out something new. +[17:28:22] So I put some of the ideas here like a covering like retrieval ranking cash +[17:28:25] capacity features stuff here. +[17:28:38] Over the fence. +[17:28:41] most wallpaper +[17:28:45] Maybe first. +[17:28:48] Yeah, I have a feeling that for some of these +[17:28:50] info related. +[17:28:51] Um... +[17:28:52] Agents? +[17:28:56] that they will potentially rely on +[17:28:57] agents or skills +[17:29:00] built on the on the infraside. +[17:29:01] So... +[17:29:02] um +[17:29:11] So they will be like actively communicating between each other, right? You would basically reach out to one of the agents maintained by by Infra. +[17:29:13] And... +[17:29:15] and the agent would provide the result. +[17:29:17] that your agent is. +[17:29:18] asking for, right? +[17:29:19] basically +[17:29:26] Yeah, because essentially what Hente said previously, we can't really +[17:29:28] go too broad +[17:29:29] Um... +[17:29:33] with any of these, right? So we have to chop them up into... +[17:29:35] into like they have to be the... +[17:29:36] the PLC. +[17:29:40] the +[17:29:41] of what they're dealing with. They have to be like the subject matter expert. +[17:29:45] So you basically, you know, you +[17:29:49] Your agent reaches out to another agent and that agent provides expert guidance. +[17:29:53] from that specific file or area. +[17:29:59] Yes, he does. Could you help us work together within your protein mask? +[17:30:00] some of the variations. +[17:30:03] Does this look like a +[17:30:05] happen to be building those agents with care. +[17:30:07] our agent can be calling their agents. +[17:30:10] Yeah, absolutely, absolutely. +[17:30:17] I think this will in the end be like one big web of agents, right? +[17:30:24] Your agents call our agents, our agents call your agents, and whatever we're all calling each other's agents, but I think it's going to be beautiful. +[17:30:29] I think we're at time. +[17:30:36] Thank you everyone for joining the session. Please continue to put in ideas and work on ideas. +[17:30:38] Yeah, good job team. +[17:30:40] At the same time, we are +[17:30:43] trying to get out volunteers to become our... +[17:30:45] domain expert of certain AI tools. +[17:30:48] So if you want to specialize in one kind of AI too. +[17:30:50] and uh +[17:30:50] Bye. +[17:30:54] basically be the go-to person so team members can reach out to you for guidance. +[17:30:57] please sign up. We will be sharing the link very soon. +[17:31:11] Okay, this is me again, Adolfo. +[17:31:15] And this is the end of my working. +[17:31:16] hours. +[17:31:19] So... +[17:31:21] Yes. +[17:31:22] process all +[17:31:24] Did they meet in? +[17:31:26] and have me +[17:31:31] have the learnings into documents or the memory so +[17:31:33] We can talk about these later. +[17:31:46] go for a service +[17:31:50] Go for it Jarvis! +[17:39:13] What do you mean, one way or another, do you see Brian in front of me? +[17:39:15] Thank you. +[17:39:21] Thank you. +[17:40:23] Let me see. +[17:40:38] This station will be to build. +[17:40:52] So like, like putting my... +[17:41:32] Before he dies. +[17:42:31] Yeah. +[17:42:48] Oh, you have one right here. +[17:42:55] I got one for the... +[17:42:56] Okay. +[17:43:08] the +[17:43:10] अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अगर अ� +[17:43:47] to the local.. you +[17:43:47] on your face or your hand? Both, both! +[17:44:00] love you +[17:44:14] the most important thing about the +[17:44:16] That's what she does. That's what she does. She does. She does. She does. +[17:44:20] You're doing a lot better. +[17:45:00] the +[17:45:49] Ready? Yeah. So I sit, I bet four. And three. +[17:45:52] Well, you know what? +[17:46:07] buenas insanas Subtítulos por la comunidad de Amara.org +[17:46:08] So what's up with the time? +[17:46:11] when to visit his cousin in the country. +[17:46:21] That's why she's called a crook. The culture knows who is happy to see is not happy. +[17:46:26] We got your mouse. +[17:46:28] จินแล้วครับ บายพูด +[17:46:33] But he was happy to share what he had with his city monks. +[17:46:44] Yes, the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the most the +[17:46:45] Yeah, riddle their families and they go to a room. +[17:46:49] As you know, it's turned out... +[17:46:50] Turn up. +[17:46:53] He snokes at the country. +[17:46:58] And he invited his classmate to have dinner with you in the city. +[17:47:17] Marble Bizer, baby, I smell. I smell? No, sir, I said that done. +[17:47:18] It's alright. +[17:47:21] Sit up for the signature. +[17:47:24] Yeah. +[17:47:29] Teach the city most。 you +[17:47:32] Yes Yes +[17:47:41] Now, spray into the gray dining room. +[17:47:49] We're still on the take-off. +[17:47:58] I'm thinking, all that was nice. I'm the other side of the night time. +[17:48:00] Suddenly they hurt. +[17:48:28] Both my friends. +[17:48:32] more than wax. Yeah, we're going to step out to do the feet there. +[17:48:37] the country monsters you meet up Bye. +[17:48:39] We'll go back to the country. +[17:48:43] What is good? +[17:48:45] Peace, fire, action. +[17:48:48] What good he's fine for. +[17:48:50] If you can't enjoy it. +[17:48:53] It is much better to clean with a piece. +[17:49:02] That's all there's to it. +[17:49:03] Yeah. +[17:49:14] Okay. +[17:49:29] Oh wow. So you don't need a, you don't want a You want me to... I can use three look, three look here too. Yes. Don't touch the arm raised. That was the last one we did. +[17:49:31] Okay, thank you. +[17:50:02] French. You're done forever? Yes, we're French. Okay, and now listen, okay, so me and my mom... +[17:50:17] សʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟʟ +[17:53:25] Oh, yeah! And then I heard the shower coming. I didn't know where to go. +[17:58:19] Whoever is coming is... +[17:58:21] Yeah, we'll be right back. +[17:58:26] But that was a good spot. Now, great, it's a good lecture. +[17:59:51] Yes please let the food in the food dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[17:59:52] Wait, what? Your turn, Grace. I was right here. Right here. What's? +[17:59:58] the +[17:59:59] Don't talk about it. +[18:03:27] right then." I turned the +[18:03:30] I don't hear because you don't know +[18:03:32] So we go +[18:03:35] Why is the baby in here? +[18:03:50] Thank you. +[18:05:08] ʕ ʔ ʔ +[18:05:53] Now we're playing hide and seek. +[18:06:06] the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey the honey honey +[18:06:09] I need a S center to do three, two, one. Yeah. +[18:06:11] Good times are good. +[18:06:13] And that's the +[18:06:19] And you didn't like it, but we're still gonna ask that. +[18:06:22] Yes? Yes? Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes +[18:06:22] Yeah. +[18:06:41] the print the lentis the lentis +[18:11:06] Yes Yes +[18:11:07] I'm not going to tell you. +[18:11:51] in the earth. +[18:11:54] Where did you hide so far? +[18:12:02] the heavens the heavens +[18:12:07] I never thought it would be. It was going to be three days and we were going to be like, Lea, don't just toss. Yeah, it was going to be epic. +[18:12:17] It's okay, yes. +[18:12:47] Oh +[18:13:18] this is the most damaged dot the dot the dot the dot the dot the dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot dot +[18:13:26] I'm gonna turn again. Let's do it, Leigh. Alright, this is how it has to be then, Grace. +[18:13:28] She's gonna count. +[18:13:30] to how long do you need to hide? +[18:13:35] So you can cut to 16 while you hide. +[18:13:40] And once you're hidden, then at least we'll go hide so you can't see him hiding. +[18:13:49] No, it's where he shaved nose. No, she didn't saw where he had it. He just... +[18:13:50] found here in Tommy. +[18:13:53] Oh, I don't feel like... It's like, see, always... +[18:14:01] right the heavens the heavens +[18:14:03] It's not a big one. It always does it. Okay, well don't do it this time Grace, okay? +[18:14:04] Okay, alright. +[18:14:08] She's not gonna do it this time, but yes, she does it this time. +[18:14:11] You come tell me and I'll give her a spank for discipline. +[18:14:15] Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! Yes! +[18:14:16] Okay. +[18:14:22] What's that around? +[18:14:27] I don't know why you're descending here. +[18:14:30] I don't know why you're defending me. +[18:14:35] Yes, the +[18:14:37] Cheers for it. +[18:14:47] I think we did it in like... +[18:14:49] Get on the way today. +[18:14:54] right +[18:14:56] I know we did all the +[18:14:58] types of agreement. +[18:15:02] Let me check it out. Open door, closed door. +[18:15:05] but we didn't get to the interactive, that's what you're saying? +[18:15:10] Can we watch the end of the video? +[18:15:14] or just start over from that part. +[18:15:22] In fact, none of these methods. +[18:15:24] That was you, right? And then me was... +[18:15:27] If we're not walking in agreement, there'll be a lack of peace. +[18:15:37] I think we were pretty good, we could just like start painting. +[18:15:40] interactive. +[18:15:58] Roeddais wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi wi +[18:15:59] I'm trying to do it right. We have volume this time. +[18:16:01] tem que dar +[18:16:02] Ah, well that's a deal. +[18:16:07] Good day to all. +[18:16:14] And then when we're struggling... +[18:16:19] Over the window." +[18:16:24] the the +[18:16:26] I was going to ask that if it didn't work. It's to the right place. +[18:16:31] Don't touch his heart, son. +[18:16:35] God bless, come on. +[18:16:39] Yes, let's go bros +[18:16:41] It'll work. +[18:16:44] I'm kidding, that's what made the chain should kill it. +[18:16:53] So, this portion they see that we can solve more. +[18:16:56] as well a blessing for this country. +[18:16:59] I'm in the libel. +[18:17:02] You have to put your holy paper on our bodies. +[18:17:08] Thank you that this material right was, and we thank you that. +[18:17:09] Thank you, guys. +[18:17:12] the most important thing about the +[18:17:19] Т lent +[18:17:28] the most ever heart.. +[18:17:29] in the wrong place +[18:18:15] Maybe we should just finish this one and skip date night. Tomorrow I'll do the next one. +[18:18:17] Yeah, probably. +[18:18:20] Use the rest of the night to work on our Texas. +[18:18:22] because I feel like our brain is just elsewhere. +[18:18:26] That's too long. +[18:18:28] Thank you. +[18:18:30] Have they come in here? +[18:18:38] They're hard to ever do, but I mean, I can just say we can't look at this week, guys. I mean, just be honest with how we're going to work. +[18:18:47] straight to interactive? +[18:18:56] the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important +[18:19:07] No, I think we need to start. +[18:19:14] It doesn't go straight. +[18:19:17] the heavens +[18:19:20] I think this was this +[18:19:24] A no ser, no hay. Esa es la razón por que cuáles los metidos anteriores es el correo. +[18:19:26] You gonna do that part? +[18:19:28] We'll skip this pair of things. +[18:19:55] You want to do it like a ping-pong? +[18:20:05] Trying to understand what is the reason for the problem. +[18:20:07] Thank you. +[18:20:39] I think it's saying none of these is a correct method. +[18:20:42] what happens when. +[18:20:45] We rely on these methods. +[18:21:05] Okay, so let's do that. I'll ask you a question you answered. I'll ask you a question you answered. +[18:21:11] the +[18:21:11] Let's go. Copy. +[18:21:13] Let's try. +[18:21:17] Entonces +[18:21:20] ¿Cuál de los métodos anteriores es el correcto? +[18:21:26] nuestro wallpaper no sucede al wallpaper ¡Suscríbete! +[18:21:28] Y que sucede si dependemos de estos acuerdos equivocados. +[18:21:29] Thank you. +[18:21:32] Lo que va a suceder es que no nos va a tener. +[18:21:33] Perfecto. +[18:21:35] y es el verdadero acuerdo de Dios +[18:21:42] let's let the。 +[18:21:44] 2 y 3 que deseen algo y no lo consiguen +[18:21:53] the most honey you've ever had in the heavens." +[18:21:56] y no pueden obtener lo que quieren. Reñen ya sin guerra. Lo tienen porque no piden. Y cuando piden no reciben, porque piden con malas intenciones. +[18:21:58] para satisfacer +[18:22:00] sus propias pasiones. +[18:22:06] Entonces, si no estamos caminando en el acuerdo que Dios da... +[18:22:08] no tenemos paz en la casa. +[18:22:15] y ese es acuerdo, abre la puerta para que el enemigo destaca +[18:22:20] Si, y que sujan frases como te lo dije, o siempre piensas que tienen la razón. +[18:22:24] Ahora vamos a tomar 3 minutos. +[18:22:25] para que cada pareja. +[18:22:28] Platicue acerca de esta pregunta. +[18:22:31] ¿Cuántos de estos métodos? +[18:22:32] Asusau. +[18:22:34] para tomar acuerdos. +[18:22:39] tomen tres minutos para platicar acerca de cuáles métodos +[18:22:42] de esos anteriores han usado +[18:22:45] y cuáles han sido los resultados en su casa. +[18:22:59] Yeah. But I'm so happy for them. +[18:23:03] like I want to try to do my best because they're getting so much out of it. +[18:23:54] Şeye wi olay hakkını çekiririz +[18:24:00] We forgot our X thing, right? +[18:24:08] the most important thing." +[18:24:09] As soon as I say that we went fighting, I say should we win. +[18:24:33] This thing with the symphony is kind of nice but it seems like a lot. +[18:24:45] armonizar +[18:24:48] No, let's reschool tomorrow. +[18:24:56] Yes, let's go. Yes, let's go. +[18:25:07] Make a cousin group chat today. +[18:25:11] My cousin Francis got engaged. +[18:25:12] Yeah. +[18:25:14] You saw it, where? +[18:25:19] Yes please。 +[18:25:21] So you know before me? That, that, that time, that's right too. +[18:25:25] It's not your job. It's not my job. She looks different. +[18:25:27] 온다 +[18:25:32] Be together forever. +[18:25:38] I'm pretty sure it's a yacht. I want you to choose yacht life. +[18:25:44] Perfectly nice at the wedding. +[18:25:49] I think I'm mostly just being prejudiced. +[18:25:52] I guess she looks at the guy that I'm taking out. +[18:25:53] I don't trust those guys. +[18:25:57] You ready? Yep. +[18:26:09] the the the the the the the the the the the the +[18:26:09] No. No. Ya estamos devueltas. +[18:26:12] the +[18:26:13] See you. +[18:26:16] formas que podemos deber, no son nada +[18:26:18] de forma correcta en la que le debemos buscar un acuerdo. +[18:26:20] Entonces, cuáles? +[18:26:24] no por no es correcto cuál es el acuerdo que podemos obtener +[18:26:25] ให้ละมาเลยนะ +[18:26:28] que Dios nos enseña ahora de la manera que Dios quiere. +[18:26:33] el primer versículo que vamos a estudiar está en Mateo +[18:26:35] Yes, you're welcome. +[18:26:40] Yes you most right. Yes you most right. Yes you most right. +[18:26:41] que descrima el valor del acuerdo, diciendo +[18:26:44] y dos de vosotros lo ponen de acuerdo en la tierra. +[18:26:47] acerca de cualquier cosa que pide +[18:26:51] le será hecho por mi padre que estar en los cielos. +[18:26:56] y entonces aquí Mateo de 78-79. +[18:26:57] Allah +[18:26:59] de Armonisar +[18:27:04] the +[18:27:05] y esto se puede ver como... como se... +[18:27:14] Is it the right thing to start the experiment? +[18:27:17] cuando escuches una sinfonía. Almarizar significa escuchar diferentes tonos, varios tonos, al mismo tiempo. +[18:27:19] una orquesta sinfónica +[18:27:21] Hay muchos instrumentos. +[18:27:23] y no todos tocan la misma nota. +[18:27:27] aunque están puesto de acuerdo y están tocando la misma canción +[18:27:31] cada instrumento toca parte de diferentes sonotos diferentes. +[18:27:34] You could have lost him. +[18:27:35] entonces están armarizando +[18:27:37] aunque no es la misma nota +[18:27:40] el sonido que produce cada instrumento +[18:27:41] En conjunto. +[18:27:43] hace que él... +[18:27:45] los que están escuchando +[18:27:47] è un po' di meraviglioso +[18:27:49] a lo que se escucha organizado +[18:27:50] Yeah, okay +[18:27:53] resulta en música. +[18:27:55] a la aldea y en emociones +[18:28:02] y de hecho cuando leemos ese versículo en la versión amplificada +[18:28:04] Eso es lo que usa. +[18:28:06] para tomar acuerdo, armonizar. +[18:28:08] con más la sifonía. +[18:28:09] Entonces +[18:28:14] estamos viendo cuál es la idea que Dios tiene acerca de tomar acuerdo +[18:28:16] Spero. +[18:28:18] ¿Cómo es que...? +[18:28:22] aprendemos a hacer las cosas de la manera que las hace Dios. +[18:28:28] Pues, primeramente tenemos que recordar que nuestro acuerdo es un acuerdo de tres partes. +[18:28:32] porque siempre debemos estar buscando hacer la voluntad de Dios. +[18:28:34] Y siempre debemos de... +[18:28:37] ponernos en acuerdo con Dios. +[18:28:41] y no al revés convencer a Dios que se ponga de acuerdo con nosotros. +[18:28:46] Entonces cuando los dos están buscando la voluntad de Dios +[18:28:47] Los dos +[18:28:50] aportan lo que Dios les ha dado. +[18:28:58] para buscar lo que realmente para ponerse de acuerdo con Dios Si, eso es como el testimonio que vemos +[18:29:00] e de que a sua força estava na terra. +[18:29:01] Good, see ya. +[18:29:04] Não pode ser funcionado por mim mesmo. +[18:29:07] Te conozco a ese justo, a ese justo. +[18:29:10] porque no buscó me voluntad +[18:29:12] sino la voluntad de mi padre +[18:29:15] it might be. If I was laying in one. +[18:29:16] 530 +[18:29:17] Thumbnail at 1. +[18:29:27] 4, 10 y 4, eso dice +[18:29:28] mi comida es que hará la voluntad del que me envió y acabe su obra. +[18:29:31] ese punto tan importante para eso. +[18:29:35] hacer la voluntad de espar y así también debe ser para nosotros. +[18:29:38] el acuerdo que le debamos alcanzar +[18:29:41] es el post gráfico y la voluntad de Dios en nuestras vidas. +[18:29:45] Si necesitamos crecer en nuestra conciencia de que Dios +[18:29:48] tiene una voluntad para todo lo que hagan sus hijos. +[18:29:51] en los apuestos a la tierra con un propósito. +[18:29:53] Entonces... +[18:29:58] en todas las cosas en vez de buscar lo nuestro, en vez de buscar. +[18:30:04] the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the love the +[18:30:05] el Cónic escucha nosotros, debemos de los dos buscar escuchar. +[18:30:06] la voz de Dios. +[18:30:12] los dos buscar entender cuál es el propósito y la voluntad de Dios para esta situación. +[18:30:16] cuando hacemos eso. +[18:30:21] tenemos la asegurancia de las promesas en la Biblia, te quedo de unos respaldas. +[18:30:26] Un ejemplo de eso viene en el primero de Juan 514. +[18:30:30] que dice esta es la confianza. +[18:30:32] que tenemos a ver. +[18:30:36] que si pedimos alguna cosa conforme a su voluntad +[18:30:40] No soy, si sabemos quién no soy. +[18:30:42] en cualquier cosa que pedamos +[18:30:45] Sabemos que tenemos las peticiones +[18:30:46] que la hayamos hecho +[18:30:52] Y también tenemos la promesa de Mateo 18-19 como ya leímos. +[18:30:54] que cuando dos están de acuerdo. +[18:30:58] en esa sinfonía, de acuerdo a la vuelta de Dios. +[18:30:59] lo que piden +[18:31:01] será hecho por nuestro Padre, no cielo. +[18:31:08] That was in back then? Yeah. +[18:31:46] Just separate it? I don't think it's separate. +[18:31:49] Or easy. +[18:32:00] So this is basically saying like if there is a little bit of a lag it should work. +[18:32:04] So we just start with that part. +[18:32:08] Entonces como podemos... +[18:32:12] com a seta lentada。 +[18:32:14] Y hay ciertas cuestiones que... +[18:32:18] No son muy difíciles de saber cuál es lo que contamos de Dios porque la palabra es clara. +[18:32:20] Next, let's do this. +[18:32:23] por ejemplo la palabra dice +[18:32:32] no tenemos que preguntarnos a nosotros mismos y Dios quiere que testifiquemos de que traigamos a otros, a Cristo. +[18:32:36] y que conozcan la verdad de por qué la palabra es clara en lo dice +[18:32:40] Marcos 1615, Mateos 1819 +[18:32:42] I neću svuna ovoču. +[18:32:44] que eso es la voluntad de Dios para nosotros. +[18:32:46] Otro ejemplo +[18:32:49] que ya sabemos muy bien es que Dios quiere que +[18:32:51] the +[18:32:52] queríamos a nuestros hijos +[18:32:53] para conocer a Dios. +[18:32:55] para caminar con Dios. +[18:33:01] Eso es muy claro también en la palabra. Entre otro nómio 11-19 +[18:33:05] 153 +[18:33:06] y 5, 1, 6 +[18:33:11] Otra cosa que nos insiste la palabra... +[18:33:13] claramente es que debemos de... +[18:33:20] pertenecer a una familia espiritual. Debemos reunirnos regularmente. +[18:33:28] Como vemos en hebreos 10-25, hechos 2-42, hechos 2-46. +[18:33:32] Entonces hay varias cosas, estos son tres ejemplos nada más. +[18:33:35] pero hay muchas cosas claras en la Biblia. +[18:33:38] donde no tenemos que buscar +[18:33:40] cuál es la voluntad de Dios. +[18:33:41] debemos más bien. +[18:33:43] someter nuestras decisiones. +[18:33:47] a lo que ya está escrito y ya sabemos que es la voluntad de Dios. +[18:33:54] Pero ¿qué hay entonces de aquellas cosas que no están tan chulas en la Biblia y que no tenemos una dirección exacta? +[18:33:55] Yeah, that was a good one. +[18:33:57] como por ejemplo +[18:34:00] Sabemos que debemos de testificar y evangelizar. +[18:34:03] Pero cuando? ¿Cómo? ¿A quiénes? +[18:34:04] ¿Con qué palabra? +[18:34:11] 以上。 +[18:34:16] the +[18:34:17] I thought you would like bounce off me for the next one. No, this is the next one, yeah. +[18:34:25] Like I was doing. We bounce with the questions you said. Eventelize, I said. Raise your children godly so then. +[18:34:27] I said, evangelize, I thought you'd say like. +[18:34:30] the next thing about the children. +[18:34:31] No, I don't care. +[18:34:32] Good night. +[18:34:35] I think we'll finish and then I just say bye. +[18:34:40] the most important part the most important part +[18:34:50] And then I said this I didn't read that but so it was like Where and how should we have asked that? How do we deal with this specific problem of having our children? What church should we join? +[18:34:56] más acá en el toro, entonces, con otro toro. +[18:35:03] wi ok +[18:35:09] Okay, I'll do the questions and you can do the... +[18:35:12] Tenemos que aprender a buscar la comandada de videos. +[18:35:21] We need to know how to seek God's will for these types of situations in our lives. +[18:35:28] It's not enough to think this choice looks better or that choice seems wiser. God has a specific will for each situation we need to learn from. Yeah, but I'm... +[18:35:37] おこもたった +[18:35:44] the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important thing in the most important +[18:35:46] determinados problemas que tienen nuestros hijos, que iglesia deberíamos de concregarlos, a que escuela o cochejes deberían de ir los hijos. +[18:35:53] o si deberíamos cambiar de casa o tomar esa propuesta de trabajo o movernos de... +[18:35:54] I'm gonna see you there. +[18:35:56] Qué trabajo vamos a tomar. +[18:36:03] Ese tipo de pregunta específica no viene contestada en la Biblia directamente. +[18:36:06] Entonces en esas situaciones necesitamos determinar. +[18:36:10] o más bien discernir y desarrollar. +[18:36:12] ¿Cómo escuchar la voz de Dios? +[18:36:16] 잇 +[18:36:20] determinar su voluntad. No es suficiente simplemente tomar acuerdos +[18:36:21] basada en la lógica de... +[18:36:24] cuál decisión parece ser más sabio +[18:36:28] cual nos conviene, cual nos gusta +[18:36:31] esos pueden ser factores pero el +[18:36:37] El factor primordial tiene que ser cuál es la voluntad de Dios y tenemos que creer y saber. +[18:36:41] que Dios sí tiene una voluntad y un