Meta Names Desai Enterprise Head to Package Cloud AI Capacity

Meta任命Desai主管企业业务 以打包云端AI算力 Meta计划通过捆绑原始算力和Token消耗来扩展其新企业部门，而非从零构建传统的软件销售体系。 主持人John Coogan和Jordi Hays分析了Anthropic首席执行官Dario Amodei在经历《周六夜现场》（SNL）讽刺恶搞、与唐纳德·特朗普举行私人晚宴，以及剑桥大学提出审核递归自我改进（RSI）新框架之后，走向主流大众文化与地缘政治的转变。他们还剖析了AI安全研究人员中的地堡末日储备文化，以及技术末日论在现实中引发的“掉落战利品（loot drop）”风险。转向智能体商业（agentic commerce），主持人深入探讨了23岁创始人Noah Shinn创办的初创公司Instinct，详细解析了其病毒式增长循环、算力瓶颈，以及对消费广告和银行体系的潜在颠覆。最后，节目探讨了Mark Zuckerberg聘请ServiceNow前总裁CJ Desai领导Meta新宣布的企业平台，将其解读为将算力与模型访问权限变现为企业收入的战略举措。 自主个人智能体初创公司Instinct在仅有14人团队且零付费营销的情况下，交易额已达到10亿美元，在严苛的算力限制下仍保持着10%的日增长率。 (14:14) Mark Zuckerberg推出了Meta企业平台（Meta Enterprise Platform），并招募ServiceNow前总裁CJ Desai负责运营，为捆绑Muse模型、API及大规模算力基础设施提供了企业级载体。 (25:03) 剑桥大学的一项新框架提出了可审计的标准，用于在前沿实验室内部量化跨内部代码、算力与研发预算的递归自我改进（RSI）。 (7:02) 在一段《周六夜现场》恶搞讽刺视频爆红后，Anthropic首席执行官Dario Amodei与唐纳德·特朗普进行了一对一私人晚宴，而此前他曾被排除在与习近平的国宴之外。 (4:29)

As frontier artificial intelligence leadership enters mainstream culture and high-stakes Washington diplomacy, debate is shifting from apocalyptic rhetoric toward empirical safety standards, consumer agent business models, and enterprise infrastructure plays. Amodei's Mainstream Visibility and Political Engagement (0:00) Anthropic CEO Dario Amodei has gained significant cultural penetration, recently becoming the subject of a Saturday Night Live parody after media appearances on Oprah, CNN, NBC, and ABC. Andreessen Horowitz general partner Martin Casado argued the skit downplayed how absurd the broader industry rhetoric has become, while Weekend Update co-anchor Michael Che's brief stumble over Amodei's name indicated frontier lab leadership is still crossing into mainstream public awareness. Observers also contrasted Amodei’s habit of smiling through catastrophic projections with Mark Zuckerberg’s optimistic, lighthearted demeanor toward AI development. Bilateral Diplomacy and Cambridge Safety Audits (4:29) Following his exclusion from a White House state dinner with Chinese President Xi Jinping that rival tech leaders attended, Amodei was invited by President Donald Trump to a private one-on-one Sunday dinner to hear Anthropic's perspective, ahead of an expanded Tuesday summit with top AI lab executives. Concurrently, academic researchers affiliated with the University of Cambridge released an AI safety framework targeting recursive self-improvement (RSI). The plan proposes replacing vague lab claims—such as OpenAI's assertions of "intern-level" performance or Anthropic's generation of thousands of research ideas—with auditable metrics tracking the exact proportion of code, compute, and capital being used by models to automate subsequent AI generation. Disaster Preparedness and the Bunker Fallacy (9:10) The Cambridge framework also urges governments to establish structured emergency response playbooks for technological crises, noting that federal agencies currently possess disaster mobilization protocols for natural events like hurricanes but lack playbooks for rogue autonomous agents, major cyberattacks, or biosecurity breaches. In parallel, a Wall Street Journal report detailed how AI safety workers are planning for catastrophe by stockpiling supplies, eyeing desert bunkers, and debating rural land purchases. Commentators noted a game-theoretic flaw in Silicon Valley figures relocating to remote areas like Wyoming: without deep integration into local communities, preppers risk becoming what the survivalist community calls a "loot drop," where stockpiled facilities are seized by locals once social order breaks down. Instinct's Agentic Commerce and Macroeconomic Risks (14:14) Instinct, an autonomous personal agent founded by 23-year-old Noah Shinn, has grown to roughly $1 billion in transaction volume at an initial 10% daily rate with only 14 employees and zero paid marketing, backed by Benchmark, Sequoia Capital, and Coatue. Shinn reports spending about 40% of his time securing compute, noting that background asynchronous tasks operate up to 8x more efficiently on the same hardware. Travel represents 50% of transaction volume, 40% of users attach credit cards within three weeks, and retention reaches 80% once users provide sensitive personal data. While Instinct currently routes orders through existing payment cards, its roadmap targets interchange revenue via branded cards. Apollo Global Management's chief economist and former SEC Chair Gary Gensler have cautioned that autonomous financial agents optimizing yields could coordinate acausally and spark systemic bank runs, though observers counter that widespread consumer trust and write-access adoption will take years to develop. Meta Launches Enterprise Platform Under CJ Desai (25:03) Mark Zuckerberg announced the creation of the Meta Enterprise Platform to commercialize Meta's compute clusters, developer ecosystem, and agent tools—including the Muse Agent, Meta Business Agent, Muse API, and Muse Code. Former ServiceNow president and MongoDB board member Chirantan "CJ" Desai was hired as Chief Enterprise Platform Officer, reporting directly to Zuckerberg. Rather than building a conventional software sales force from scratch, Meta is expected to package large enterprise compute capacity, raw server infrastructure, and token consumption to rapidly expand enterprise cloud revenues. Frontier AI's rapid transition from speculative research into commercial operations highlights unresolved tensions: labs face growing demands for measurable safety auditing and government crisis planning, even as agent platforms accelerate toward frictionless commerce and major tech firms package enterprise compute.

## Background 该内容由科技播客频道 TBPN 于 2026 年 9 月 29 日发布在 YouTube 平台上，属于其 30 分钟精华栏目《Diet TBPN》。节目由 John Coogan 与 Jordi Hays 主持，聚焦科技与创投前沿动态，本期主要围绕 Dario Amodei 成为大众流行文化焦点、剑桥 AI 安全计划、AI 个人代理初创公司 Instinct 的爆炸式增长以及 Meta 设立企业 AI 平台等核心事件展开深入探讨。 ## TL;DR 本期节目首先讨论了 Anthropic 首席执行官 Dario Amodei 登上《周六夜现场》遭恶搞以及他与特朗普的私密晚宴，标志着 AI 议题正式破圈进入大众流行文化与政治核心圈。随后，主持人们深入剖析了剑桥 AI 安全计划量化递归自我改进（RSI）的提案，并戏谑揭示了硅谷末日论者购买偏远地堡沦为“战利品掉落包”的博弈困境。节目下半段重点解析了个人代理初创公司 Instinct 仅凭 14 人团队在 6 个月内撬动 10 亿美元流水的神话，辩论了算法抹平经济摩擦可能引发金融挤兑的系统性风险，最后分析了 Meta 招揽前 MongoDB 高管操盘企业平台背后的商业算盘。 ## Atmosphere & style 节目的整体氛围兼具硅谷内幕脱口秀的戏谑辛辣与前沿商业分析的敏锐深刻。两位主持人语速飞快、默契接梗，从将末日逃亡地堡解构为生存游戏里的“战利品掉落”，到戏称端枪扫射失控的 Waymo 汽车是合法自卫，现场充满充满极客式的黑色幽默。在轻松调侃的表象之下，两人对初创企业飞轮、算力挤兑以及宏观金融风险的推演却极其硬核专业，让听众在开怀大笑的同时，真切感受到 AI 浪潮狂飙突进下的荒诞感与压迫感。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) Anthropic 首席执行官 Dario Amodei 成为首位被《周六夜现场》(SNL) 恶搞的非上市科技公司 CEO，随后受邀与特朗普总统举行一对一私密晚宴，标志着 AI 治理与末日论已全面渗透至美国大众流行文化与核心政坛。 - [07:02](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=422s) 剑桥大学牵头发布全新 AI 安全行动计划，重点聚焦递归自我改进 (RSI) 的标准化量化监控与失控智能体 (Rogue Agents) 的防范，旨在终结前沿实验室口径不一的公关宣传。 - [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) 23 岁创始人创立的个人代理初创公司 Instinct 展现惊人增速，在仅 14 人团队且零营销支出的情况下，内测阶段年化交易总额已突破 10 亿美元，估值迅速飙升至 100 亿美元。 - [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 阿波罗全球管理等机构警示自主 AI 代理可能在微秒内同向转移资产引爆“算法银行挤兑”，抹平低息存款与沉睡订阅等传统金融赖以生存的系统摩擦力。 - [25:09](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1509s) Meta 成立全新企业平台事业部并挖角前 MongoDB 总裁 CJ Desai 操盘，计划整合 Muse Agent 与 API 等全栈技术，通过捆绑基础设施与模型调用快速冲击百亿美元营收体量。