{ "updated_at": "2026-10-22T14:30:00+08:00", "page": "https://aitier.app/personal-agents/", "note": "Founder / team / official YouTube interviews only for /personal-agents/ Interviews & videos. Verified via oEmbed and yt-dlp. Takeaways are from the source file only (subtitles, chapters, or show notes as marked there) — nothing invented. HTML in index.html is the crawlable source of truth — keep this file in sync (copy the video-card block). No iframes on load. No creator reviews. No CNBC Altman clip. Lauren Tan & Roshan Sadanani: page label is 'Grok Bot team' only (no job titles). Official Muse tour: label 'Official demo'. Lee Robinson Stanford CS146S lecture: chip 'Talk', role from his X bio. Omit duration when the source file has none.", "groups": [ { "id": "claude-duet", "label": "Claude Duet", "note": "Official Claude-channel intro. No founder or team-member interview found yet (Karan Ahuja is named as team lead in Anthropic’s article; Mike Krieger is quoted in press about a launch video — this is the Claude channel upload).", "videos": [ { "id": "Fb-68zqcSVs", "url": "https://www.youtube.com/watch?v=Fb-68zqcSVs", "title": "Introducing Claude Duet", "channel": "Claude (official)", "published": "2026-10-22T03:35:00+08:00", "durationSeconds": 99, "speakers": [ { "name": "Claude", "role": "Official demo" } ], "kind": "official-demo", "sourceFile": "YouTube description (oEmbed author Claude / @claude; uploadDate 2026-10-21T12:35:35-07:00; lengthSeconds 99). Auto-captions could not be fetched at edit time.", "descriptionEn": "Official Claude-channel intro to Claude Duet: a new app where Claude comes to you with what needs your attention, keeps working after you put your phone down, decides what needs you and what can wait, and checks with you before it acts. Research preview waitlist for US Pro and Max.", "takeawaysZh": [ "Claude 官方频道短片《Introducing Claude Duet》（Official demo，约 1 分 39 秒）。", "片源简介：Claude 会主动把需要你关注的事找上门，手机放下后仍继续工作。", "它会判断什么需要你、什么可以等，并在代你行动前先征求确认。", "研究预览：美国 Pro / Max 用户可排队；简介链到 claude.com/claude-duet 与官方介绍文。" ], "takeawaysEn": [ "Official Claude-channel short “Introducing Claude Duet” (Official demo, about 1 min 39 sec).", "From the YouTube description: Claude comes to you with what needs your attention and keeps working after you put your phone down.", "It decides what needs you and what can wait, and checks with you before it acts on your behalf.", "Research preview waitlist for US Pro and Max; description links to claude.com/claude-duet and the official article." ] } ] }, { "id": "grok-bot", "label": "Grok Bot", "note": "Product lead and team interviews, plus a team lecture (Lee Robinson, Stanford CS146S; chip “Talk”). No founder interview found (Michael Truell, Aman Sanger, Sualeh Asif, Arvid Lunnemark).", "videos": [ { "id": "maSdsTLaMuU", "url": "https://www.youtube.com/watch?v=maSdsTLaMuU", "title": "How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)", "channel": "Lenny's Podcast", "published": "2026-09-08T20:30:00+08:00", "durationMinutes": 82, "speakers": [ { "name": "Roman Ugarte", "role": "Grok Bot product lead" } ], "kind": "interview", "descriptionEn": "Roman Ugarte, Grok Bot product lead, on Lenny's Podcast: a small standalone team built an internally usable version in 4 weeks and shipped publicly 3 weeks later, as a product of its own rather than inside Cursor. Early decisions were fully cloud, a computer per bot, and a colleague-pilled interface that hides internals; he also talks about Grok Bot as an always-on information filter and chief of