propósito. +[18:36:43] nuestras vidas. +[18:36:58] Thank you. +[18:36:59] should I focus on this and join it? +[18:37:01] Okay. +[18:38:52] dann ist es kein dot.. da muss was kein dot +[18:38:53] ¿Cuáles vamos a notar, mi Dios? +[18:38:55] en una situación como esta +[18:38:58] es algo que debemos hacer juntos. +[18:39:02] porque así como dice Mateo, cualquier cosa +[18:39:03] ¡Qué pena! +[18:39:05] the right +[18:39:07] No basta con que uno de los coñoguees +[18:39:09] que busque a Dios, efectiva de Dios. +[18:39:14] sino que Dios nos ha dado una sabahuarda contra errores. +[18:39:19] al permitir que los dos ojejamos +[18:39:21] y confirmemos el sentido del mío. +[18:39:32] Sí, porque de otra manera estarías usando una versión espiritual. +[18:39:35] de lo que vimos de los acuerdos humanos de... +[18:39:40] de depender de uno solo para tomar la decisión y llamarlo acuerdo. +[18:39:46] No podemos decir el hora más que yo así que que él tome la sesión, que él busque la sesión. +[18:39:52] sabemos que Dios donará a los dos pues +[18:39:56] para que una pareja pueda armonizar de esta forma con la voluntad del dios +[18:39:58] ambos tienen que escuchar +[18:40:01] Yes +[18:40:05] the most important thing about the +[18:40:05] Y creer que Dios les va a dar... +[18:40:11] al escucharlo al lentir +[18:40:11] So, I'll end that. +[18:40:14] Dios nos dice en su palabra que... +[18:40:17] Cada uno de nosotros podemos oír su voz. +[18:40:24] Entonces no debemos de dejar de engañarnos a nosotros mismos diciendo no, yo soy el que puede escuchar a Dios. +[18:40:24] Four. +[18:40:28] quizá también venga ya no se decir no ya no puedo escuchar adiós +[18:40:34] quizá a tu respuesta, no, yo soy muy nuevo y no tengo las cosas de Dios todavía. +[18:40:34] ¿No se si? +[18:40:36] que buscamos con fe +[18:40:40] y confiar que el Señor realmente puede dar a ti. +[18:40:43] Asher sal herself'ı bul heavens.. +[18:40:45] Y eso me recuerda que... +[18:40:51] a veces en una pareja hay uno más experimentado que otro de las cosas de Dios +[18:40:52] But +[18:40:55] En Santiago dice que... +[18:40:57] Todo el barco es navegado por él. +[18:40:59] ¿Cómo se llama el timbre? +[18:41:02] la cosita pequeña que +[18:41:03] vamos a hacer así +[18:41:07] y afecta toda la dirección del barco. +[18:41:13] No sé cómo se llama. Pero opone el volante del carro. +[18:41:18] Es una sola parte, es una parte pequeña del carro. +[18:41:23] y no necesariamente más importante el motor es lo que hace que el carro va. +[18:41:25] pero hace mucha. +[18:41:28] diferencia en la dirección del cartón. Entonces... +[18:41:31] Recuerden que cuando vimos lo de... +[18:41:33] los roles y de sinergía. +[18:41:40] vimos que Dios pone dones en la esposa y pone dones en el esposo y permite que la esposa vea +[18:41:43] ciertas cosas, ¿entiendes ciertas cosas? +[18:41:46] recibe ciertas cosas y les puso otras cosas. +[18:41:48] Entonces, no... +[18:41:51] No se, menos presia, no se juzgen por... +[18:41:55] ¿Cuántos años tiene el señor? ¿O qué dones tiene el chino? +[18:41:58] Permiten que cada uno aporte lo que Dios le da. +[18:42:00] y cuando es de parte de Dios. +[18:42:02] Does this look like a +[18:42:09] Makir ya.. Is.. Terima kasih kerana menonton! +[18:42:11] sinfonía de la cual estamos hablando. Tampoco que menos presión al otro. +[18:42:13] We're glad you're all okay. +[18:42:16] Yo recibo esto, como son las cosas. +[18:42:18] ok, vios +[18:42:21] love the! the the the the the the the the the the the the the the the the the +[18:42:23] A dos dois, ao mesmo ponto, até onde devemos caminhar. +[18:42:26] y esa es la parte importante de buscar juntos. +[18:42:28] ¿Cuál es esa dirección? +[18:42:32] Over the fence the +[18:42:33] al mismo tiempo es importante que ambos +[18:42:34] everything. +[18:42:37] imponer su propia opinión. +[18:42:38] cuando +[18:42:40] están buscando orar. +[18:42:44] es muy fácil cuando borras en la mente +[18:42:46] horas en tu entendimiento. +[18:42:50] que oras con tu opinión, con cómo tú ves las cosas. +[18:42:51] Entonces... +[18:42:52] quando ora se l'ha menti +[18:42:54] es muy... +[18:42:59] ayuda mucho buscar escritura para quitar tu oración. +[18:43:01] al mismo tiempo +[18:43:04] the +[18:43:07] no es malo confesar tu opinión a Dios pero tienes que hacerlo en una manera humilde +[18:43:12] para someterlo, para soltarlo, para que Dios te conteste, para que escuches la voz de Dios. +[18:43:16] Sin embargo, recomendamos también que oren juntos en lenguas. +[18:43:18] en el Espíritu. +[18:43:21] porque así Dios te puede hablar. +[18:43:25] más allá de tu opinión, más allá de tu lógica, más allá de tu mente. +[18:43:27] y tu espíritu se haificado +[18:43:31] y tú puedes recibir de parte de Dios, tal vez lo que... +[18:43:33] en la mente no pudiste ver. +[18:43:40] puedes recibir el espíritu. Entonces también, especialmente si tienen un fuerte desacuerdo. +[18:43:45] o si tienen una opinión muy fuerte acerca de algún tema, es importante orar. +[18:43:49] no solo en el entendimiento, pero también en el espíritu. +[18:43:53] y permitir que el Espíritu Santo les hable y les desea visiones. +[18:43:55] sia scrittura +[18:43:57] en la manera que Dios te hable a ti. +[18:44:00] igualmente yo creo que si estás limitado el fe +[18:44:06] en ese momento no tiene esa idea de por dónde podría ser la oportunidad de Dios. +[18:44:08] y los lenguas también liberan. +[18:44:09] esa parte de ahí +[18:44:12] tener que encontrar una solución que tiene sentido. +[18:44:16] o algo que está de acuerdo con lo que puedes ver con tus ojos naturales. +[18:44:19] That's what they have all out. +[18:44:21] como hace para lo ignorar el entendimiento. +[18:44:22] Hey +[18:44:26] Permites que el Espíritu se conecte con Dios aún, porque no entiendes. +[18:44:30] lo que estás diciendo, los espíritus apunente con Dios se recibe. +[18:44:39] Не нужно wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper wallpaper +[18:44:39] Por ahí yo agrego algo que no viene en el libro, pero creo que también es importante que especialmente... +[18:44:42] ser para todos, pero especialmente para los que son nuevos. +[18:44:45] en las cosas de Dios en escuchar la voz de Dios. +[18:44:47] Anoten lo que reciben. +[18:44:49] Siapa? +[18:44:55] Is the Yes +[18:44:57] muy directo o si es algo más como el espiritual, una visión, lo que sea. +[18:45:00] 。 +[18:45:01] sometan juntos lo que reciben a... +[18:45:03] liderazgo. +[18:45:09] Sí, como pareja, tienen la responsabilidad de buscar a Dios. +[18:45:11] BESU PARTE! +[18:45:13] Pero yo también nos ha dado líderes. +[18:45:17] más experimentados en la fe, en la vivia. +[18:45:21] ayudarnos a... +[18:45:24] y discernir lo que estamos recibiendo de parte de divas. +[18:45:25] Entonces... +[18:45:26] na tingan +[18:45:32] Miedo de ir con sus líderes con los padres en la fe que Dios les ha dado. +[18:45:33] para someter esas cosas. +[18:45:34] ¡Tomen el! +[18:45:37] Tomen la responsabilidad de orar primero. +[18:45:39] y hablar uno con el otro. +[18:45:43] Pero, por ejemplo, don Fillo, aunque estemos muy de acuerdo en algo. +[18:45:46] No es punto de desacuerdo. +[18:45:48] casi la mayoría de las veces +[18:45:54] Los sometemos, en nuestro caso al apóstol libre, que es nuestra cabeza aquí. +[18:46:01] puede ser placaciones, puede ser la educación de los niños, puede ser un trabajo, casa. +[18:46:03] è una copra grande +[18:46:12] cuando tomamos igual del desayuno, que vamos a desayunar, pero cuando tomamos decisiones grandes, aunque... +[18:46:15] estemos de acuerdo entre nosotros. +[18:46:18] siempre queremos como esa +[18:46:20] extra protección +[18:46:26] de eso meter lo que creemos que hemos recibido de parte de Dios a los que Dios nos ha dado. +[18:46:30] entonces eso es como un paréntesis que no viene en el libro pero +[18:46:34] También puede ser una ayuda y como una red de seguridad para ustedes. +[18:46:35] Tiiä näissä partuu, että var. +[18:46:39] responsabilidad de orar y de buscar también entra +[18:46:48] y compartir entre ustedes lo que han recibido entonces cuando ambos creen haber recibido algo del Señor de acuerdo a esta decisión que están buscando tomar +[18:46:51] was that where the Istanbul is kind of in, right? +[18:46:53] deben de comprar las respuestas. +[18:46:58] Si Dios es el que habla, lo van a recibir dos cosas distintas. +[18:47:03] porque en Dios no hay confusión, como dice el Primario de Corintios 1433. +[18:47:06] y no va a dar más respuestas. +[18:47:08] diferentes haciendo que +[18:47:12] y ustedes serían como una persona de 12 años. +[18:47:13] And that's what I'll play with any of these. +[18:47:14] That's it. +[18:47:16] cuando se sienten ya recibido +[18:47:18] Vosca es muy comparada. +[18:47:20] lo que han recibido de Dios. +[18:47:22] si no es lo mismo +[18:47:24] que es el que tiene que seguir buscando. +[18:47:29] y ahí tengan cuidado otra vez de no juzgar el nuevo. +[18:47:32] No es que yo tengo razón y tú estás mal. +[18:47:36] sino que los dos necesitan seguir buscando. +[18:47:38] A veces sí es porque estás... +[18:47:41] pensando mucho en tu propia opinión todavía. +[18:47:42] y hay que seguir +[18:47:48] buscando lo que Dios está diciendo más allá de tu propia opinión, tu propia visión. +[18:47:50] Otras veces Dios +[18:47:54] parece que está diciendo dos cosas distintas +[18:47:56] Pero él tiene un plan. +[18:47:58] que ninguno de los dos ha pensado. +[18:47:59] and +[18:48:03] la manera que lo que Dios les está dando, la manera que armoniza. +[18:48:07] es en algo que no han considerado, no han pensado. +[18:48:11] Entonces, en ese caso también tienen que seguir orando. +[18:48:14] tanto en el espíritu como en la mente +[18:48:16] hasta recibir +[18:48:19] el plan de Dios que él da a los dos en acuerdo. +[18:48:23] Jetzt ist ein Prozess okay. +[18:48:25] puede tomar tiempo sobre todo así. +[18:48:28] Es la primera vez que está al nas. +[18:48:30] buscando al señal de esta forma. +[18:48:33] porque al principio puede ser que todavía no es tu propia opinión. +[18:48:34] for the sales. +[18:48:37] de estar influyendo en lo que podemos creer escuchar. +[18:48:41] mientras este sigue haciendo caso +[18:48:44] No hay que desesperarnos, hay que seguir. +[18:48:48] buscando ahorrar en el espíritu, buscando la dirección del Señor. +[18:48:51] o aferrándonos a la respuesta que... +[18:48:56] Oh, that didn't feel like 40 minutes. +[18:49:00] Was it? Yeah. No. +[18:49:05] No, aferrándonos a la respuesta que creemos. +[18:49:07] It's like we are receiving most of all. +[18:49:08] tener un corazón. +[18:49:10] abierto y disponible +[18:49:14] para recibir realmente que el Señor tiene para nosotros. +[18:49:15] We continue on. +[18:49:18] la conversación en comparado lo que recibemos. +[18:49:20] con amor +[18:49:22] como de cualicias impulsarnos +[18:49:25] Okay +[18:49:30] Lo mejor que podemos encontrar es la dirección que Dios quiera para nosotros. +[18:49:37] entonces vale la pena el esfuerzo de tener esa confianza que vamos caminando haciendo la voluntad de Dios +[18:49:41] Where were you then? +[18:49:45] I think I read from the top here. +[18:49:54] al cost heavens the most y continuo acoporando los puestos y volviendo a la rara. +[18:49:56] Okay, so now we're going to start with agreement reached this way is let us great of God for you to get that part +[18:50:00] Did you already do Luke 11.10? +[18:50:02] Yeah. +[18:50:08] Como al albergar al albergar al albergar al albergar al al albergar al al albergar al al albergar al al albergar al al albergar al al albergar al al al albergar al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al al +[18:50:09] like average this point I didn't read the +[18:50:22] अटीके अटीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीके टीक +[18:50:22] Yeah, I think we'll mention this tonight. I won't continue on. I won't watch him fallibly. +[18:50:25] It's not a burger. +[18:50:40] Okay Yeah +[18:50:52] Alright, so how do we know this is from God? One, we're in agreement. +[18:50:55] to wet peace +[18:51:00] Three in aligns with the word of God. +[18:51:35] before I would add again. +[18:51:36] leadership +[18:51:51] Alright, I can do that paragraph. +[18:51:55] So then you want to go with the hostess? Yeah. +[18:52:00] all the way to bringing God's will to pass on the earth. +[18:52:04] Might as well take a break. How's that? One more thing. +[18:52:10] But the husband for me is a big long career griff. +[18:52:12] Yeah. +[18:52:20] I might need your help. +[18:52:37] By achieving? Working out a lot. +[18:53:08] Makes me think we shouldn't have taken the Facebook job +[18:53:13] But I can't imagine that, I mean... +[18:53:15] That's... that's... +[18:53:17] decision to me is like +[18:53:22] one to this day to understand like +[18:53:24] Did we do right or wrong because? +[18:53:29] It feels wrong in the sense of like we didn't know that was our last year with Apostle John. +[18:53:33] I don't know how much we could have grown and known him and received. +[18:53:35] in that time. +[18:53:40] But then the other side of it is it feels so right in that like... +[18:53:42] We're so... +[18:53:43] Set up. +[18:53:45] to be here. +[18:53:46] Like. +[18:53:49] Which is so incredibly best. And everything like... +[18:53:51] falling into place now. +[18:53:54] Yeah, and all of it like... I always have to change the remote thing they want here, huh? +[18:53:56] We have mentioned. +[18:53:57] Come back. +[18:53:59] Just a house. +[18:54:02] I think the fact that Universal John said like God won't let me +[18:54:04] like prohibit you from going +[18:54:07] was like as close as you could get him to say +[18:54:10] Yeah, I like it. +[18:54:17] We went around, we took the decision. +[18:54:19] It's just like that was a hard one. +[18:54:31] Because of this thing of the deadline where it was like that? Yeah, I think that's like a complicated, like this is where it's like, no, not coming back to see until I have deadline. +[18:54:36] I got the end of the day there are things to do, like not making a decision is also taking a decision. +[18:54:46] I'm not going to be afraid to look this one over. +[18:54:51] the tack the most now the the the the +[18:54:51] You're making the decision not to go. +[18:54:58] And it could be like it cuts right there, but at the same time I understand it in the sense that like +[18:55:02] Look at the job compared to the house. +[18:55:05] The house had a deadline, not exactly a deadline, but the house had... +[18:55:09] a limitation of like if we don't buy it somebody else might +[18:55:09] Yeah. +[18:55:10] But +[18:55:13] We didn't rush into it just because we wanted it. +[18:55:16] God, push us to move quickly. +[18:55:19] My husband at the job, we were like, God. +[18:55:23] We're asking you to give us something else by this date. +[18:55:27] the the the the the +[18:55:30] Well, it wasn't just that the money right now. The job had a deadline. +[18:55:32] to respond. +[18:55:34] And we said, like, God. +[18:55:37] We think this is from you. +[18:55:45] So we're gonna take this job, if you don't give us something else, anything else, like give us any sign. So I feel like in that sense, it wasn't so much like. +[18:55:48] Oh, we're just like, we didn't pray about it, we didn't like... +[18:55:51] You know, we just took the job because the debt was a deadline, but like... +[18:55:55] Yeah, we were like, I think we were going to. +[18:56:05] possibility that was going to be the same second, so that's just like two more moments or something like that. +[18:56:12] I just think in general like we could keep growing in terms of what this chapter is teaching like I think we make a lot of decisions. +[18:56:17] trusting that like, oh we have been following God for so long that like +[18:56:18] This makes sense. +[18:56:21] But then again I feel like would you pray? +[18:56:29] Alright, well that's hard for me too, but I'll just do my best. I'll just make it subtle. +[18:56:36] I'm going to make it less about deadline and more about pressures. +[18:56:41] Okay, so I go first, right? +[18:56:47] Entonces al final... +[18:56:53] ¿Cómo pueden tener la confianza de que la decisión que están tomando es de parte de Dios o es la voluntad de Dios? +[18:56:55] Há quatro factores. +[18:56:59] que son testimonio o evidencia. +[18:57:01] de que Dios les hará una respuesta. +[18:57:03] Primeramente, como ya dijimos... +[18:57:06] Dios no va a dar dos respuestas contra ellas. +[18:57:11] si han tomado la tarea de orar y Dios ha hablado a las cosas +[18:57:14] y ha hablado lo mismo al esposo. +[18:57:16] están en simbolía y armonía. +[18:57:20] y están en acuerdo, esa es la primera evidencia. +[18:57:21] Segundo +[18:57:23] esa decisión de parte de Dios les dará paz. +[18:57:26] De Dios es la paz. +[18:57:27] y él no va a dar. +[18:57:30] una convicción en algo que les va a quitar la paz. +[18:57:34] sea de su relación o de su vida, de su casa. +[18:57:37] Entonces cuando los dos tienen paz +[18:57:38] en la decisión +[18:57:41] eso es la segunda evidencia. +[18:57:45] Me acuerdo que esto sucedió con Elias, nuestro primer hijo cuando era bebé. +[18:57:48] y estamos en el debate de +[18:57:51] como entrenarlo para dormir en la noche. +[18:57:54] y eran muy difícil que las emociones dejanos llorar. +[18:57:57] pero huramos +[18:58:00] y tomamos la decisión de seguir el consejo de... +[18:58:04] de la iglesia en Nueva York. +[18:58:06] que son +[18:58:10] han recibido mucha revelación acerca de cómo crear a los hijos en las cosas de Dios. +[18:58:12] and +[18:58:14] y empezamos a... +[18:58:17] a usar esa manera de entrenamiento +[18:58:19] y aunque fue difícil. +[18:58:25] escucharlo llorar o pasar por las pruebas de ese entrenamiento. +[18:58:27] los llevamos a los dos pasos. +[18:58:31] y eso fue como una evidencia de que estamos haciendo las cosas bien. +[18:58:35] La tercera evidencia es que... +[18:58:37] se tiene que alinear con la palabra de Dios. +[18:58:42] K wallpaper on +[18:58:44] lo que Dios nos ha dicho ya, como dijimos, como comentamos. +[18:58:45] Hay cosas. +[18:58:47] escritas en la Biblia que... +[18:58:52] ya no tenemos que preguntar sino que la voluntad de Dios está escrito +[18:58:54] Entonces cuando... +[18:58:55] Hey +[18:58:58] nos da su respuesta. +[18:59:01] él no va a contra decir su propia palabra. +[18:59:05] 1 +[18:59:05] cuarto +[18:59:10] debe estar en sometimiento a Lleras Coquedos nos ha dado. +[18:59:12] no debemos de tomar una decisión en contra +[18:59:16] the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the +[18:59:17] lo que hemos recibido de parte de nuestros líderes espirituales. +[18:59:21] Y solo comento... +[18:59:24] acerca de estos cuatro evidencias que +[18:59:28] mientras más leemos la Biblia y más conocemos la Biblia y... +[18:59:31] y diariamente estamos pasando tiempo en la palabra de Dios. +[18:59:33] Eso. +[18:59:35] no solo ayuda a... +[18:59:40] 여기 +[18:59:40] Conozcamos lo que dice Dios para... +[18:59:43] comparar lo que estamos recibiendo +[18:59:46] asegurarnos que están de acuerdo con la vivienda pero también +[18:59:50] Hace ese proceso que menciona en romanos 12, 1 y 2 +[18:59:55] que estamos transformados por la renovación de nuestra mente. +[18:59:57] Mientras más leemos la Biblia. +[19:00:02] más sometemos nuestras propias opiniones y recibimos la mente de Cristo. +[19:00:05] la mente de Dios que está dentro de su palabra. +[19:00:07] Entonces, en general... +[19:00:11] tener una vida de oración y una vida de lectura de la Viria. +[19:00:15] nos va a ayudar también en este proceso de tomar decisiones juntos. +[19:00:16] That's what I'm here for. +[19:00:20] y en este punto de buscar a Pueblos +[19:00:24] y es posible que tiene una... +[19:00:27] un papel importante como cabeza. +[19:00:31] y el cruz importante que tenemos es el... +[19:00:34] நிக்லனா. +[19:00:35] que nosotros +[19:00:38] olis insan +[19:00:39] No vamos a actuar. +[19:00:41] en un asunto determinado. +[19:00:43] a menos de que hayamos. +[19:00:44] que han encontrado ese juego. +[19:00:51] ¿Qué quiere decir esto? No quiere decir que el hombre va a ser que decide cuantinal de cuentas cuál es. +[19:00:55] el acuerdo correcto, sino que el hombre tiene esa postura. +[19:00:59] Está firmes. +[19:01:01] in the continent." +[19:01:02] la voluntad de Dios +[19:01:04] the start the most +[19:01:07] Over the fence.. +[19:01:09] a oración y a buscar de Dios +[19:01:12] hasta que tenemos un acuerdo genuino y sincero +[19:01:16] y esto no niega el papel de cabeza que tiene el esposo. +[19:01:21] sino que refuerza en una dimensión adicional que es de consejo y de confirmación. +[19:01:27] Una palabra a la más preferencia, son programios 11-14 y 15-22. +[19:01:30] y entonces algo +[19:01:34] aquí que nos muestra esta lección es que vamos la misma palabra +[19:01:36] se debe confirmar consigo mismo. +[19:01:40] si esto es el caso para la palabra que es +[19:01:42] la voluntad de Dios escrita para nosotros. +[19:01:48] 。 +[19:01:51] Nosotros necesitamos que este acuerdo y la dirección que sentimos del Señor sea confirmado. +[19:01:52] con la palabra +[19:02:00] Was that supposed to come in here? No. +[19:02:12] This paragraph is so weird. It's like three different topics in one paragraph. +[19:02:38] I think it's a slightly dangerous the way I mean I feel like we +[19:02:41] fix it fine like just by mentioning leadership but it's like +[19:02:44] It's slightly dangerous. +[19:02:45] to be like. +[19:02:48] If the two of you receive the same thing +[19:02:50] It's from the Lord. You know, it's like... +[19:02:54] The process is good, the exercise is good, but like... +[19:02:57] Just because you both think the same thing doesn't mean it's from God. +[19:03:07] Maybe it does if you're both mature and you're really receiving from the Lord. Depends how mature you are in hearing the voice of the Lord. +[19:03:17] El esposo sigue siendo el responsable delante de Dios. +[19:03:19] sobre la dimensión +[19:03:20] Okay. +[19:03:22] Toma com a minha família. +[19:03:24] Sin embargo... +[19:03:26] Te vamos a lograr. +[19:03:28] Thank you. +[19:03:30] confirmar estas decisiones. +[19:03:32] And what else, yeah. +[19:03:34] y sobre todo no debemos dejar. +[19:03:39] las constancias o las presiones de la vida, los deslanzo +[19:03:41] One more second, yeah. +[19:03:44] y pensemos que tenemos que tener una decisión para este tiempo. +[19:03:46] Is that OK? +[19:03:47] una expresión a tomar +[19:03:49] una decisión a la etapa. +[19:03:51] si no quedamos de confiar. +[19:03:55] que vamos a buscar la voluntad de Dios y que Dios no la abeja. +[19:03:58] creo que cuando necesitemos una respuesta +[19:04:00] el tiene para nosotros. +[19:04:02] tiene la mejor respuesta +[19:04:04] y que no la da hasta su tiempo. +[19:04:10] He +[19:04:12] para concluir. +[19:04:16] Para mí este es uno de los puntos más importantes. +[19:04:18] ¿Cuáles? +[19:04:24] el más grande beneficio de empezar a tomar acuerdos en esta manera buscando la voluntad de Dios. +[19:04:26] Pues es que... +[19:04:28] en esta manera +[19:04:32] nos ejercemos como instrumentos de la voluntad de Dios en la tierra. +[19:04:35] es parte de nuestra transformación. +[19:04:39] de buscar nuestra propia... +[19:04:40] Voluntary. +[19:04:43] a llegar a entender nuestra identidad. +[19:04:46] como