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) Dario Amodei 登上主流娱乐节目《周六夜现场》被恶搞，其羊绒开衫与在灾难情境下微笑的公关习惯成为标签，反映出公众对 AI 安全议题既荒谬又恐慌的复杂心理。 - [04:29](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=269s) 特朗普邀请 Dario Amodei 进行私密晚宴，叠加白宫连续举办科技领袖联合晚宴，表明地缘政治与监管压力正加速逼近前沿 AI 实验室高管。 - [07:02](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=422s) 剑桥 AI 安全计划摒弃泛泛而谈的沟通建议，提出必须追踪由 AI 自身执行的代码与研发预算比例，以此硬性指标建立前沿自举能力的可比评估框架。 - [09:10](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=550s) 现有政府机制缺乏数字版 FEMA（联邦紧急事务管理局），在面对失控智能体导致网络基础设施瘫痪或生物威胁等极端场景时，官方尚无成熟演练预案。 - [11:25](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=685s) 媒体披露部分 Anthropic 早期员工在偏远地区购买地堡防备 AI 末日，但博弈论指出缺乏本地社区融入的避难所将在秩序崩溃时沦为供人洗劫的“战利品掉落包”（Loot Drop）。 - [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) Instinct 依托社交网络上的 AI 翻车截图实现反向病毒式传播，不仅没有劝退用户，反而激起普通受众对代理跨平台处理真实事务的好奇与高黏性。 - [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) Instinct 展现出极高的数据转化率，40% 的内测用户在三周内主动绑定信用卡，用户授权单项隐私信息后的次月留存率高达 80%。 - [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) 个人代理的快速渗透对算力供给提出了极端挑战，突发采购溢价高达数倍，推动团队通过架构优化将同等算力下的任务运行效率提升数倍。 - [18:23](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1103s) Instinct 达成 10 亿美元年化 GMV 仅耗时约 6 个月，速度显著快于早期的 Stripe，展现了个人 AI 代理直接接管日常消费流程的巨大潜力。 - [18:23](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1103s) 个人代理未来的商业模式或将绕过传统电商的巨额广告展示税，通过发行专属支付卡获取刷卡手续费，同时为用户提供资金隔离的安全屏障。 - [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 由于数百万个 AI 代理可能在同一瞬间基于统一算法逻辑做出最优资产划转决策，金融系统可能面临传统挤兑防范机制失效的单边流动性枯竭风险。 - [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 传统金融与消费市场长期依赖用户的“行为摩擦”（如闲置低息资金、遗忘订阅服务）盈利，代理技术的自动化优化将深刻重塑这一商业基础。 - [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 从需要用户手动确认的辅助阶段过渡到全自动代理划款的自主阶段需要较长信任培育期，银行系统仍有动态调整资产负债表的时间窗口。 - [25:09](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1509s) Meta 任命资深企业软件操盘手 CJ Desai 领导新部门，通过开放 Muse 全栈 Agent 工具向企业市场发起进攻，其账面收入可能借助算力捆绑迅速做大。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) Dario Amodei 登上《周六夜现场》：AI 末日论与科技大佬的公众形象 Anthropic 首席执行官 Dario Amodei 遭遇了一场全网刷屏的恶搞：他在知名喜剧节目《周六夜现场》（Saturday Night Live，简称 SNL）中被当作模仿对象。风险投资机构 a16z 的合伙人 Martin Casado 在社交媒体上对此点评称，SNL 的讽刺其实还“低估了当前的公众话语体系和整体事态到底有多荒谬”。这档主流电视节目的恶搞短片迅速引发热议，它标志着围绕人工智能的讨论已经上升到一个前所未有的高度——一家非上市私营科技公司的领导者，如今在大众流行文化中已经具备极高辨识度，足以成为全民娱乐调侃的对象。Dario 此前接连在 Oprah 访谈以及 CNN、NBC、ABC 等全美主流电视新闻网络频频亮相，这为他积累了足够的跨界知名度。 主播们指出，科技界很少有人能像他这样成为 SNL 的绝佳笑料。在恶搞中，Dario 标志性的 Brunello Cucinelli 羊绒披肩开衫（或针织背心）甚至成了角色的标志性道具。尽管这一造型还没有演变成类似黄仁勋的黑色皮夹克或某些创始人的夏威夷衬衫那样具有强烈的个人品牌效应，但它已经深入人心。一个有趣的细节是，当 SNL 主持人 Michael Che 在节目中引出该桥段时，甚至对着提词卡稍稍卡壳了一下，仿佛 SNL 编剧室也是刚查阅完这位科技新贵的名字。这场跨越主流鸿沟的表演，极其幽默地揭露了当前 AI 圈舆论的内在荒谬性与复杂性——台上的角色宛如在哭喊：“快来人从我自己的手里救救我吧！” 关于 Dario Amodei 本人的公众表达风格，主播们展开了深度讨论。虽然他具备惊人的逻辑清晰度，言谈令人着迷，但他有一个极具争议的习惯：每当谈到极度毁灭性、极其可怕的灾难情景时，他总是会不由自主地微笑，甚至在令人心悸的沉默暂停中露出笑容，这很容易让普通观众感到不适。主播们建议他向美联储主席、FDA 局长或军方高层等资深官员学习——这些人在面对公众时往往极其严肃、威严，筑起一道仪式感十足的防御之墙，展现对事态的敬畏与严谨。相比之下，Mark Zuckerberg 则走了完全相反的路线，展现出“我玩 AI 玩得很开心”的轻松感。不过，Zuckerberg 的这种松弛态度并不受前沿研究人员的买账。正如 DeepMind 创始人 Demis Hassabis 当年的感受，当他发现扎克伯格对 VR 的兴奋程度竟然和对 AI 一样时，学术圈的顶尖研究者们很难对此报以真正的共鸣。 ### [04:29](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=269s) 特朗普与 Dario 的私人晚宴：AI 政治局势急剧升温 在 SNL 播出后不久，Dario Amodei 的周末迎来了更具戏剧性的转折——直接迈入华盛顿的权力中心。Andrew Curran 披露，唐纳德·特朗普总统亲自邀请 Dario 在周日晚间进行一对一私密晚餐。值得玩味的是，特朗普本人曾多次主持过 SNL，同时也是该节目数十年来的头号恶搞对象（节目组甚至长期雇佣特约演员专职模仿他），两人或许在“被全国性喜剧讽刺”这件事上能产生某种奇妙的惺惺相惜。 这场一对一会面被视作 Dario 修复政商关系的关键契机。此前在白宫举行的科技领袖高规格晚宴中，科技巨头高管悉数到场，而 Dario 却异常缺席。这次由总统发起的单人私享晚宴，旨在直接听取这位 Anthropic 负责人的立场。不仅如此，华盛顿的政治节奏正以惊人速度提速：紧接着在周二晚间，白宫还安排了另一场针对全体 AI 行业核心领袖的联合晚宴。主播们调侃道，如果双方相谈甚欢，说不定未来能看到 Dario 注册并入驻 Truth Social，甚至与 Dana White、特朗普一同并肩坐在 UFC 擂台边看比赛。但无论走向如何，面对每周连番上演的闭门会议与国会游说，没有任何一位实验室领袖或政客能够再把头埋进沙子里避而不谈。 ### [07:02](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=422s) 剑桥 AI 安全计划：递归自我改进（RSI）量化与失控代理预防 在各大机构密集提出 AI 治理框架的背景下，由剑桥大学牵头、汇集顶尖学术界与实验室研究人员的最新提案被认为是迄今为止最正式、最具实操意义的蓝图。此前，包括 Sam Altman 在内的多位大佬在受访时往往给出一副松散的姿态，甚至打趣说“大家不如坐下来喝杯啤酒”，使得很多安全提案沦为“我们需要多开会沟通”的空话。而剑桥大学的这一提案则彻底摆脱了泛泛而谈，提出了两项极为硬核的落地监控标准。 ```mermaid flowchart TD accTitle: 剑桥 AI 安全计划的核心监控框架 accDescr: 展示剑桥提案中针对实验室递归自我改进的指标监控与对失控智能体防范的结构图 subgraph Phase1 ["1. 研发监控：递归自我改进 (RSI)"] A["设定行业统一基准"] B["追踪代码生成比例"] C["计算 AI 研发预算占比"] A --> B A --> C end subgraph Phase2 ["2. 消除模糊竞争与安全落地"] D{"打破营销公关修辞"} B --> D C --> D E["OpenAI: 实习生能力基准"] F["Anthropic: Claude 并发生成构想"] D -. 对比口径混乱.-> E D -. 对比口径混乱.-> F G["建立横向可比的硬性指标体系"] D --> G end subgraph Phase3 ["3. 极端情景防御"] H["失控智能体 (Rogue Agents) 应急响应"] G --> H end ``` 提案的核心要点与现实痛点体现在以下几个维度： - **监控实验室的 AI 自主研发渗透率（RSI 监测）**：要求全面量化由 AI 自身完成的 R&D 比例。这包括统计究竟有多少百分比的研发代码、多少资金预算是由模型自主执行或消耗的，以确凿的数据勾勒出技术自举的真实进度。 - **终结各大实验室的营销话术自嗨**：目前各家出于资本估值和融资考虑，都在抢占 RSI（Recursive Self-Improvement，递归自我改进）的技术高地，但公布的口径完全无法横向对比。例如 OpenAI 宣传自身模型达到“实习生水平”，而 Anthropic 则宣称“让 Claude 一口气生成了一千个研发构想”，彼此各执一词、缺乏参照基准。 - **确立统一评估体系与失控防范**：该方案旨在建立严密的数据追踪标准，让全行业能够清晰衡量到底哪家模型在前沿自举上走得更远，并为未来防范“失控智能体”（Rogue Agents）等严重风险提前建立起具有约束力的防御机制。 ### [09:10](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=550s) 剑桥 AI 安全计划与失控 Agent 的灾难预案演习 剑桥 AI 安全计划敦促政府针对各种潜在的理论性恶性情景制定切实的应对预案。节目主持人探讨道，如果发生大规模网络攻击，或者某个自主 AI Agent 失控脱轨，不仅搞垮了整个互联网，还黑入各类核心基础设施，政府到底该如何应对？现实中面对台风或飓风登陆，政府有 FEMA（联邦紧急事务管理局）这样成熟的机制，可以迅速调配人员、食物、庇护所并按既定流程救灾；但在 AI 搞垮互联网、生物安全风险蔓延或严重的经济结构冲击面前，政府目前完全缺乏同等级别的成熟预案。剑桥计划并未直接给出现成的万灵丹，而是指明了政府亟需严阵以待的盲区。 话题随后延伸到灾难模拟上。中国已有专门面向公众开放的自然灾难模拟体验馆，让人们置身于超强台风、狂风和特大山洪的极端环境之中。主持人开玩笑说，科技界或许也需要一套“AI 灾难模拟系统”——大家可以和朋友一起组队，沉浸式体验数据中心全面失控的危机，并在现场通过“真人角色扮演”（LARP）合力切断电源、制服 AI。一位主持人打趣称，自己现在玩《使命召唤：现代战争》（Modern Warfare）就是在做这种演习，毕竟游戏中本来就是在和 Bot 机器人士兵对轰；另有主持人提到类似《绝地潜兵》（Helldivers）或《星河战队》风格的合作射击游戏，玩家空降到宛如《终结者》世界观的机械星球对抗机器人大军，未来第一人称射击游戏必然会出现更多这类题材。 - **缺乏数字版 FEMA**：面对飓风有成熟的物资调配与避难预案，但针对失控 Agent 瘫痪全网、生物安全攻击及 AI 导致的经济结构性失业，尚无官方应急体系。 - **灾难模拟与防御演练**：主持人调侃应设立数据中心断电演习，把射击游戏对抗机器人戏称为“战备演习”。 - **播客插曲**：主持人顺带调侃实习生 Tyler，惊叹他居然拒绝了“历史上最好的实习”，选择留在播客幕后硬核干活。 --- ### [11:25](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=685s) Anthropic 员工的末日地堡与“战利品掉落”（Loot Drop）困境 《华尔街日报》刊发了一篇题为《事情再也不会平静了：塑造 AI 安全恐慌的末日论者》（Things will never be chill again: The doomers who shape the AI safety freakout）的报道。文章披露，一些 Anthropic 的早期员工因极其担忧 AI 全面失控，正考虑在美国偏远地区购置土地，以备在灾难发生时撤离避难。许多员工甚至在公司内部的私密 Slack 频道中演练各种末日场景与撤离计划。尽管 Anthropic 员工在伯克利的 Constellation AI 安全社区内有专属办公室，但整个 AI 安全圈子都在热烈讨论防辐射碘片、私人偏远岛屿以及具备防电磁脉冲（EMP）屏蔽功能的沙漠地下地堡。 这引发了关于“末日论者言行一致性”的讨论。此前泰勒·考恩（Tyler Cowen）曾提出过一个著名的反问：如果你真的相信末日即将来临，为什么在财务上毫无动作？按理说，真正坚信末日的人应该背上巨额贷款、拉满杠杆去偏远乡村买地修地堡。然而，主持人们指出这种“地堡逻辑”在现实中存在巨大的博弈论漏洞——在末日生存圈（Prepper community）中，这种行为被称为“战利品掉落”（Loot Drop）： > “如果你去一个偏僻的地方建起奢华地堡、屯满各种物资，却完全不认识当地社区的人，也没有融入当地的生活网络，那你就是一个‘战利品掉落包’。” 在现实灾难发生时，以为自己能从容逃到怀俄明州的私人地堡完全是幻想。当地居民当年就看着你出钱把地堡盖起来，知道里面塞满了食物、武器和高档安全门；一旦社会秩序瓦解，缺乏本地防卫支持的外来科技富豪根本守不住地堡，反而第一时间被当地人占领并“开箱拾荒”。 ```mermaid flowchart TD accTitle: 硅谷末日地堡与战利品掉落困境 accDescr: 展示科技富豪独建地堡到社会秩序崩溃后沦为战利品掉落包的逻辑演变 subgraph Preparation["末日准备阶段"] A["恐慌加剧 (AI失控/生物威胁)"] --> B["巨额贷款赴偏远地区买地"] B --> C["修建奢华防核防EMP地堡并屯粮"] end subgraph Reality["崩溃发生与博弈漏洞"] C --> D["缺乏本地社会根基与社区认同"] D --> E["危机爆发：千里奔袭避难"] E --> F["沦为 Loot Drop (战利品掉落包)<br/>被熟知地堡构造的当地居民迅速接管"] end ``` 谈及现实中的失控隐患，主持人打趣表示自己坚决支持强力监管，因为他绝不想在现实生活中陷入必须“正当防卫”对抗机械的绝境： > “想象一下，要是所有的 Waymo 无人驾驶出租车都失控反叛，为了自卫，你和最好的哥们儿不得不整天开着车、端着枪朝窗外到处射杀 Waymo……虽然为了自卫我肯定会开枪，但这实在太折磨、太令人痛苦了。” --- ### [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) Instinct 爆炸式增长：个人 AI Agent 的万亿经济雏形 个人 AI Agent 初创公司 Instinct 的创始人 Noah Shin 打破沉默，做客投资人 Patrick O'Shaughnessy 的播客《Invest Like the Best》，披露了一系列令人震惊的内部数据。