staff, and says the moat was discovered, not planned.", "takeawaysZh": [ "一个很小的独立团队用 4 周从零做出内部可用版本，3 周后公开发布。", "刻意做成独立产品，而不是加进 Cursor 里（章节 08:40）。", "团队亲自手动 onboarding 了近 300 名早期用户（章节 11:20）。", "早期两个关键决策：完全上云、每个 bot 有自己的电脑；以及“colleague-pilled”理念，隐藏内部机制，让委派任务像交给同事（章节 14:29、30:02、39:20）。", "还谈到 Grok Bot 作为 always-on 的信息过滤器和“幕僚长”，以及护城河是“发现出来的，不是规划出来的”（章节 51:04、1:11:45）。" ], "takeawaysEn": [ "A very small standalone team built an internally usable version from scratch in 4 weeks and shipped it publicly 3 weeks later.", "They deliberately built Grok Bot as a standalone product instead of adding it inside Cursor (chapter 08:40).", "The team personally hand-onboarded nearly 300 early users (chapter 11:20).", "Two early decisions: fully cloud, and each bot gets its own computer; plus a “colleague-pilled” idea — hide the internals so delegating a task feels like handing it to a colleague (chapters 14:29, 30:02, 39:20).", "They also talk about Grok Bot as an always-on information filter and “chief of staff” (幕僚长), and say the moat was discovered, not planned (chapters 51:04, 1:11:45)." ] }, { "id": "A63sedG-p5Q", "url": "https://www.youtube.com/watch?v=A63sedG-p5Q", "title": "Are Agents About to Replace Software Engineering? | Lauren Tan and Roshan Sadanani", "channel": "MTS", "published": "2026-08-14T06:00:00+08:00", "durationMinutes": 28, "speakers": [ { "name": "Lauren Tan", "role": "Grok Bot team" }, { "name": "Roshan Sadanani", "role": "Grok Bot team" } ], "kind": "interview", "descriptionEn": "Lauren Tan and Roshan Sadanani of the Grok Bot team on MTS, the day after launch: a live demo of what Grok Bot can do, how it first found product-market fit inside Cursor, the Grok 4.6 announcement and how model efficiency opens new uses, plus a goal of mainstream users — not only developers — and a preview of later features.", "takeawaysZh": [ "现场演示 Grok Bot 能做什么（00:04 Live demo）。", "讲 Grok Bot 怎么先在 Cursor 内部找到 PMF。", "同场宣布 Grok 4.6，并谈模型效率如何带来新用法（00:07、00:24）。", "目标是主流用户，不只是开发者（00:11）；预告了后续功能（00:22）。" ], "takeawaysEn": [ "A live demo of what Grok Bot can do (00:04 Live demo).", "How Grok Bot first found product-market fit inside Cursor.", "They announced Grok 4.6 on the same show and talked about how model efficiency opens new uses (00:07, 00:24).", "The goal is mainstream users, not only developers (00:11); they also previewed later features (00:22)." ] }, { "id": "2Hi7GVb1yqo", "url": "https://www.youtube.com/watch?v=2Hi7GVb1yqo", "title": "How Cursor's Browser Experience Shaped Grok Bot: Nick Miller Interview", "channel": "Callstack", "published": "2026-10-12T13:18:00+08:00", "durationMinutes": 10, "speakers": [ { "name": "Nick Miller", "role": "SpaceXAI Lead Field Engineer" } ], "kind": "interview", "descriptionEn": "Nick Miller, SpaceXAI Lead Field Engineer, on Callstack: Cursor and SpaceXAI started with borrowing Colossus to train models, then that relationship became an acquisition. Years of Cursor work on the agent harness and a built-in browser got Grok Bot to an internal beta in about a month. Bots run in isolated VMs and finish tasks with an API, an MCP server, or a browser UI.", "takeawaysZh": [ "Cursor 和 SpaceXAI 的关系始于借用 Colossus 训练模型，后来发展成收购。", "Cursor 多年做 agent harness 和内置浏览器的积累，让 Grok Bot 大约一个月就进了内部 beta。", "bot 跑在隔离的虚拟机里，按情况用 API、MCP server 或浏览器界面完成任务。" ], "takeawaysEn": [ "Cursor and SpaceXAI started with borrowing Colossus to train models, then that relationship became an acquisition.", "Years of Cursor work on the agent harness and a built-in browser let Grok Bot reach an internal