individuo y como pareja. +[19:04:49] de establecer la voluntad y el reino de Dios en la tierra. +[19:04:52] cuando practicamos +[19:04:53] Este ejercicio +[19:04:57] de orar, de buscar el espíritu, de buscar la palabra. +[19:05:03] de tomar sesiones siempre buscando la voluntad de Dios es un ejercicio que no significa +[19:05:07] como instrumentos de la voluntad de Dios en la tierra. +[19:05:14] y ese es nuestro principal propósito para estar aquí. +[19:05:22] Estamos aquí con hijos de Dios para establecer la voluntad de Dios en la tierra, como leímos en el recículo de Juan. +[19:05:25] que Jesús siempre tenía esa mentalidad. +[19:05:27] Yo estoy aquí para hacer la voluntad de mi padre. +[19:05:29] Entonces... +[19:05:33] esta lección es como +[19:05:35] un ejercicio que conecta +[19:05:39] the the +[19:05:40] lo que llamamos microcosmos, el pequeño mundo de nuestra casa. +[19:05:42] con el cosmos. +[19:05:45] el macrocosmos de la Tierra Antena. +[19:05:52] pueden ignorar esas palabras grandes si no les parece pero el punto es que +[19:05:58] Hacemos el ejercicio de establecer la voluntad de Dios en nuestra casa. +[19:06:03] y nos prepara como embajadores del reino para establecer la voluntad de Dios. +[19:06:08] y me hace como a Jesús, me hace una disporación. +[19:06:10] Have a good day. +[19:06:15] así como este en los cielos +[19:06:16] You're completely gone. +[19:06:20] le será hecho por mi padre que está en los días. +[19:06:27] Sí, entonces mientras oramos y confiamos. +[19:06:29] y en la respuesta de Dios que viene. +[19:06:32] llegamos a hacer esos hijos en la tierra. +[19:06:34] Депускай! +[19:06:38] poner la tierra de cuero con los cielos. +[19:06:42] Sí, también lo veo ahora como así. +[19:06:45] esa forma de entrenamiento +[19:06:46] la familia +[19:06:48] en la relación de los posos. +[19:06:51] esa forma de ser entrenados +[19:06:54] y sobre todo de caminar junto +[19:06:56] Thank you very much. +[19:07:00] Hey, hey, hey. How am I going to listen to the call of God? +[19:07:04] sobre nosotros primero, sobre nuestras vidas, como pareja, como familia. +[19:07:08] pero también nuestro oído es siempre entrenando como escuchar. +[19:07:10] La voluntad de Dios +[19:07:12] en nuestro entorno, en nuestro... +[19:07:14] la cultura en nuestro país. +[19:07:18] Qué bendición. +[19:07:22] the most amazing thing about the +[19:07:23] este entrenamiento del señor y saber que... +[19:07:25] Sus promesas, hasta la próxima. +[19:07:28] y cuando busquemos, encontraremos. +[19:07:30] Thank you. +[19:07:36] Si su voluntad es que caminemos juntos, él nos va a ayudar a encontrar los acuerdos correctos de la forma correcta. +[19:07:37] Come here. +[19:07:42] Entonces como siempre tienen 5 minutos para hacer preguntas de aclaración. +[19:07:45] si algo necesiten a explorar de la enseñanza. +[19:07:48] 1 2 3 4 5 6 7 8 9 8 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 +[19:07:49] van a recibir su tarea para esta semana. +[19:07:54] y van a identificar unas maneras que han tomado colos en el pasado. +[19:07:55] 1 +[19:07:59] también practicar lo que es ahorrar juntos para tomar acuerdos. +[19:08:13] wi al final del verano. Hola, mi hermanito. +[19:08:14] Entonces los amamos y los bendecimos y nos vemos en la próxima seda. Hasta la próxima. +[19:08:18] Oh, we have physical intimacy. +[19:08:20] Did we skip that one when they were here? +[19:12:41] Okay, what is done for today? Carvis. +[19:12:45] Jarvis. diff --git a/workspace/transcription/2026-03-03.log b/workspace/transcription/2026-03-03.log new file mode 100644 index 0000000..3b7127a --- /dev/null +++ b/workspace/transcription/2026-03-03.log @@ -0,0 +1,543 @@ +[11:57:57] Tudo bem? +[11:58:03] Today is the second day of the AI focus week on my work +[11:58:07] So, same than yesterday +[11:58:09] We're just gonna be... +[11:58:12] listening to the meetings and paying attention to them. +[11:58:15] After +[11:58:21] All that's done, I'm gonna ask you to summarize. I might ask you in the middle to summarize like certain meetings and things like that. +[11:58:25] let's do it Jarvis +[12:00:31] commit commit commit commit Thank you. +[12:00:32] all the life sessions were buzzing, folks were leaning in hard, learning. +[12:00:33] बिल्डिंग +[12:00:35] Mir commit Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir Mir +[12:00:36] Really really cool. Thank you so much +[12:00:38] We're gonna continue this for the next. +[12:00:39] 4 days. +[12:00:42] and we crank up our learnings and +[12:00:45] intensity and building and +[12:00:46] It's just gonna be... +[12:00:49] really awesome and thank you again for all your participation. +[12:00:54] Day 2, we are going to kick it off with a very very special guest. +[12:00:58] commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:00:59] You know, he, he. +[12:01:01] peace +[12:01:03] The legend in in in in meta, you know +[12:01:04] See. +[12:01:05] is a core head up. +[12:01:10] Engineering for all of Meta, Heads Thief, the RL Foundations team. +[12:01:12] Boris +[12:01:13] It's got a long, long. +[12:01:17] Factor commit. +[12:01:17] you know, an amazing industry, I think spanning. +[12:01:21] He fights the kids, he brings us. +[12:01:26] Wealth of experience, wealth of knowledge and wealth of inspiration. +[12:01:29] Everyone please welcome. +[12:01:30] The one and only. +[12:01:34] Mahir commit +[12:01:37] Sabah. +[12:01:38] Good morning. +[12:01:43] Your mutating. +[12:01:52] Right now. +[12:01:54] Can you get me? +[12:01:55] Yes. +[12:01:58] Yeah, I would say you can think about how old I am. +[12:02:04] Sabah is 21. +[12:02:08] Forever turning on. +[12:02:09] Yes, exactly. So thank you so much. +[12:02:11] for joining us. Really, really. +[12:02:13] Yeah, for sure. +[12:02:15] Yeah, I mean I think +[12:02:17] There's none other than you. +[12:02:21] to inspire us on how to kind of learn, build and adapt in the ZA world. +[12:02:37] quite a while and... +[12:02:41] And then of course you came in as one of the... +[12:02:45] You came to Facebook, you joined Facebook actually, so you came to kind of... +[12:02:51] had pretty important efforts within Facebook and then you went around your journey doing various other things in Meda. Okay. +[12:02:55] Now, Saba, tell us about your journey. Okay, all these tech waves. +[12:03:00] all the decades and what's special about this particular one? +[12:03:07] Yeah, so yeah, I never worked on Gobo work. +[12:03:09] I'm not that old. +[12:03:10] So... +[12:03:19] No, I started IBM when IBM was trying to build the Unix version of back then called System 5. +[12:03:22] from AT&T and BST 4.3. +[12:03:25] They were thanks for the dinner and my job was actually +[12:03:33] taking every piece of code that is slow and making it faster by wrapping it in assembly. +[12:03:37] to something that I may call power DC, which later on became +[12:03:41] you know, the incarnation of... +[12:03:47] moving away from SIS versus risk and this was the risk kind of an instruction set. +[12:03:58] You're probably gonna have to cut the Wikipedia half of the stuff that I say, but because they don't matter anymore. It was an interesting time because... +[12:04:00] back then. +[12:04:04] IBM was the only name in the endowment. +[12:04:06] I believe it or not. +[12:04:12] You know all the work that they were doing they would be paid to have the rest of the industry +[12:04:13] Um +[12:04:24] So what was interesting to see like you know a giant control of the industry and then obviously that went down then I went to the continuous operating system work to +[12:04:25] I work on Windows. +[12:04:30] that also became a giant control of the industry. +[12:04:39] And then later on with open source, Linux, you know, you next came back in a major way, especially with the mobile. +[12:04:42] I think Microsoft missed that wave in a major way. +[12:04:46] So I kind of went back to my roots with the UNIX. +[12:04:50] And more get on mobile stuff and this is what the that need a Facebook +[12:04:54] in here. +[12:04:56] on various projects. +[12:04:58] Um, minus three doesn't matter. What, what? +[12:05:00] I think it's the most interesting. +[12:05:03] Bar with me, Johnny, for Facebook where... +[12:05:07] commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:05:12] you know set the tone to what a stark comparison to my previous period to joining here. +[12:05:17] I showed up and they're like, hey, we hear that video is growing. +[12:05:20] on the internet and people are using video. +[12:05:22] We want the best video for Facebook. +[12:05:26] Here is the team of five people. +[12:05:30] We're also going to give you 10 headcount and go beat YouTube. +[12:05:31] And I'm like... +[12:05:39] No wonder they didn't offer this job to anybody else. No one else would make it. So I didn't even know how to spell video. +[12:05:42] But that's where we started. +[12:05:45] I learned a lot quickly. +[12:05:51] that risk taking is something is very, very encouraged in this company. And I love that match my. +[12:05:53] personality in my preference so +[12:06:06] Since then, I've been taking lots and lots of risks, taking on different projects, and some of them succeeded, some of them didn't. It didn't matter. The most important part that we've got. +[12:06:17] All of this, you know, gave me a wealth of information about like how we do stuff, how we set up teams, how you write code, how you should care for quality. +[12:06:23] until you know the AI wave showed up. +[12:06:25] and they I wave, you know, starting. +[12:06:30] 2017, you know, the famous papers and people minted it and so on. +[12:06:33] But it really did not hit us here at Meta. +[12:06:45] and Bill, I would say six months ago, there were people like dabbling with it, working with it, doctors, and some people just want to learn a new technology and they were interested in it. +[12:06:47] But all of a sudden... +[12:06:50] With advances with AI. +[12:06:53] You see a huge huge separation between. +[12:06:56] people who are producing that power to the maximum. +[12:07:05] And then people who are like on it, just use it as a tool. And we want to do more and more to maximize the usage of AI. +[12:07:06] and uh +[12:07:10] commit commit +[12:07:10] You hear what I'm doing, just like the rest of my journey is... +[12:07:12] It doesn't matter what got me here. +[12:07:14] Like everything I know. +[12:07:16] when it comes like for AI. +[12:07:18] I'm an ICP. +[12:07:32] I was in IC3, I'd say three months ago or four months ago in that domain. And I kept on working and building and so on. I think I'm getting closer to IC7 and I'm hoping to be like an IC8 soon. +[12:07:33] in that domain. +[12:07:44] But that's the thing, it's such a great way to level the field, independent of the experience, the tenure, the function and all that stuff. +[12:07:50] commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:07:56] and your creativity and the power of AI you can accomplish so so much more than you've ever accomplished in your life. +[12:07:58] and many of your, you know... +[12:08:01] people have been before you, what they have accomplished. +[12:08:04] just because it's an unbelievable unbelievable +[12:08:06] a powerful way to do it. +[12:08:10] in your basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic basic +[12:08:10] Awesome. What a journey, Sabah. +[12:08:15] You know, I think you mentioned, you know, ambition and, you know, creativity. +[12:08:16] But this +[12:08:25] There's one thing about you which pops out really for me is also curiosity, which is an important piece of the puzzle as we're kind of learning and adapting, you