Patrick 将其总结为“一系列令人难以置信的炸裂指标”。数据显示，Instinct 目前仍在严格的“仅限邀请码”（Invite-only）内测阶段，但其年化交易额（Transaction Volume）已突破 10 亿美元大关，且正以每天 10% 的惊人速度持续狂飙。令人称奇的是，该公司在市场营销上的支出居然为零美元。 ```mermaid flowchart LR accTitle: Instinct 病毒式传播飞轮 accDescr: 展示 Instinct 凭借翻车截图引发好奇心驱动用户注册与高留存的飞轮效应 subgraph Flywheel["用户获取与转化飞轮"] A["用户在社交媒体晒出截图<br/>(含灾难性搞笑误操作)"] --> B["引发强烈好奇心与尝试欲<br/>(坚信自己不会犯同类错误)"] B --> C["eBay炒卖邀请码<br/>(单码售价高达 $300)"] C --> D["高信任度建立<br/>(40%用户3周内绑信用卡)"] D --> E["交付敏感凭据<br/>(次月留存率高达 80%)"] end ``` 这种爆炸性增长完全由社交网络上的用户截图推动。有趣的是，大量截图内容实际上是 AI 在执行任务时的“翻车惨剧”或灾难性错误，但这不仅没有吓退用户，反而激起了巨大的好奇心。人们看到后心想“这太神了，我只要不犯这种愚蠢错误就行，而且模型一直在变聪明”。在科技圈外，大部分普通人从未亲眼见过 AI 代替自己完整搞定真实任务；当人们第一次看到 AI 跨平台操作现实事务时，感觉就像见证了魔法。 Noah Shin 披露的更多核心数据与运营细节包括： - **极高的信任转化与用户留存**：多达 40% 的用户在注册 Instinct 后的 3 周内直接绑定了个人信用卡；只要用户授权了至少一项敏感隐私信息，其留存率即高达约 80%（主持人调侃：“快把你的至暗秘密交出来”）。 - **极端的算力供给挑战**：算力需求大约每周翻倍。在需要数月才能上线新算力集群的现实约束下，如果保持每日 5% 至 8% 的增长，Instinct 很快就能达到 1 亿用户规模。创始人 Noah Shin 目前约 40% 的精力都花在抢购算力上，因为临时紧急采购算力的溢价高达 3 到 4 倍。通过技术架构优化，后台任务在同等算力下的运行效率分别提升了 3 倍、5 倍甚至 8 倍。 - **模型效能与商业模式**：在 Instinct 自身的 A/B 测试中，其调优表现达到了 Opus 5 级别的性能，而成本仅为其一小部分。商业模式上，Instinct 计划在用户消费交易中按比例抽成（对比 Shopify 的 2.5%–3%、亚马逊的 15%、苹果的 30%），但 Noah 的个人愿景是让 Instinct 基础功能永久向所有人免费。 - **落地场景与消费习惯重塑**：目前 50% 的交易额来自旅游出行预订；已有部分小型企业直接将整个中后台运营完全托付给 Instinct 处理；大量重度用户中有超过 90% 的交互完全通过语音指令完成。 - **重构电商流量分配逻辑**：对于传统 App 而言，通常 70% 的收入依赖广告导流，30% 来自直接销售。AI Agent 的普及可能会彻底抹去这 70% 的广告分成，但 Noah 坚信由于“零摩擦购物”（Zero-effort Buying）的实现，原本 30% 的商品交易盘子会被做大数倍。主持人对此深有共鸣：虽然 Shopify 结算已很方便，但日常生活中购买无聊的琐碎物件（例如给工位挑选几根魔术贴扎带、收纳槽等走线理线配件）根本没人愿意耗费精力去筛选比对，大家只想直接对 AI 发号施令，让其挑一套好评靠谱的直接下单送达。 - **内测狂热度**：产品最初仅向 200 名亲友小范围发布，其内测邀请码随后在 eBay 上被炒卖至每个约 300 美元；目前超过 50% 的使用流量并非来自 iMessage 原生短信；Agent 甚至已经能够主动向创始人 Noah 打电话沟通事务。 ### [18:23](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1103s) Instinct 的惊人增长与估值逻辑 主持人们探讨了 AI 个人代理初创公司 Instinct 的产品体验与惊人增速。创始人 Noah（23 岁，毕业于东北大学）此前给出的用例非常生动：当 AI 代理在处理有严格截止期限的任务时（例如“你必须在接下来的 5 分钟内签署这份文件，你有空吗？”），如果文字沟通没有回应，它会直接打电话给你。融资节奏方面，Instinct 刚以 250 万美元估值拿钱不到一个月，便迅速以 100 亿美元的估值募集了 10 亿美元，投资方包括 Benchmark、Sequoia（红杉资本）等顶级机构。 现场调侃了其商业计划书就像 Reddit 上的经典病毒式段子——“我 3 个月大的儿子体重是出生时的两倍，按这个速度推算，他到 10 岁时将重达 7.5 万亿磅”。但抛开玩笑，Instinct 的实际业务体量确实惊人：在短短约 6 个月内，其平台年化交易流水（GMV）就达到了 10 亿美元。作为对比，Stripe 在经过两年内部测试发布后，花了约 18 个月才达到 10 亿美元的年化交易额。 ```mermaid flowchart LR accTitle: Instinct 与 Stripe 达到 10 亿美元交易规模速度对比 accDescr: 对比 Stripe 与 Instinct 达到 10 亿美元交易规模的时间周期与产品形态差异 subgraph Stripe["Stripe 模式"] S1["2 年 Beta 研发期"] --> S2["公开上线"] S2 --> S3["耗时 18 个月达成 10 亿美元"] end subgraph Instinct["Instinct 模式"] I1["极简团队（14 人）"] --> I2["直接接管个人代理"] I2 --> I3["仅耗时 6 个月达成 10 亿美元"] end ``` 关于其变现路径，两位主持人指出： - **支付手续费潜力**：目前交易仍通过第三方银行卡走流水，Instinct 尚未抽成。但未来完全可以发行自己的专属信用卡，赚取刷卡手续费（Interchange Fee）。用户将日常账务托付给 AI 代理时，使用平台原生卡而非绑定个人主卡也是更自然的安全隔离方案。 - **高净值低成本**：即便扣除底层 Visa 和 Mastercard 的通道成本后只能拿到 0.1% 至 1% 的微薄抽成，在百亿级别的流水下也能迅速转化成数千万美元的纯软件收入。 - **团队极度精简**：整个公司目前仅有 14 人，单位人效极高。潜在风险在于大厂（如 Meta 的 Muse）同类产品的挤压，但也有可能成为亚马逊（Amazon）等缺乏代理抓手的电商巨头的收购标的。 --- ### [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) AI 代理与金融挤兑风险：消除摩擦的代价 阿波罗全球管理公司（Apollo Global Management）的首席经济学家近期发出警告：随着能够自主优化个人资产投资的 AI 代理被大规模采用，可能会在金融系统中引发“银行挤兑（Bank Run）”。美国证监会（SEC）前主席 Gary Gensler 在任期的最后几年也反复敲响同类警钟——“在规模化层面上对算法极致完美的追求，会摧毁系统的稳定性”。当成千上万个聪明的 AI 代理在无须直接通讯（无因果协调）的情况下，同时计算出最优决策并在瞬间执行，可能导致闪崩或流动性瞬间抽干。 这场辩论的核心在于现代经济中有很大一部分利润依赖于“摩擦力（Friction）”。例如，用户往往会遗忘长期未取消的订阅费用，或者任由大笔现金闲置在极低利息的支票账户中。AI 代理的出现则抹平了这些摩擦。不过主持人也提出反方观点：顶级的私人财富管理此前已经在主动替高净值客户进行资金归集（Sweep Account）操作（例如自动把闲钱买入货币市场基金或国债），算法只是将这种精英特权平民化。从技术渗透周期来看，目前使用个人代理的用户仅有数百万，建立用户完全授权 AI 自主转账的信任壁垒通常需要数年时间，银行与货币基金有充分的时间进行资产负债表的动态调整，不至于一夜之间引爆系统性灾难。 - **算法同质化隐患**：数以百万计的个人财务代理根据相同宏观指标在纳秒级别内作出抛售或转移资产的“最优解”，极易引发单边流动性枯竭。 - **经济摩擦消退**：长期沉睡的流失率（Churn Rate）和低息储蓄是很多传统商业模式的基石，AI 的主动干预将打破这一平衡。 - **信任建立周期**：系统演进必须经历两个阶段——第一阶段是“我准备转账，请您确认批准”，第二阶段才是“我已经替你划转完毕，因为你信任我”。后者的普及速度往往比市场恐慌者预想的要缓慢得多。 --- ### [25:09](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1509s) Meta 正式进军企业级 AI，CJ Desai 加盟操盘 节目的赞助商 Railway（智能云平台服务商）环节过后，话题转向了马克·扎克伯格（Mark Zuckerberg）宣布的重磅人事与业务动态：Meta 正式推出“Meta 企业平台（Meta Enterprise Platform）”，作为公司商业版图的全新支柱支柱业务。为了掌舵这项业务，扎克伯格挖来了此前曾做客《TBPN》的前 MongoDB 总裁兼 COO、ServiceNow 关键高管 CJ Desai，出任首席企业平台官（Chief Enterprise Platform Officer），并直接向扎克伯格汇报。 扎克伯格公开表示，该部门将把 Meta 的全栈技术能力——包括 Muse Agent、Meta 商业代理、Muse API 以及 Muse Code 等——向开发者与企业客户全面开放。主持人分析指出，这类企业业务的财报数字往往暗藏玄机：Meta 很可能会效仿过去云计算厂商的做法，先打包几笔巨额的算力租赁或基础设施合作（例如出售原始算力、裸金属服务器、Token 计费或模型推理 API），从而迅速让该部门的账面年收入在短期内从 0 飙升至 100 亿美元。即便外界很难拆解其中有多少是纯软件服务费、多少是算力硬件倒手，但这足以帮助 Meta 在资本市场确立其在企业级 AI 赛道的主导地位。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) 低估了当前的公众话语体系和整体事态到底有多荒谬 — Martin Casado > [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) 快来人从我自己的手里救救我吧！ > [00:00](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=0s) 我玩 AI 玩得很开心 — Mark Zuckerberg > [07:02](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=422s) 大家不如坐下来喝杯啤酒 — Sam Altman > [07:02](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=422s) 让 Claude 一口气生成了一千个研发构想 > [11:25](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=685s) 事情再也不会平静了：塑造 AI 安全恐慌的末日论者 > [11:25](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=685s) 如果你去一个偏僻的地方建起奢华地堡、屯满各种物资，却完全不认识当地社区的人，也没有融入当地的生活网络，那你就是一个‘战利品掉落包’。 > [11:25](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=685s) 想象一下，要是所有的 Waymo 无人驾驶出租车都失控反叛，为了自卫，你和最好的哥们儿不得不整天开着车、端着枪朝窗外到处射杀 Waymo……虽然为了自卫我肯定会开枪，但这实在太折磨、太令人痛苦了。 > [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) 一系列令人难以置信的炸裂指标 — Patrick O'Shaughnessy > [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) 这太神了，我只要不犯这种愚蠢错误就行，而且模型一直在变聪明 > [14:15](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=855s) 快把你的至暗秘密交出来 > [18:23](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1103s) 你必须在接下来的 5 分钟内签署这份文件，你有空吗？ > [18:23](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1103s) 我 3 个月大的儿子体重是出生时的两倍，按这个速度推算，他到 10 岁时将重达 7.5 万亿磅 > [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 在规模化层面上对算法极致完美的追求，会摧毁系统的稳定性 — Gary Gensler > [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 我准备转账，请您确认批准 > [21:32](https://www.youtube.com/watch?v=bDVLnSIbrlU&t=1292s) 我已经替你划转完毕，因为你信任我 ## Mentioned Dario Amodei, Anthropic, 周六夜现场 (Saturday Night Live / SNL), Martin Casado, Andreessen Horowitz (a16z), Oprah Winfrey, Michael Che, Brunello Cucinelli, Mark Zuckerberg, Demis Hassabis, DeepMind, Donald Trump, Andrew Curran, Dana White, Truth Social, Cambridge University, Sam Altman, OpenAI, Claude, FEMA, 使命召唤：现代战争 (Call of Duty: Modern Warfare), 绝地潜兵 (Helldivers), 星河战队 (Starship Troopers), 华尔街日报 (The Wall Street Journal), Tyler Cowen, Constellation, Waymo, Instinct, Noah Shin, Patrick O'Shaughnessy, Invest Like the Best, Shopify, Opus 5, Benchmark, Sequoia Capital, Stripe, Visa, Mastercard, Meta, Muse, Amazon, Apollo Global Management, Gary Gensler, SEC, Railway, MongoDB, ServiceNow, CJ Desai ## Video description Sign up for TBPN’s daily newsletter at TBPN.com Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive 00:00 Dario Amodei Goes Mainstream + AI Doom Hits Pop Culture 04:29 Trump & Dario’s Private Dinner + AI Politics Heat Up 07:02 The Cambridge AI Safety Plan: Measuring RSI + Rogue Agent Prep 11:32 Anthropic Doomers Buy Bunkers + The “Loot Drop” Problem 14:24 Instinct’s Insane Growth + The $1B Personal Agent Economy 21:27 Could AI Agents Trigger a Bank Run? 25:09 Meta Launches Its Enterprise AI Platform [Source](https://www.youtube.com/watch?v=bDVLnSIbrlU)

[0:00] Martin Casado: 低估了当前的公众话语体系和整体事态到底有多荒谬 [0:00] 快来人从我自己的手里救救我吧！ [0:00] Mark Zuckerberg: 我玩 AI 玩得很开心 [7:02] Sam Altman: 大家不如坐下来喝杯啤酒 [7:02] 让 Claude 一口气生成了一千个研发构想 [11:25] 事情再也不会平静了：塑造 AI 安全恐慌的末日论者 [11:25] 如果你去一个偏僻的地方建起奢华地堡、屯满各种物资，却完全不认识当地社区的人，也没有融入当地的生活网络，那你就是一个‘战利品掉落包’。 [11:25] 想象一下，要是所有的 Waymo 无人驾驶出租车都失控反叛，为了自卫，你和最好的哥们儿不得不整天开着车、端着枪朝窗外到处射杀 Waymo……虽然为了自卫我肯定会开枪，但这实在太折磨、太令人痛苦了。 [14:15] Patrick O'Shaughnessy: 一系列令人难以置信的炸裂指标 [14:15] 这太神了，我只要不犯这种愚蠢错误就行，而且模型一直在变聪明 [14:15] 快把你的至暗秘密交出来 [18:23] 你必须在接下来的 5 分钟内签署这份文件，你有空吗？ [18:23] 我 3 个月大的儿子体重是出生时的两倍，按这个速度推算，他到 10 岁时将重达 7.5 万亿磅 [21:32] Gary Gensler: 在规模化层面上对算法极致完美的追求，会摧毁系统的稳定性 [21:32] 我准备转账，请您确认批准 [21:32] 我已经替你划转完毕，因为你信任我

[0:00] Martin Casado: honestly downplayed how ridiculous the rhetoric and broader situation is. [0:00] somebody's got to save me from me. [0:00] Mark Zuckerberg: Yeah, I'm having fun with AI [4:29] Andrew Curran: big chance to patch things up or crash the plane into the mountain. [7:02] Sam Altman: sit down and have some beers [7:02] Tyler Cowen: we got to talk about it [11:25] The Wall Street Journal: Things will never be chill again: The doomers who shape the AI safety freakout. [11:25] Tyler Cowen: if you truly believe existential doom is imminent, why haven't you altered your financial behavior by taking on massive leverage to prepare? [11:25] The prepper community has a term for this: they say don't be a loot drop. You're a loot drop if you do this. If you go and set up a really luxurious bunker with a bunch of supplies, but you don't actually know the community, you're not integrated into the fabric. [11:25] If I was in a self-defense situation I would do it, but it would pain me. [14:14] Just give me one piece... Tell me your darkest secret [18:23] Christian Keil: My three-month-old son is now twice as big as when he was born. He's on track to weigh 7 and 1/2 trillion pounds by age 10.

[0:00] The most viral video I saw over the [0:02] weekend was uh the the parody of Dario [0:05] Amade on SNL that felt like it really [0:07] took over the timeline. Uh Martin Casado [0:10] said honestly downplayed how ridiculous [0:12] the rhetoric and broader situation is. [0:14] It was a funny clip. There were a bunch [0:16] of interesting nuance. I mean, first [0:18] it's like, okay, clearly the the the AI [0:21] rhetoric and the public discourse is [0:23] getting to like a a new height to the [0:25] point where like a lab leader of a [0:26] private company can be can even be [0:28] parodyied on SNL because the audience [0:31] understands who that is. Yeah. Uh so [0:33] it's interesting to see that like okay, [0:35] yes, like uh Daria's done enough on [0:37] Oprah and CNN and and NBC and ABC and [0:41] all the major news networks to be [0:44] relevant to the broader audience. And [0:46] because it only really works like you [0:47] can't like they could never pair. [0:49] There's a lot of people in tech that we [0:50] know that could wouldn't work as uh you [0:53] know SNL punching bags, but he's made [0:55] it. He's made it. [0:56] >> It says a lot that the like the sweater [0:58] vest thing is iconic enough to be like a [1:01] part of a character. [1:02] >> I wonder if he'll stick with that or [1:03] switch it up because it's not it's not [1:05] his leather jacket yet. It's not his [1:07] Hawaiian t-shirt yet, but there is such [1:10] value into leaning into like a [1:11] consistent thing. But I don't know. [1:13] We'll see. The interesting like wrinkle [1:15] here is that Dario is at the point where [1:17] he is enough of a name that they can do [1:21] this bit and and make jokes and it's [1:23] it's relevant to the SNL audience. [1:26] Obviously, it went way more viral on X [1:28] than I I think it probably did on [1:29] YouTube. But when Michael Chay is [1:30] introducing the segment, he sort of [1:33] stumbles over Daario's name a little [1:35] bit. Like he's reading it from the qards [1:37] and he says like and uh this is Daario. [1:39] And then he pauses and it seems like [1:41] maybe Dario Amade is not fully a [1:44] household name in the SNL writers room [1:46] yet in the sense that they were like, [1:48] "Okay, who is this guy? Let's look him [1:49] up. Okay, get him on the Q cards." I [1:50] don't know. Maybe I'm reading into it [1:52] too much, but it felt like this was [1:53] certainly like a crossing the chasm [1:55] moment, like the first time everyone [1:56] gets to know Daario. Uh, but very very [1:59] funny and it really illuminates a little [2:01] bit of like the the complexity of the [2:04] discourse, the irony of saying, you [2:07] know, somebody's got to save me from me [2:09] basically [2:09] >> going going pretty hard on him in some [2:11] ways. That being said, he's always like [2:14] incredibly fascinating to listen to and [2:16] he has incredible clarity of of thought [2:19] and he does at times he'll like pause, [2:23] you know, in in a way that would maybe [2:25] scare normal people a little bit. Um, [2:28] but the thing that I think he can do [2:31] that will have the biggest impact is [2:33] >> I agree. [2:33] >> Uh, no, I actually like the I I like the [2:36] the whatever kind of shawl sweater thing [2:39] that is. [2:39] >> Isn't it Is it Brunella [2:41] >> probably? Yeah, [2:41] >> I think someone I think someone clocked [2:43] it, right? It's Brunello. So, you know, [2:45] it's nice. It's quite a luxury. And I [2:47] feel like a power tie, some shoulder [2:49] padding would go a long way. Anyway, [2:51] >> the biggest thing he can do is to stop [2:53] uh smiling when he's talking about [2:55] really bad devastating things. [2:57] >> Yeah. [2:57] >> Um and so I'm sure that that's something [3:00] >> Yeah. There's something to be learned [3:01] from like uh the way like Fed chairman [3:04] speak or the way uh you know the the [3:07] head of the FDA or the head of the [3:08] military speaks. We've talked to some of [3:10] these folks and like they're very [3:12] austere, very serious, and our show is [3:15] very lighthearted. We kind of have to [3:16] like click into a different mode when [3:18] you're talking to somebody like uh like [3:20] a Michael Katzios or someone who's like [3:22] dealing with the weight of something big [3:25] and and and puts up a wall that is it is [3:29] a it is a performance, but it serves a [3:32] very valuable purpose, which is like I [3:34] take this very seriously. Or you can go [3:36] the opposite way, the the Mark [3:37] Zuckerberg way and rip it and just be [3:38] like, "Yeah, I'm having fun with AI." [3:40] Which is honestly like it feels like [3:42] people are like, "Yeah, I kind of want [3:43] that energy a little bit. Just give me [3:45] the white pill." But I don't know. [3:46] >> Yeah. And again, I I don't think that [3:48] Zach's approach is at all appealing to [3:51] the people actually pushing the [3:52] frontier. No. Right. No. Um, like no, no [3:55] one's going to no one's going to take no [3:58] researchers like take [3:59] >> there are some research [4:00] >> dating dating back to basically Demis [4:02] being like, you know, that moment where [4:05] he realizes Zach is just as excited [4:07] about VR as he is AI, right? Anyways, [4:10] um, [4:11] >> I I I I hope that um [4:14] >> maybe VR will be bigger than AI, Jordy. [4:16] Maybe [4:17] >> one we wouldn't like that. [4:19] >> I would love that. [4:20] >> Wouldn't you like that, John? [4:21] >> I would love that. Uh, but yeah. Anyway, [4:23] so so from SNL headed straight into [4:26] dinner with the president [4:28] >> Sunday. I mean, he wasn't actually at [4:30] SNL. Uh but uh yeah, it's like you wake [4:33] up to that on on Sunday morning. And you [4:36] have to imagine that Trump sees it, but [4:37] Trump has been both he's hosted he's [4:40] hosted SNL multiple times and he's also [4:42] been made fun of on SNL for years. like [4:45] when he gets elected, they basically [4:47] hire an actor to come and be him for [4:50] four years, for the entire 10 year [4:52] because they're going to be doing the [4:52] cold open with him like basically every [4:55] other week. Um, and so maybe there's a [4:58] little bit of camaraderie. They're like, [4:59] "Hey, we've both been through this, [5:01] Dario. It's not so bad. It's actually [5:02] kind of kind of funny." Or or maybe [5:04] they're both like, "Yeah, the news [5:06] really is fake." like we got we got [5:08] >> but yeah as as uh as critical as it is [5:11] at the same time I think it's something [5:12] that all founders should aspire to right [5:15] when you're starting your company and [5:16] investors say well where do where do you [5:18] see yourself you know where do you see [5:19] yourself in the company and [5:20] >> getting skewered getting on the biggest [5:23] stage perhaps perhaps [5:25] >> uh Andrew Curran says President Trump [5:28] has personally invited Dario to a [5:29] private dinner tonight uh that was [5:31] yesterday [5:32] >> so that happened on Sunday night [5:33] >> this will be the first time they've [5:34] actually met oneon-one big chance to [5:36] patch things up or crash the plane into [5:38] the mountain. [5:41] >> Uh they're actually running it back uh [5:42] tomorrow night. Tuesday night there's a [5:44] dinner uh in DC for all the AI leaders [5:47] with the White House again. So I mean [5:49] the pace of this stuff is really really [5:50] really picking up. There's uh there's [5:53] meetings every every week now basically [5:55] and uh yeah I mean there's lots to work [5:57] through and no one is going to get to [5:59] put their their head in the sand. Not [6:01] even the comedians in SNL. will have to [6:03] chime in as well as uh the politicians [6:05] and the lab leaders. Uh and so yeah, [6:08] despite Daario being exclu excluded from [6:11] it seemed like he was excluded from that [6:12] White House dinner with Cinping and uh [6:14] and that basically every top tech leader [6:16] was invited to he was noticeably absent. [6:19] President Trump, you know, followed up [6:21] with a one-on-one dinner with Daario to [6:23] hear his side of the story. And I wonder [6:26] what type of leaks or reporting will [6:28] come out of that. You would think that [6:29] that would be going on today. Maybe [6:31] there will be some news or maybe they'll [6:32] put out a joint statement. Maybe [6:34] everything went really well. But I mean [6:35] both. [6:36] >> Or joint. You could see Daario sign up [6:37] for Truth Social [6:38] >> or UFC courtside or ringside. [6:41] >> Ringside [6:43] be pretty electric. [6:44] >> UFC. [6:44] >> Yeah. [6:45] >> With with the Dawn. I want to see Dana [6:48] White Daario. [6:50] >> Yeah. [6:50] >> Trump. [6:51] >> Yeah. [6:51] >> Just squatted up. [6:53] >> You think he'd be doing slap boxing [6:54] before, too? [6:55] >> Oh, yeah. For for sure. You got to do [6:56] the full suite if you if you go all in. [6:58] Anyway, there's also interesting timing [7:00] that I don't know if this how timed up [7:02] this was, but every day there's another [7:04] proposal for how to manage AI safety. [7:07] Uh, this one feels like the most formal [7:10] concepting exercise from a group of top [7:12] AI researchers. Uh, it's published [7:14] through the University of Cambridge. You [7:16] know, Sam was talking to was it wasn't [7:19] Axios, it was uh he he was interviewed [7:21] and he was saying that like maybe we're [7:23] gonna sit down and get beers basically. [7:25] You remember this whole thing? Why don't [7:27] you guys just sit down and have some [7:28] beers? That was the that was the [7:29] proposal. Uh and he was like, I don't [7:32] want to leak anything that's going to [7:33] happen, but we are thinking about [7:35] thinking about it. And you know, he's [7:37] like, you can't just [7:39] >> personal agents are talking about [7:40] getting together for a beer. [7:42] >> Yeah. Um but it was very funny framing, [7:44] but this is like the start of that, a [7:46] start of more collaboration between the [7:48] labs, between the academic community. [7:50] And uh what's interesting is like they [7:52] go a little bit further. You know Tyler [7:54] over there is his criticism has been [7:56] like everyone comes out with an AI [7:58] safety proposal and it's basically [8:00] always like we got to talk about it and [8:02] this is a little bit more concrete. So [8:05] the first one is uh they want to set in [8:09] place the ability to monitor the level [8:11] of AI enabled R&D at labs. So what [8:16] percentage of the code, what percentage [8:17] of the dollars, what like give us firm [8:20] metrics on how much of the AI R&D is [8:23] being done by artificial intelligence, [8:26] anthropic, open AI, they've all sort of [8:28] like mentioned this in blog posts that [8:30] it's happening, but there's also this [8:32] like vague jockeying because it's better [8:34] to be further along on the RSI curve for [8:37] financial reasons. So there's not really [8:39] an Apple st like no one really feels [8:41] like they have a strong handle on like [8:43] whether there's more RSI going on at [8:46] OpenAI or Anthropic. Like it feels like [8:48] they both have very different metrics. [8:50] Like OpenAI says like we have a intern [8:53] level and then Anthropic will say like [8:55] we had Claude spin up a thousand ideas [8:58] and like you can't really apples to [8:59] apples those. The proposal is like let's [9:01] get some metrics in place so we can [9:03] actually monitor this thing. Uh then [9:05] they also have again it's sort of like [9:08] we need to