beta in about a month.", "Bots run in isolated VMs and finish tasks with an API, an MCP server, or a browser UI as needed." ] }, { "id": "L1OBRtfPS34", "url": "https://www.youtube.com/watch?v=L1OBRtfPS34", "title": "Stanford CS146S: How always-on agents work", "channel": "leerob", "published": "2026-10-12T01:58:00+08:00", "durationMinutes": 45, "speakers": [ { "name": "Lee Robinson", "role": "Superintelligence at SpaceX, previously Cursor" } ], "kind": "talk", "descriptionEn": "Lee Robinson’s Stanford CS146S lecture on how always-on agents work, using Grok Bot as the example: an agent that runs on a server with its own cloud Linux computer, wakes on messages and events, and knows when to stay quiet. He covers the architecture (servers and Firecracker VMs that sleep and wake, durable workflows), a harness whose only client tool is send to user, context engineering with prompt caching and compaction, layered memory, and the eval-and-training flywheel.", "source": "YouTube auto-captions (English) + YouTube chapters; the description is one line", "takeawaysZh": [ "always-on agent 跑在服务器上：合上电脑、重启、崩溃都不停；它有自己的云端 Linux 电脑，会积累记忆，知道什么时候不说话，可以被 Slack、邮件等事件唤醒（章节 00:00、02:34）。", "架构：服务器和电脑平时休眠、需要时唤醒；电脑是可快照的 Firecracker 虚拟机，agent loop 放在服务器而不是 VM 上；用 Temporal 做 durable workflow，失败重试从出错那一步接着跑（章节 10:04、14:05）。", "harness：客户端和服务器之间只有一个工具“send to user”；工具按需发现，只把工具名放进上下文；先用对话记忆、插件/MCP、搜索、浏览器、整台桌面，都不行才去打扰用户；子 agent 帮手让主对话保持精简（章节 18:06）。", "安全与上下文：auto review 由另一个模型检查每条 shell 命令，发邮件这类不可逆操作先问用户；提示词前缀保持不变以命中 prompt cache，长输出写进文件，趁缓存还热就做压缩；记忆分为核心用户事实、按日期的日志和临时草稿（章节 27:24、32:50）。", "飞轮与展望：出错变成 eval，修 harness，再训练新模型；Cursor 和 SpaceX 的原则是“delete the product”，为下一代模型而做；他预计 agent 很快能连续工作数周甚至数月，而模型每月一发时如何评估跑一个月的 agent 还是难题（章节 34:29、37:43）。" ], "takeawaysEn": [ "An always-on agent runs on a server, so closing your laptop, restarts and crashes don’t stop it; it has its own cloud Linux computer, builds up memory, knows when to stay quiet, and can be woken by Slack messages, email or other events (chapters 00:00, 02:34).", "Architecture: the server and the computer sleep and wake as needed; the computer is a Firecracker VM that can be snapshotted, and the agent loop lives on the server, not the VM; durable workflows on Temporal mean a retry after a failure resumes at the failed step (chapters 10:04, 14:05).", "Harness: the only tool between client and server is “send to user”; tools are discovered on demand with only their names in context; the agent tries conversation memory, plugins/MCP, search, the browser and the full desktop before bothering the user; sub-agent helpers keep the main conversation sparse (chapter 18:06).", "Safety and context: auto review has a separate model check every shell command, irreversible actions like sending email ask the user first; a static prompt prefix keeps the prompt cache warm, long outputs go to files, and compaction runs while the cache is still warm; memory is layered into core user facts, dated logs and ephemeral scratchpad notes (chapters 27:24, 32:50).", "Flywheel and outlook: failures become evals, the harness gets fixed, and new models train on them; the Cursor and SpaceX principle is to “delete the product” and build for the next model generation; he expects agents to work for weeks or months soon, and says evaluating a month-long agent when models ship monthly is still unsolved (chapters 34:29, 37:43)." ] } ] }, { "id": "instinct", "label": "Instinct", "note": "Founder interview. No other Instinct founder or team-member video interview found.", "videos": [ { "id": "Am7IWP8IpEc", "url": "https://www.youtube.com/watch?v=Am7IWP8IpEc", "title": "The $10B AI Assistant Challenging Meta's Muse, Grok's Bots, and OpenAI's Dots", "channel": "Invest Like The Best", "published": "2026-09-28T20:00:00+08:00", "durationMinutes": 85, "speakers": [ { "name": "Noah Shinn", "role": "Instinct founder and CEO" } ], "kind": "interview", "descriptionEn": "Noah Shinn, Instinct founder and CEO, on Invest Like The Best: Instinct has no app — its own phone number and computer — so users text, call, or email it, and it can call them. Invite-only, about 10% daily growth, no marketing spend; a take-rate model with more than $1 billion annual GMV on the platform; compute is the biggest challenge (40% of his time), and he also talks about Muse.", "takeawaysZh": [ "没有 App：Instinct 有自己的电话号码和电脑，用户发短信、打电话、发邮件就能用，它也会主动给你打电话。", "仅限邀请，日增约 10%，没花营销费（据播客节目说明）。", "讲用户要多久才肯把信用卡、收件箱和日历交给 agent；“Instinct to Instinct”可信网络让两个人的 agent 直接协调。", "商业模式：不做广告，靠抽成（take rate），平台上每年流过的交易额超过 10 亿美元（据播客节目说明）。", "算力是最大挑战，他说自己 40% 的时间在想算力；还专门谈了对 Muse 的看法（播客章节 01:02:30）。" ], "takeawaysEn": [ "There is no app: Instinct has its own phone number and computer. Users text, call, or email it, and it can call them back.", "Invite-only, about 10% daily growth, with no marketing spend (from the podcast show notes).", "How long it takes users to trust an agent with a credit card, inbox, and calendar; an “Instinct to Instinct” trust network lets two people’s agents coordinate directly.", "Business model: no ads, a take rate, and more than $1 billion in annual GMV flowing through the platform (from the podcast show notes).", "Compute is the biggest challenge — he says he spends 40% of his time on it — and he also talks about Muse (podcast chapter 01:02:30)." ] } ] }, { "id": "muse", "label": "Muse", "note": "Founder / Chief AI Officer interviews, plus Meta’s official product tour labeled Official demo.", "videos": [ { "id": "S6l3aRsecuE", "url": "https://www.youtube.com/watch?v=S6l3aRsecuE", "title": "Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta", "channel": "David Senra", "published": "2026-10-11T20:00:00+08:00", "durationMinutes": 72, "speakers": [ { "name": "Alexandr Wang", "role": "Meta Chief AI Officer, Scale AI co-founder" } ], "kind": "interview", "descriptionEn": "Alexandr Wang, Meta Chief AI Officer and Scale AI co-founder, on David Senra: Muse started from OpenClaw inside MSL — his first use was a self-analysis he compared to three years of therapy — then a February prototype and seven months to a reliable consumer product. Meta chose personal and consumer AI over coding; a good product needs a single point of view (Nat Friedman’s on Muse), plus trust, Zuckerberg’s product obsession, and rebuilding Meta’s AI lab.", "takeawaysZh": [ "起点是 OpenClaw：Nat Friedman 最先在 MSL 内部深度使用 OpenClaw，Wang 跟着用，第一件事是一次“相当于三年心理治疗”的深度自我分析。", "2 月就有了第一个原型，Muse 的很多元素当时已经在了；之后花 7 个月把“偶尔很神奇、多数时候会坏”的原型打磨成可靠的消费级产品。", "为什么做个人 agent 而不是 coding：年初 Anthropic 在 coding 上遥遥领先，业界普遍认为只有 coding agent 重要，Meta 则认定个人和消费级 AI 才是自己能做出特别东西的领域。", "好产品需要“单一视角”，他说 Muse 上最明显的就是 Nat Friedman 的视角；也讲了用 meme 帮产品传播。", "还谈到怎么建立用户对 agent 的信任、扎克伯格对产品的执着，以及在 Meta 重建 AI 实验室。" ], "takeawaysEn": [ "It started with OpenClaw: Nat Friedman used it deeply inside MSL first; Wang followed, and his first use was a deep self-analysis he compared to “three years of therapy” (三年心理治疗).", "A first prototype existed in February, already with many Muse elements; they then spent 7 months turning a prototype that was occasionally magical and usually broken into a reliable consumer