know. +[12:08:30] And I think you've been an epitome of that. Let me digress and give a super big story for our friends here. This was... +[12:08:38] I know Savai, we talked about it before. This was two decades ago. I was a very early and career engineer working on Windows operating system. +[12:08:41] And all of a sudden I get a mail from this guy who said +[12:08:48] Hey, your source code, take this diff and apply it. Your source code is gonna improve by 4% the CPU compute. +[12:08:51] I was like every possessive junior engineer, who's this guy? +[12:08:57] you know updating my source code and telling me that I should improve performance and I look it up it's at Mahir Sabha. This is... +[12:08:59] Many, many years ago. +[12:09:05] You know, and like, of course, some of you are in the Windows internals performance team at that point and then +[12:09:08] You're working on something really deep, you know. +[12:09:08] And then +[12:09:13] I remember my second interaction was about maybe 10 years later, we were working on the secret phone. +[12:09:14] project and again. +[12:09:20] My boss at the point tells me Sabah has got some voicemail problems go debug and this is 9pm in the night. +[12:09:25] And you were with me trying to debug this to the core and stuff like that. And of course. +[12:09:32] I had an opportunity to work inside. But what keeps you going? Like not just the breath, I think you go deep. +[12:09:34] Like you're curious you want to learn more and +[12:09:41] Like, like, is that a formula you have? Like, how can we, like, is that a repeatable formula or is just the inbuilt, like, you got it? +[12:09:51] No, I mean I love to get my hands dirty. Just let me technologist. I'm an engineer. I was born that way So the more I can get into the details of fixing stuff +[12:09:55] You know, the more I find that stuff fascinating. +[12:10:07] I've never looked at the challenge and I said, oh my god, I want to move away from that. You know, in fact, I get excited about doing it. I think a couple of exciting or interesting stories here. +[12:10:11] When I first got into computing I was 14 years old. +[12:10:16] My dad bought me this computer box in flair, I don't speak English, I don't speak English so later. +[12:10:17] And uh... +[12:10:21] You put these letters together. +[12:10:22] The computer will do something. +[12:10:30] So I was basically programming in basic without knowing English. And I was just like memorizing that these tokens means things. +[12:10:32] And that's how I was coding. +[12:10:39] you know at 14 years old with zero support from everybody and so on and so and that's like I +[12:10:48] I basically thought myself like, you don't even need the language, you don't need the instructions, like there's this weird machine that does what you do. +[12:10:50] what you've done it to do. +[12:10:53] and then he built that relationship. +[12:10:59] The other one is in my experience was in my first +[12:11:01] Computer class. +[12:11:06] God was in a language called Pascal, which many of you haven't heard of it, but they used to teach Pascal because it was like... +[12:11:12] very well, uh, syntax the language so you don't go out and learn to see and go wrong. +[12:11:15] So they will be just passed out first. +[12:11:17] And the professor goes... +[12:11:21] I have this program that does this. +[12:11:24] and the +[12:11:27] I'm gonna, you know, I'm gonna time your programs to see how fast your solutions are. +[12:11:30] and mine, you know, came slow in seconds. +[12:11:33] And here are his and his two milliseconds. +[12:11:34] And here goes. +[12:11:36] Your own will never beat me. +[12:11:37] Alex. +[12:11:43] God damn it, like why you? What does he know that I don't know? I mean besides being my professor and knowing everything. +[12:11:48] So I went and spent hours upon hours, maybe days, just trying to beat them. +[12:12:11] us and once I learned how to get down to the operating system, yeah, I didn't know or I came very close to him. So again, like, you know, +[12:12:28] And that's what I keep doing. +[12:12:32] commit +[12:12:32] Awesome, it makes sense. Very inspiring. +[12:12:37] If I may ask you, you know, one of the reasons why we're having this gifty-wweak Noisberry, one of the feedback made. +[12:12:40] Things are changing very fast. There's so many things. +[12:12:49] And one of the feedback we heard was, oh boy, we just don't have time to do this. You know, we just like always, we have half goals, we have team commitments and things like that. +[12:12:54] and Thomas Grace-Euthanel to give us this week and says, okay, you know what, this is a learning week and go do stuff. +[12:12:58] But you know, I've also seen you kind of learn things on the fly. +[12:13:01] Is it a habit like you build in? Like do you allocate like? +[12:13:04] time in a week or a time in a day where you're going to +[12:13:16] Go listen to new things, listen to new tech, especially being a co-head of engineering and media, you have to keep yourself up here with so many things out in the industry as well. Just curious, just from a learning perspective, for us to pick up any good habits here. +[12:13:24] commit commit time. +[12:13:32] I mean, I get excited by stuff. So once I get excited, I just go deep as much as I can, and this time allows. But yeah, time blocks is usually the best way to do this. And so. +[12:13:35] I had a good time, but I was getting good. +[12:13:38] a way, you know, for me to go there. +[12:13:43] Like I don't say like, oh my god, I'm gonna grab a hour to order something new. +[12:13:57] No, usually the way it works the opposite of like, oh, here's something new. Oh my god, this is awesome. I need to learn more about it. I need to figure out what's happening. OK, now let me go allocate the time to get it done. So each time I find something I'm interested in. +[12:14:04] I just go on my calendar and put the time for that. Now the hard part here for me is to... +[12:14:13] Hold off my excitement for the next few hours or the tomorrow so I can spend it in there. That was so good because I like it. +[12:14:27] I want to learn about this, but there's always a million things. But I remember that I have commitment to those million things. So I go spend the time on them, and give them their due attention, and importance, and time and effort. +[12:14:39] And then actually I'll be like, okay, cool. Now I have the time to go do this. And it's like opening a new book or just having an experience or going on a journey or an adventure. +[12:14:49] like what a great feeling. And so that's like, you know, I kind of built some silly suspense to get into that time to go and work on new things. +[12:14:59] Makes sense. I think I'm just coming back to the secret word. I think you just extremely curious Which is a which is an important skill all of us to pick up on +[12:15:05] I just saw this amazing hack project you did a few weeks ago. It is mind blowing. +[12:15:12] Can you walk us through? Why do you stop? How do you even stop there? You know, just building on your previous... +[12:15:13] you know, response. +[12:15:16] So you want to do something. +[12:15:29] then what went through it, okay? I'm gonna apply this, I need to build an agent. Maybe I should just like, you know, walk me through the process with went through a head, then you build the hack. And for context for the friends here, if you can also talk about the hack enough, not sure everybody is, you know. +[12:15:35] can commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:15:38] gone pretty viral all of mid-air but just the hack itself and the journey. Like how did you start that process? +[12:15:41] Yeah, it was like really... +[12:15:43] Open weekend. +[12:15:45] The open slot or the... +[12:15:46] and then we'll have to +[12:15:54] your web spotlight. +[12:15:56] And then all of a sudden everybody starts finding their mac minis and... +[12:15:59] Let's start installing it and using it and making it +[12:16:02] The videos about it and so on and so forth. +[12:16:10] And of course, like this is cool. So I went to this folder that I built it and I gave it all sorts of things, you know, but then like +[12:16:12] God, this is really my assistant. +[12:16:14] Why am I typing to it? +[12:16:16] I wanna really talk to you. +[12:16:23] And I'm like, but the phone is too small. And my sister, that wanted to be next to me. +[12:16:25] In the house of them like +[12:16:29] We have these portals that we used to make. +[12:16:33] And they're lying around the house and I'm like... +[12:16:46] Because of course every time we do something I buy like two or three of them just to give a different to them. And so I'm like, I can't find one and so on. So I call some people and they're like, okay, here was the number one for you. So I got one. +[12:16:47] And I'm like +[12:16:52] This would be the perfect device because now this assistant based on open block. +[12:16:52] is +[12:16:57] You know, it has a lot more power. It has a speaker +[12:17:02] It has a microphone, it has a camera, it has a nice screen, it has a touchscreen. +[12:17:04] all sorts of like +[12:17:05] capabilities that +[12:17:10] But I think I can use to make my interactions with much better. +[12:17:11] And, uh... +[12:17:13] And +[12:17:22] I went and I connected it and I went through all the, you know, connected the serial connections of what have you to my depth server and what have you. +[12:17:26] And I said, hey, take open claw. +[12:17:29] and force it onto the Rostov device. +[12:17:35] So I asked FLA to lamp open bar and port it on the port of device. +[12:17:38] And I'm thinking like this guy is taking 2-3 days. +[12:17:42] Happy now our day, there's life, here it is, it's off and running, a lens. +[12:17:43] Bye. +[12:17:48] I would have asked for a probole if it was the best. +[12:17:50] You know, look how good I am. +[12:17:51] I got it working. +[12:17:55] Alright, so I got to work and I'm like, okay, this is good. +[12:18:02] And then I went and I, but then I was still typing and I like, I don't like this. And I'm like, all right, I wanted to +[12:18:13] I want a video talking back to me. So I went to MetaAI and I said, create me this generic image of an agent. +[12:18:14] It's surely not great, it's one. +[12:18:21] That's it animated, animated talking, animated listening, animated speaking, all the stuff, animated those downloadable videos. +[12:18:23] and I went to vlog and I'm like... +[12:18:28] Here are my states, here are the videos that match this, make it look like this. +[12:18:29] Oh +[12:18:31] Happy hour later, it's up and running. +[12:18:36] I'm like, okay, I wanted to listen to me not to... +[12:18:39] that do need typing. +[12:18:51] And it says, great. And each time they were like, oh, you need those APIs. Go spend five bucks in here. Or you need to spend the API. Go spend this up here. So I put it up. And now we can talk and listen to me using the tabs. +[12:18:52] I'm like okay cool. +[12:18:58] I could tap on the screen and ask it to do something and it does it. And then I said, how's the weather around me? +[12:19:01] had no idea what's going on. +[12:19:08] Okay, this sucks. Then I, well, your depth server can have surge, and your depth server is locked down, and the real server is this bad. +[12:19:12] Alright, I'm gonna grab my Mac mini and I'm gonna attack my chipsorter. +[12:19:16] So I like to rewrite all the code and put it all on the Mac and the editor server. +[12:19:20] and close it. +[12:19:23] You know what? If you're going back, Bob, why you're paying money for Obis... uh... ...the... uh... thing. +[12:19:28] Why did I go bring llama and put it on the back of the head? So I brought llama and put it in there because I was running llama. +[12:19:36] I like, okay cool. And it says, oh, you want to talk to us, why don't we go grab this open source to put it in there so the process speech. +[12:19:40] You're enough. It was a fun thing that you did in the park. +[12:19:45] And now I can just, you know, not a, oh, and I have to go buy some... +[12:19:50] $16 and like okay sweet I'll spend the money +[12:19:55] And now I can search at the local restaurant. I can do this. I can do that. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. I'm going to go to the local restaurant. +[12:19:57] And as I'm doing all the things, uh... +[12:19:59] Somebody's at the need. +[12:20:06] What's with this topic as I'm showing them? Well, I'm like, yeah, you're right. I'm like, it has a camera. I'm like, OK, each time I want to look at it. +[12:20:08] I want to be camera or hold a gun. +[12:20:09] I think that is his gaze. +[12:20:13] And then my assistant will say, hey, how may I help you? +[12:20:16] Like basically I'm starting this project in British. +[12:20:24] You know, by the way, can you say languages, what, you can talk to the train, and what would you like to talk about, or the German, or whatever? +[12:20:27] OK. +[12:20:31] I'm just going to leave it at that. This is the method used. And now each time I'm going to play football, I'm going to help you. I'm back. +[12:20:38] Some of the about 90, some of the about 40, it's not working working. And I'm like, but what's that before the, it's like, you know... +[12:20:43] That's where the market do this, where the market do that, just where the creativity comes in. +[12:20:49] Eight days later, I went to the fact-flash insights and said, yeah, we've modified over a million lines of code. +[12:20:51] Oh my god. +[12:20:54] million lots of, I mean it was 700 ads and +[12:20:58] you know subtract. +[12:20:59] very much. +[12:21:06] Never in my life I wouldn't even think about me, you know, like there were only few people in this company who +[12:21:09] have like the many lives of folks themselves and they contact it. +[12:21:10] So +[12:21:14] For me to have this ability to do this is fantastic. +[12:21:16] And then after I got the insights. +[12:21:18] It came to me and said +[12:21:20] By the way, you're pretty good at using flaws. +[12:21:21] Bye. +[12:21:28] Let me tell you a few more things that you can even make yourself better. Oh my god, I'm having a performance review with this AI, I think. +[12:21:37] And it went and helped me to tell me exactly what I need to do. And so I learned as I went. So basically, pick any project. +[12:21:41] Your curiosity will carry you for the first couple days. +[12:21:49] Cre commiti cualquier yourseful dwe commit commit +[12:21:51] and you stopping at the checkpoints and asking how I'm doing so far will teach you and value into whatever you want to. +[12:21:59] And it's like a very natural thing because that's how we are technologists do like no one told you the stuff you did it and that's why you're here +[12:22:00] Yeah +[12:22:15] What a story, very inspiring. Thank you. Thank you, Sabha. Sabha, we talked about how you learn, and then you share as a nice little story on how you build the school stuff, and you continue to build your forever 21. I expect you to keep building things. Now, +[12:22:21] I want to change this a little bit and talk about adapting to this work. +[12:22:22] What a good bed I just had. +[12:22:26] Things are changing very fast and we don't even know what's gonna be that now. +[12:22:28] Like how are you spring? +[12:22:31] What tips you have for us as... +[12:22:34] engineers, leaders and excellent partners here. +[12:22:46] on how we should all look at this journey, on how we should go about adapting to this phase. You know, there's various things that can help, you know, people paint a picture of doom, there's a picture of optimism, there's a picture of... +[12:22:55] you know, it's gonna be great. This is a great, I'm sure you're reading all of this and hearing all of this too. But what's your perspective? And if you have a few nuggets of wisdom on. +[12:22:58] how we should all adapt to this what would that be? +[12:23:00] Yeah, I mean, my... +[12:23:04] My thing here is you have to change your mindset. +[12:23:08] Every time you have any challenge. +[12:23:10] Any questions? +[12:23:14] and you find yourself typing looking for the search, you're not typing into some. +[12:23:15] 또 한 번 더. +[12:23:21] commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:23:23] and you'll probably get a waste a lot more time even though you know where the answer might be. +[12:23:31] Just go push yourself to go use AI to figure out how to get the master to school to get out of it. +[12:23:33] Second one. +[12:23:36] and AI wants to please. +[12:23:38] Poderá pregui-me. +[12:23:38] is +[12:23:39] You can forward it. +[12:23:44] press it into any answer, you can do this in case you want to do any answer you want to do in the FHP. +[12:23:45] Yeah, I get it. +[12:23:51] Well, okay, but you're not actually harnessing the power of it. You're trying to do much something. +[12:23:53] and install +[12:23:55] The bugger's on, right? +[12:24:04] Actually, Mark says this all the time. And I don't want to quit the barbecue world. I think it's a bit... But we can... +[12:24:05] This is alright. +[12:24:07] commit +[12:24:08] Optimists are successful. +[12:24:12] So my advice to all of you is to be optimistic. +[12:24:14] So you can be successful. +[12:24:18] It's not about these things, this is how right I am. So I leave it on you and you give me the wrong thing. +[12:24:25] It's more like focus on where I use this to actually help me do the better thing and help me improve the thing. +[12:24:28] So basically, you're on the side of light. +[12:24:35] I don't want to tell you how bad AI is. It's basically the deal inside of life. This is how AI is actually happening. +[12:24:39] And the more you do that and learn how to use it and harness its power. +[12:24:43] commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit commit +[12:24:47] You can get a lot more productive, you can harness your own creativity. You're gonna get to the point where I'll be dying. +[12:24:56] Yes, boss, go ahead, do your thing. Stop asking me questions. Yeah, yeah, yeah, so on. And then at the end, I will go and say to it, you know. +[12:24:57] Okay, you've wrote enough code. +[12:25:00] go do a football review +[12:25:05] Go find more bugs and go put some of yourself and come back to me. So basically go down the path. +[12:25:09] Tigra, continue Tigra. Not all the problems along the way. +[12:25:13] put them together as a thick backlock and then go fix them. +[12:25:18] The good news here is you can do that on a daily basis. This is not like something you do. +[12:25:20] on the weekly basis. +[12:25:25] Every time you face any challenges, the problem will always work for you. So, you have to use AI. +[12:25:44] learning and how to use it with a back multiple pulls. +[12:25:46] And that's the video to push back that I want to do. +[12:25:50] out of this. There is no bad investment in learning about AI. +[12:25:54] Awesome. I think I am. +[12:25:58] That sentence you said, you know, just being in the head, you know. +[12:25:59] Just missed a ride +[12:26:09] Optimists are successful. So in a sense, I think you're saying sir, you just leave the whole thing with optimism, you know, if we all have to adapt and win and be successful. +[12:26:17] in this way we just need to lead with optimism in every day what we do right if you're pessimist about something I think it's gonna get you so +[12:26:18] Optimism. +[12:26:41] I'm using a lot of code a lot. Yeah. There's a couple other stuff coming. I don't want to re-enounce any product in here. OK. OK. +[12:26:42] that we're using. +[12:26:45] Okay, that's awesome. I am not putting some of them. +[12:26:53] But I'm like, why am I dog pudding myself? So I got an agent that went out and did the dog pudding for me. And it's crazy. +[12:27:06] I'm like the best dog fooder for that product because I'm not doing anything I just figured out how to use AI without putting AI product that we're gonna do things and this is where the Beautiful virtual virtual cycles gonna come in +[12:27:07] Yeah, amazing. +[12:27:09] Soba! +[12:27:12] Thank you so much for all your time and inspiration. Really appreciate it. +[12:27:13] And, uh... +[12:27:17] Yeah, I said it for this amazing future. +[12:27:19] And one day I'll get up to you and say... +[12:27:25] Your code needs to be faster by 4% and you can do it. Yes, I bet I'm waiting. +[12:27:32] I think that is still stuck in my head. You know, I was like, who's this guy? He said he wanted to beat himself. All right. Thanks, everyone. +[12:27:33] until next week. +[12:27:34] Thank you. Cheers. +[12:27:38] Good folks. Thank you so much again. +[12:27:42] Hopefully this session was interesting and inspiring. +[12:27:48] sub-ass-credit inspiration so continue to follow him. He does quite amazing things on a lot of projects. +[12:27:51] And yeah, so coming back to regular programming here. +[12:27:53] They for they to here +[12:27:59] So we're going to have the usual programs. You know, the next hour we're going to have an FBI session. +[12:28:04] and after that at 1030 we're going to break off to all the functional tracks. +[12:28:08] Yesterday we saw amazing engagement in the functional track, so keep keep at it. +[12:28:09] And again... +[12:28:17] Today is the day you want to kind of have a good idea of the hack project. It's an important stuff We want pretty much everyone to participate in the hack +[12:28:20] You want to see 7000 plus entries in that. +[12:28:21] and it's not optional. +[12:28:24] So just a reminder on that, so you should start thinking through that. +[12:28:28] अद नुन तुडे लेटिस पाई नुन अद नेर नुन +[12:28:51] And one thing which came up in conversations is that there's a lot of amazing sessions, but I actually know some of the stuff, what should I do? +[12:28:54] Like some of you might feel that, you know, you might be advanced. +[12:29:01] Number one, I think reputation is the best form of learning. Sometimes it's actually go through some of the sessions again, because it actually can strengthen your fundamentals. +[12:29:01] But +[12:29:14] you know, at least like some of the sessions we have crafted in such way that the levels of these things graduate over the days. Yesterday was like one on one style, today will be two on three on one, so that's how we have crafted it. It will get pretty. +[12:29:15] you know, pretty... +[12:29:20] you know, you know, higher level as you go, especially in engineering, it's been crappier than that way. +[12:29:30] Do that. But in the meantime, if you have, if you want to go deeper on your own, work on some very deep projects, you know, there's some big refactoring you have to do, you know, code base to make it air ready. +[12:29:35] This is the time, this is the week. This is the week of learning, this is the week of building. You're welcome to do that. +[12:29:40] Connect the intensity of your learning and building as much as you want to solve in. +[12:29:49] your hands and our hands. So just want to pull it out back. With that, I think we're ready to go to the next session. I think it's going to be Chris and team picking it up from +[12:29:50] from me. +[12:29:53] Chris, do you want to... +[12:29:54] Get on stage. +[12:29:56] Happy to do so. +[12:30:00] F commit. +[12:30:00] Let me get set up here. +[12:30:08] Alright. +[12:30:10] Buckle