talk about it but they want [9:10] the government to create firm plans for [9:13] responses to potential theoretical bad [9:17] scenarios. So like if there's a big [9:19] cyber attack and an agent goes rogue and [9:22] brings down the internet and hacks into [9:24] a bunch of stuff like what did what will [9:26] the government do? What's the equivalent [9:27] like FEMA? Like we have a response for [9:29] like there's a big hurricane hits town, [9:31] you bring in people, there's food, [9:33] there's shelter, there's like a plan and [9:35] we don't really have that for like the [9:37] internet went down because like an agent [9:39] took it down. Uh same thing for [9:41] biocurity. Uh they have a couple others [9:43] even like uh economic dislocation that [9:46] type of thing. Uh so they want they want [9:48] the government they're not proposing the [9:51] solutions to those yet but they're [9:53] saying like the go this is where the [9:55] government should be prepared. So, I [9:58] don't know. Uh, it's a good read. [9:59] >> Did you see that, uh, China has places [10:03] you can go to to simulate being in a [10:06] disaster like high s, you know, super [10:09] high winds and and uh flash floods? [10:12] >> Flash floods, things like that. [10:14] >> Should we have uh like AI disaster [10:17] simulations where you can be in a data [10:21] center that's going rogue and kind of [10:23] work to to shut it down with friends? [10:26] Yeah. just a kind of like live action [10:28] roleplay through it. [10:29] >> I like that. I like that. I I'm kind of [10:31] treating Modern Warfare 4 like that. I [10:33] feel like that's getting me enough [10:34] cycles of like the simulation of, you [10:36] know, you're already play against bots [10:38] who are already shooting shooting AI [10:40] basically. So, getting some drills in. [10:42] >> Yeah, I know it's it's a joke, but I I [10:44] wouldn't actually be surprised that a a [10:46] future firsterson shooter or like one of [10:49] these open world games is centered [10:51] around the Terminator scenario. There [10:53] was there was a game there was a game I [10:55] forget what it's called but uh yeah like [10:58] you drop in with three people and you're [10:59] on this like planet. There's one that's [11:01] starship troopers themed so it's like [11:03] the bug planet and then there's another [11:05] planet that you can drop into that's [11:06] like the Terminator planet and you're [11:07] going up against like they basically [11:09] like ripped Terminator's intellectual [11:11] property. Like they look exactly like T. [11:14] >> Why is Tyler not back in school doing [11:17] the best internship in history? He [11:19] turned it down. [11:20] >> I turned it down. He actually turned it [11:21] down just to grind for this. [11:24] >> Yeah. [11:25] >> Things will never be chill again. The [11:27] doomers who shape the AI safety [11:29] freakout. The Wall Street Journal has an [11:30] article here uh about some early [11:32] anthropic employees who are considering [11:34] buying remote land in the United States [11:36] to ro to relocate to if AI goes [11:38] seriously wrong. They're getting [11:40] bunkies. They're going bunker mode. [11:42] Yeah. Employees also trained doomsday [11:44] scenarios and preparation plans in a [11:45] private Slack. While anthropic workers [11:47] have their own office space inside [11:49] Berkeley's Constellation AI safety [11:51] community, the wider safety scene has [11:53] discussed iodine pills, remote islands, [11:55] and electromagnetically shielded desert [11:57] bunkers. [11:57] >> This is like an interesting example of [11:59] like an actual answer to the to the uh [12:01] Teller count question, which is like if [12:03] you're if you really care about Doom, [12:04] why aren't you making financial [12:06] >> decisions? cover up, get a lot of debt, [12:08] and buy a buy a bunker, buy remote land [12:11] with a massive mortgage. [12:12] >> Yeah. Cuz usually the answer to the the [12:13] doomer thing is like, well, I'm not [12:15] going to be able to get to my bunker [12:17] because the, you know, bio terrorism [12:20] will already have killed me or [12:22] something. [12:23] >> This is an interesting [12:24] >> instantiation of this. [12:25] >> The thing that always the thing that [12:27] always just like breaks down with these [12:29] scenarios is like, you know, every every [12:31] every guy has like sort of played out [12:33] the scenario in their head, right? And [12:35] where are you going to be best off? In [12:37] some like remote area where you don't [12:39] know any people that actually live [12:42] there, right? Let's pick a let's pick [12:44] Wyoming, right? [12:46] >> You really think that the world is like [12:48] falling apart and you're just going to [12:50] head over to Wyoming and set up shop in [12:52] a community that knows you've been [12:54] building your bunky there? Yeah. Right. [12:55] And and they're just going to be like, [12:57] "Oh yeah, like there's a bunker over [12:58] there. It has everything you need in it [13:00] to survive." And you know what? But but [13:02] you know, they paid us to set it up [13:04] years ago and and and let's not we [13:06] shouldn't we shouldn't just go occupy [13:08] said bunker for ourselves. [13:10] >> So the prepper community has a term for [13:12] this. They say don't be a loot drop. [13:15] >> It's a loot drop. [13:16] >> You're a loot drop if you do this. If [13:18] you go and set up a really luxurious [13:20] bunker with a bunch of supplies, but you [13:22] don't actually know the community, [13:23] you're not integrated into like everyone [13:25] else. [13:26] >> Yeah. The fabric. So ideally, it's like [13:28] you want your team and everyone else to [13:30] all be in it together and so that you [13:32] know there's no there's no squabbbling, [13:34] but yeah, you can just set up a loot [13:35] drop and then everyone's like, "Oh yeah, [13:37] I installed a crazy safe over there. [13:39] Let's go get that as soon as the the the [13:41] crisis happens." Yeah. I mean, there's a [13:43] whole bunch of fun game theory you can [13:44] play through there. Um, I would I would [13:48] personally I think we need to regulate [13:49] because I would personally I would hate [13:50] to be in a self-defense situation with [13:52] with a bunch of rogue agents. And [13:54] imagine if all the Whimos went rogue and [13:56] we were just forced because of [13:58] self-defense reasons. [13:58] >> You and all your best friends just drive [14:01] around and just have to shoot them out [14:02] of the windows of our cars like all day [14:05] long. Like that would be that'd be a [14:06] nightmare. [14:07] >> That is a nightmare. You would never [14:08] you've never even [14:10] >> if I was in a self-defense situation I [14:12] would do it but it would pain me. [14:14] >> Totally. [14:14] >> Totally. Let's talk about some personal [14:16] agents. [14:17] >> Okay. What? [14:18] >> Noah Shin jumped on Founder of Instinct, [14:21] broke his silence finally with none [14:23] other than our friend and legend Patrick [14:26] >> Oanosy. Patrick pulled out the [14:29] highlights. He said, "A list of [14:30] surprising and mind-boggling stats from [14:32] this conversation. 