product.", "Why personal agents instead of coding: early in the year Anthropic led coding by a wide margin, and the industry treated coding agents as the only thing that mattered; Meta decided personal and consumer AI was where it could make something distinctive.", "A good product needs a “single point of view” (单一视角); on Muse he says the clearest one is Nat Friedman’s. He also talks about using memes to help the product spread.", "He also covers how to build user trust in an agent, Zuckerberg’s product obsession, and rebuilding Meta’s AI lab." ] }, { "id": "XfdIhd1g6EI", "url": "https://www.youtube.com/watch?v=XfdIhd1g6EI", "title": "Zuckerberg's $100 Billion Bet", "channel": "Cheeky Pint", "published": "2026-10-13T05:10:00+08:00", "durationMinutes": 134, "speakers": [ { "name": "Mark Zuckerberg", "role": "Meta founder/CEO" }, { "name": "Alexandr Wang", "role": "Meta Superintelligence Labs Chief AI Officer" } ], "kind": "interview", "descriptionEn": "Mark Zuckerberg (Meta founder/CEO) and Alexandr Wang (MSL Chief AI Officer) on Cheeky Pint with John Collison: viral Muse uses such as sourcing rare teas from small farms in Taiwan and Japan; “the model is the product,” with much of Muse’s product work on privacy and safety; Muse Wonder trained end-to-end for personal and professional agents with active memory; agentic commerce; glasses as a multimodal agent entry; AI not winner-take-all.", "takeawaysZh": [ "开场聊 Muse 的爆款用例，比如 Wang 让自己的 Muse 每周从台湾、日本的小茶园淘冷门茶叶（章节 00:24）。", "“The model is the product”：产品越来越薄，模型越来越像产品本身；Muse 背后大量的产品工作在隐私和安全模型上。", "新模型 Muse Wonder 是端到端为个人和职业 agent 训练的，重点投入“active memory”，他们认为记忆是可靠 agent 的必要条件。", "Agentic commerce：需求会逐步从现有渠道转向 agent 渠道，就像电商之于实体店。", "眼镜会成为纯多模态的 agent 入口；他们也认为 AI 不会赢者通吃（章节 01:20:06、01:34:06）。" ], "takeawaysEn": [ "They open with viral Muse uses, including Wang having his Muse source uncommon teas each week from small farms in Taiwan and Japan (chapter 00:24).", "“The model is the product”: the product layer gets thinner and the model looks more like the product itself; a lot of Muse product work is on privacy and safety models.", "The new Muse Wonder model is trained end-to-end for personal and professional agents, with a focus on “active memory,” which they treat as a requirement for a reliable agent.", "Agentic commerce: demand will gradually move from existing channels to agent channels, the way e-commerce moved demand from physical stores.", "Glasses become a fully multimodal agent entry point; they also argue AI will not be winner-take-all (chapters 01:20:06, 01:34:06)." ] }, { "id": "wHn0hTjvFoo", "url": "https://www.youtube.com/watch?v=wHn0hTjvFoo", "title": "Take the full tour of Muse, Meta's personal AI agent.", "channel": "Muse (official)", "published": "2026-09-09T03:00:00+08:00", "speakers": [ { "name": "Alex Cornell", "role": "Official demo" } ], "kind": "official-demo", "descriptionEn": "Official demo: Alex Cornell gives the full tour of Muse, Meta’s personal AI agent — buying movie tickets, planning family weekends, and helping with chess, covering chat, a personalized feed, proactive ideas, goals, and the library.", "takeawaysZh": [ "Meta 官方 Muse 完整导览（Official demo），演示者是 Alex Cornell。", "演示包括买电影票、规划家庭周末、陪练国际象棋。", "导览覆盖聊天、个性化信息流、主动建议、目标追踪和资料库。" ], "takeawaysEn": [ "Meta’s official full tour of Muse (Official demo), presented by Alex Cornell.", "The demo includes buying movie tickets, planning family weekends, and helping with chess.", "The tour covers chat, a personalized feed, proactive ideas, goals, and the library." ] } ] } ] }