up buttercups. Hey Facebook +[12:30:17] Let's talk about how easy it is to take an idea and bring it to life with Vibe Designing and Figma Make. +[12:30:21] So I'm Chris Maverick and I'm here with Jason Wing. +[12:30:25] and special thanks to Ash Clair Clyburn and Karen Lannister contributors. +[12:30:30] So first we're going to give you the quick summary of what vibe design is. +[12:30:34] And then we're going to show you how to use FigmaMake as a by design tool. +[12:30:43] And finally close out with a quick example. This is just a crash course. There's a lot more that you can do here, but we want to inspire you and give you the basic tooling and understand what this is all about. +[12:30:47] So let's start with an overview of what Vibe Designing is. +[12:30:55] designing is designed with AI using props. That's the TLDR most basic summary of what you're going to do. +[12:30:59] And typically it's to create quick interactive prototypes. +[12:31:02] So instead of having an idea trapped in your head. +[12:31:07] or attempting to describe it to others, you can liberate it by manifesting your concept visually. +[12:31:12] to better understand it yourself, iterate to a better solution, or represent it to others to get by in. +[12:31:18] or put another way, the phrase in everybody's lips is to help you move into a world of demos. +[12:31:22] Nam commit. +[12:31:25] And then the question here is who can vibe sound? +[12:31:28] It's literally anyone, truly. It really is anybody. +[12:31:29] including you. +[12:31:37] Because like other AI experiences, like talking to Metamate or ChatGPT, it's all about prompting to get information or a result. +[12:31:43] So Vive Designing is ultimately prompting with an experiential lens in mind to manifest your idea. +[12:31:44] or a design. +[12:31:47] So take this quick prompt during an example. +[12:31:49] I saw a feature on another app. +[12:31:52] where you can see posts from friends on a map view. +[12:31:55] And I want to see what that looks like for Facebook. +[12:31:58] So I'm attaching a screenshot of the experience that I saw. +[12:32:03] I want you to use Facebook's visual design language. That's my general prompt. +[12:32:04] And boom! +[12:32:07] We've got various components already in place. +[12:32:17] But the map is missing, and so it goes in the Wild West of AI. We're starting to get familiar with this. Things don't always go as we expect. So we iterate and we keep pushing. +[12:32:20] So I asked for a map of Sydney, Australia, to fill the void. +[12:32:21] Got one. +[12:32:24] But it's 3D, so I asked for it to be 2D. +[12:32:26] Getting closer, almost there. +[12:32:38] but it doesn't look like the typical mobile app. Real quick, we're not seeing your slides in advance. It looks like maybe a different app might be selected when it throws your screen share. We're still on that title card. +[12:32:39] Oh. +[12:32:42] Can you see it moving now? +[12:32:48] Yeah, I can see your screen, everything live there like that. I'm not sure what was happening with your slides, but. +[12:32:49] Let's do it. +[12:32:50] Okay. +[12:32:56] don't know why it's increasing there. +[12:32:58] Can you see it now? That's. +[12:33:00] That's great. Carry on. Thank you. +[12:33:01] Alright, cool. +[12:33:06] Let me get back to where it was. So nobody saw these screens. +[12:33:09] No, it was just that title card. +[12:33:10] Oh, weird. OK. +[12:33:12] Let's forget something on my side. +[12:33:13] So let me just back up just for a second. +[12:33:17] So starting with this five design journey, this is helpful to see. +[12:33:21] So like I was saying, you have an app that you recently released or that you saw in the wild. +[12:33:29] and then you wanted to be able to make a version of this via vibe designing. So this is a really simple prompt for you to describe what you saw. +[12:33:35] you'd like the five design tool to be able to build this design using Facebook visual design and you attach a screenshot. +[12:33:37] So you get this +[12:33:39] particular set of components which looks pretty good. +[12:33:42] But it's missing the actual map, right? +[12:33:44] So the best thing to do is to iterate. +[12:33:56] So you have the math you asked to add, in this case, Sydney, Australia. But lo and behold, it is 3D and not 2D. So then you ask to make the map flat and two-dimensional, and then it erases that way. +[12:33:59] but it's still not looking like the map that you're probably gonna want. +[12:33:59] So +[12:34:07] In this case, I'm wanting it to follow a map that I'm familiar with and to follow a style reference so I update an actual screenshot. +[12:34:08] And then boom. +[12:34:12] we've got the thing that we were actually looking for. It's very close to the original idea. +[12:34:17] And then if I wanted to say make it interactive, like if you pressed on the bottom tab button there, the input field. +[12:34:23] Then you could just say, hey, tap on that. And when a user taps it, create a bottom sheet that pops up. +[12:34:31] for being able to search for friends and boom, you have all of that in truly just a matter of minutes. You go from your idea to getting something ready to go. +[12:34:41] Now, all of this though, starts with a really clear understanding of prompting if you really want to get a better understanding of vibe designing, aka vibe coding in this case. +[12:34:44] and effective prompting is clear specific. +[12:34:50] iterative and collaborative. And by collaborative, I mean, you can actually collaborate with the AI as you're designing. +[12:34:56] you could have an idea, show a screen and say what are some different ways that I could actually design or visualize this. +[12:34:58] work with the AI and decide how you want to proceed. +[12:35:05] but it really builds on a lot of the prompting information that y'all heard yesterday and you'll continue to hear throughout the week. +[12:35:12] So bear with me on this one. I know you're seeing a bunch of different prompt frameworks. This is just another best class example. +[12:35:18] And this is one that works particularly well in five design tools like Sigma Make, which we're going to go over soon. +[12:35:20] So remember TC, EBC. +[12:35:22] So you have the task +[12:35:23] of what you're designing. +[12:35:27] and then like the actual thing that you're trying to accomplish here. +[12:35:31] Then you have context like user problem, audience, use case. +[12:35:36] elements or features like create a bottom sheet that's going to do XYZ. +[12:35:40] or UI components like a button or an input field. +[12:35:43] you have behavior like interactions. +[12:35:47] Like if the user is going to do X, I would like Y to be the result. +[12:35:49] and then you have constraints. +[12:35:52] So things like the format like an iOS mobile app. +[12:35:57] You have the style, like look like Facebook, look like Spotify, look like Pinterest, look... +[12:36:09] medieval, whatever it might be for you. Boundaries, just anything that you're wanting the app or the experience to do versus not do, and then references, like you can include a screenshot to say I want to actually look something like this. +[12:36:18] And then you could do something complex where you could do a really big prompt, describe your big idea, put everything in there. +[12:36:22] But AIs tend to struggle with a lot of complexity too quickly. +[12:36:25] So it's typically best to prompt iteratively. +[12:36:28] Starting with the base template or the foundation of your idea. +[12:36:30] And then you add a feature. +[12:36:41] then you can make it interactive, etc. And so I'm going to give you some idealized examples just to give you the idea. And again, you don't need to remember all of this. This is to give you an idea of how this is working. +[12:36:46] So starting with this prompt journey that I was talking about, you can create a base template. +[12:36:47] And then +[12:36:54] An okay version of a prompt would be create a home screen for task following TCEBC. +[12:36:57] In this case, task is whatever you want the home screen for your experience to be. +[12:37:02] But that's going to be vague, and so it's going to leave a lot to interpretation for the AI. +[12:37:06] So what's better would be to create a home screen for your task. +[12:37:12] based on context, like the user problem, or what you're hoping that the user is going to want to accomplish. +[12:37:17] A best version of that problem would be something like create a home screen for the task. +[12:37:25] based on the context, include these elements that you may have in mind, like I want there to be an input field, or I want there to be pins on a map. +[12:37:29] and then use these constraints like visual language. +[12:37:30] So +[12:37:35] An example prompt there being create a mobile iOS social media home screen. +[12:37:38] for Gen Z users to see we're friends, post the stories on a map. +[12:37:43] That's all the context and the task of what you're trying to accomplish. +[12:37:47] Include navigation buttons at the top and circles on the map. +[12:37:48] representing front stories. +[12:37:52] And again, that's the elements or pieces of the components. +[12:37:57] and then follow Facebook's design language. So that's trying to get to look like Facebook, that's constrained. +[12:38:05] Your prompt could be a lot simpler than this. This is just really trying to spell it out so you can get a sense of how you could approach whatever it is you're trying to do. +[12:38:09] And now you've got that face template that's on the original face. +[12:38:13] then you just start adding a feature. See, you got that, now let's build on top of it. +[12:38:18] so that we can really control the trajectory of what we're making with the bi-design tool. +[12:38:22] So an okay way to add a feature would be something like, hey, add an element for a feature. +[12:38:31] Let's start going to be vague once again. So a better way would be add an element or feature to the home screen to help the user with a context like a use case. +[12:38:36] And best would be adding additional concepts beyond that like user problem. +[12:38:38] So an example prompt there would be. +[12:38:40] design a search bar. +[12:38:41] at the bottom of the screen. +[12:38:55] for users who quickly want to find friend posts because they get overwhelmed by too many options. So you're describing overall the experience that you're trying to make while also explaining what this feature is trying to solve for. +[12:38:59] where friends are wanting to find their friend posts or users want to find their friend posts? +[12:39:10] And then the user problem being they get overwhelmed by too many options. So the AI is able to internalize this and help you experientially get a design out the door. +[12:39:13] And the third being make it interactive. +[12:39:25] So an okay version of that would be like when a user performs a behavior or interaction, like tapping on the input field at the bottom of that screen, show an element or a feature, have something happen. +[12:39:34] But again, we want to get more specific. So a better version of that would be when a user performs that interaction, show this feature that helps the user do. +[12:39:35] context. +[12:39:43] And then again, more specificity, keep adding to that prompt to get as specific as you can, like prioritizing these constraints. +[12:39:51] And here's an example of that once again, where you have the overall task. You're specifying the feature that you want to appear that solves this user problem. +[12:39:54] solves for this user desire, what they want to do. +[12:40:00] your constraint.