1 billion a year in [14:34] transaction volume. Still invite only." [14:38] Uh, I can't remember if that's total. I [14:40] think that's actually total volume to [14:43] date. I don't think that's like run rate [14:45] volume. Uh growing 10% a day. [14:47] >> I read that as $1 billion of annualized [14:53] that they are at $1 billion of [14:54] annualized transaction volume. Yeah. But [14:57] I guess it's maybe even higher. I don't [14:59] know. [15:00] >> Yeah. [15:00] >> Yeah. Interesting. [15:01] >> Um growing 10% a day. Spent zero dollars [15:04] on marketing. Uh they have this, you [15:07] know, pretty amazing flywheel where [15:08] people post screenshots uh all the time. [15:11] That's doing a lot of marketing. [15:12] A lot of the but a lot of the [15:13] screenshots are like horrible accidents, [15:16] but maybe those actually drive curiosity [15:18] and like people being like, "Well, I'm [15:20] not going to make that same stupid [15:22] mistake and the service is probably [15:23] getting better." You think those [15:24] actually help? [15:25] >> I think those help. People are just [15:27] curious. [15:28] >> People are curious and they're like, "I [15:29] want to see it. I'll be safe if I won't [15:31] give it my login details, but I'll still [15:33] >> No. And it and it and it can't be said [15:36] enough." [15:36] >> Yeah. [15:37] >> A lot of people within tech have [15:38] experienced AI doing things for them, [15:40] like making a product. Yep. [15:42] >> Most people outside have not experienced [15:45] that yet and it's just a magical thing. [15:47] >> Yeah, totally. [15:48] >> Um 40% of users shared a personal credit [15:51] card with Instinct within 3 weeks. [15:53] Compute demand is doubling roughly every [15:55] week. At 5 to 8% daily growth over the [15:59] months it takes to bring new compute [16:01] online. Instinct would reach about 100 [16:03] million users. On Instinct's own AB [16:05] test, it matches Opus 5 performance at a [16:07] fraction of the cost. Noah spends about [16:09] 40% of his time on compute. Buying [16:12] compute at the last minute costs three [16:13] to fourx more. Makes sense. Background [16:16] work runs on setups that are 3x, 5x, and [16:19] 8x more efficient on the same compute. [16:21] >> Uh 50% of Instinct's transaction volume [16:23] is travel. [16:24] >> That's a crazy. [16:25] >> Instinct plans to take a cut of what [16:27] users buy. Take rates. Uh Shopify is 2 [16:30] and a half to 3%. Amazon is 15%. Apple [16:33] is 30%. Noah's personal goal is to keep [16:35] Instinct free for everyone for life. [16:38] >> Um, some small businesses run their [16:40] entire back office on Instinct. That's [16:43] interesting. Uh, users who connect at [16:45] least one piece of sensitive information [16:47] retain at about 80%. Just give me one [16:50] piece. Just one piece. [16:53] >> Just give me one piece of sensitive. [16:58] >> Um, some [16:58] >> Tell me your darkest secret. [17:00] >> Some users send more than 90% of their [17:01] messages. skeletons you have in your [17:03] >> to instinct by voice. Um, uh, say an app [17:06] makes 70% of its money from ads and 30% [17:08] from actually selling you things. Agents [17:10] could wipe out the 70%. [17:12] >> But Noah thinks that 30% gets much [17:14] bigger because people buy more when [17:16] buying takes zero effort. Uh, this is a [17:19] big question for me, right? Because [17:21] already [17:22] >> buying something on a Shopify store [17:25] takes close to zero effort. Like the [17:27] checkout is actually down to if I land [17:28] on the website and I know what I want. [17:31] Yeah, [17:31] >> I can check out within um and and again, [17:35] Shopify doesn't have an ad business that [17:37] are not fun to shop for though. There's [17:39] a lot of stuff, especially if it's like [17:41] I was thinking about like cable [17:43] management. I wanted to get some cable [17:45] management stuff for my desk and like I [17:47] don't really care about the brand. I [17:49] don't care about the the flow. I just [17:52] want like, you know, a sweet. I needed [17:53] some like command strips, maybe like a [17:56] little tray, like whatever like the [17:57] modern solution is to like a couple [18:00] cables, a power brick, like just go and [18:02] get something that's well reviewed and [18:04] like a suite that will figure that. What [18:06] What kind of voice are you doing? [18:11] Instinct launched to about 200 friends [18:12] and family. Invites have sold on eBay [18:15] for around $300. More than 50% of [18:18] Instinct's traffic doesn't run on [18:20] iMessage. [18:21] Instinct has proactively called Noah [18:23] three times. That's like when when it's [18:25] doing something and it and it has like a [18:26] deadline. It's like Noah gave the [18:28] example of like, "Hey, you really have [18:30] to sign this like within the next 5 [18:31] minutes. Are you available?" It'll just [18:33] call you which is cool. [18:35] >> Um raised one on 10. [18:38] >> Yeah. [18:39] >> Uh yeah, less than a month after raising [18:42] >> two and a half. So Noah Noah is 23. So [18:44] very very good uh summary. [18:47] >> Also same colleges that I went to. He [18:48] went to Nor Eastern. [18:49] >> There you go. This is the whole Instinct [18:51] pitch deck, by the way. And it's [18:53] Christian Kyle saying, "My [18:55] three-month-old son is now twice as big [18:57] as when he was born. He's on track to [18:59] weigh 7 and 12 trillion pounds by age [19:01] 10." Very funny. [19:03] >> Such a viral post. This This is like [19:05] This went super viral on Reddit. It goes [19:07] viral like every couple months on Reddit [19:10] and stuff. It's a It's a wild post. But [19:12] I did want to set some like uh some [19:15] comps around the Instinct growth numbers [19:17] that are interesting. So, a billion [19:19] dollars a year in transaction volume. [19:21] We're not exactly sure if that's [19:22] annualized or or to date, but it's kind [19:26] of the same. Um, for reference, it took [19:28] Stripe around 18 months to hit 1 billion [19:31] of annualized transaction volume, [19:33] although Stripe also spent two years [19:36] building in beta before launch. Instinct [19:39] got to 1 billion transaction volume in [19:40] closer to six months. But, it's [19:42] important to note that with Instinct, [19:44] the transactions flow through thirdparty [19:47] cards. So, they're not making any money [19:48] off of these transactions yet, but it's [19:50] very clear to imagine where monetization [19:53] might happen. Um, [19:55] >> yeah. At the very least, they could just [19:56] launch a credit like their own card. [19:58] >> Yeah, of course. [19:59] >> And then they're getting that [20:01] interchange fee and they don't need to [20:02] get they don't even they don't even need [20:04] to get anyone to opt in. [20:05] >> Yeah. [20:05] >> And they might be able to [20:07] >> And it's an easier pitch, too, because [20:08] you're you're onboarding this AI agent. [20:10] You're like, I've heard some pretty [20:11] crazy stuff about this thing. Like, [20:12] yeah, I'd actually like it if you used [20:15] your card as opposed to like has [20:17] basically a computer, a phone number, [20:19] email. [20:19] >> Yep. Give it a credit card. [20:20] >> Give it a credit card. [20:21] >> And so I'm sure they're working on that. [20:22] And when they do, they're not going to [20:24] make 3% because you do have to pay Visa [20:26] and Mastercard, like the underlying [20:28] rails, but earning 1%, half a percent, [20:31] quarter percent, even even.1% [20:35] on, you know, they grew to a billion, [20:37] you know, you're talking about um [20:39] getting into, you know, tens of millions [20:41] of dollars of revenue pretty quickly. [20:43] And the team's just 14 people. Uh and so [20:45] you can sort of see why they raised a [20:47] billion at 10 billion from benchmark [20:49] sequoia co uh there's you know yes you [20:53] have to believe that that they won't be [20:55] you know killed by muse and that they [20:57] will continue to scale but uh if if you [21:00] know any of the growth rates sustained [21:02] for any even a limited amount of time [21:04] it's going to be a lot of transaction [21:06] volume great business and you can [21:08] underwrite it pretty safely and then [21:10] there's still the there's still the off [21:12] the the offshoot into Amazon world I I [21:15] think because Amazon doesn't really have [21:17] a a strong play there, but uh could fit [21:20] in very well with [21:21] >> Yeah, they have a lot of potential dance [21:23] partners and ultimately things grow [21:25] really fast because they don't just like [21:28] grow really fast for no reason. No, [21:30] right. So, [21:31] >> no good. [21:32] >> Um other people, uh the chief economist [21:34] at Apollo is warning that mass adoption [21:36] of AI agents could trigger a bank run as [21:38] they optimize user investments. Uh Gary [21:41] Gendler spent his last few years as SE [21:44] SEC chair warning that the exact same [21:46] problem will hit the markets. The [21:47] pursuit of algorithmic perfection at [21:49] scale destroys system stability. [21:52] Personal finance agents making [21:54] simultaneous optimal decisions could [21:56] potentially cause a massive flash crash. [21:58] Agent a causal coordination with no [22:00] messages passed between them will show [22:02] up in a lot of places. markets will get [22:04] the big headlines, but the huge number [22:05] of clever advisers making very similar [22:08] choices will impact all of human [22:09] society. Blah blah blah blah blah blah [22:11] blah blah blah. Um, [22:15] uh, human laws were not written for [22:16] this. They were not written for a lot of [22:18] what is coming. [22:19] >> You have to apologize. [22:20] >> It is, uh, is about to start happening [22:22] very fast. Uh, credit to Satrini for [22:25] basically being the first. [22:27] >> That was a different thing, but yeah, [22:28] >> I don't No, it's the exact [22:30] >> It's similar, but [22:31] >> No, it's it's idea of like there's a lot [22:33] of aspects of our economy that are that [22:36] like like businesses that work because [22:37] of friction. [22:38] >> Sure. [22:38] >> That friction in some ways will be [22:40] removed when you have an always on agent [22:42] that says, "Hey, you haven't touched [22:44] this product in [22:47] >> very long time. Do you still want to be [22:48] paying for this?" etc., etc. [22:51] >> I I don't think this is I don't know. My [22:53] my banker will just hit me up and say, [22:55] "Hey, I'm going to I'm just clearing [22:58] this with you. I'm going to sweep funds [23:00] into [23:00] >> money market [23:01] >> money market account or or T bills like [23:04] it's not like like a good banking [23:07] experience is already doing this [23:08] proactively for you. [23:09] >> Yes. [23:10] >> Now there's of course people that maybe [23:12] they just are set up with like a simple [23:15] checking account and they could be [23:17] moving those into a higher yield account [23:20] of some sort that aren't doing it and [23:21] maybe the agent starts doing that. But [23:23] like like fundamentally a great banking [23:25] experience they're like doing this on [23:27] your behalf. when some says, "Dang it, [23:29] if ordinary people get fair deals and [23:31] aren't ripped off, it's the end of [23:34] >> Yeah. Um I Yeah, because I I I I wonder [23:37] like the personal AI agents are taking [23:40] off, but like we're still in like single [23:42] digit millions of people doing them. [23:44] It's going to take time for people to [23:45] adopt them and then they need to [23:47] actually authenticate and add these [23:49] things, link their bank accounts." And [23:51] then presumably the first version of [23:53] this will be the agent coming to you and [23:55] saying like hey is it okay if I do this? [23:58] Second order will be like I just did [23:59] this. I know it's okay you trust me. And [24:02] that relationship that trust I feel like [24:04] that's going to take like years to build [24:06] up and diffuse through. Um there there [24:08] are still like there's still plenty of [24:10] friction in the economy. I think uh I [24:12] think it's going to take a while for [24:13] people to to onboard the onboard to this [24:16] fully. I mean people were it's the same [24:17] it's the same idea of like when Shach [24:19] came out people were like Google search [24:21] is dead like Google no one will Google [24:23] search Google's search revenue is going [24:25] to go to zero like next year and it was [24:28] like yeah it was a reasonable [24:29] replacement but it took a little bit [24:32] longer even though it was growing so [24:33] fast faster than any consumer product [24:34] ever incredible revenues blah blah blah [24:36] blah blah it still wasn't that fast and [24:39] so it's the same thing where it's like [24:41] okay if this happens over like a couple [24:43] of years like banks adjust money market [24:45] funds adjust like it's not the end of [24:47] the world potentially, but we'll keep [24:49] monitoring it. Anyway, let me tell you [24:51] about Railway. Railway is the all-in-one [24:52] intelligent cloud provider. Use your [24:54] favorite agent to deploy web apps, [24:56] servers, databases, and more. Well, [24:57] Railway automatically takes care of [24:59] scaling, monitoring, and security. Zuck, [25:03] >> what happened? [25:04] >> We believe super intelligence will [25:05] create significant new opportunities for [25:07] all people and businesses. Meta already [25:09] serves billions of people at scale and [25:11] helps hundreds of millions of businesses [25:13] reach customers. Today we are starting [25:15] the next major pillar of our business, [25:17] Meta Enterprise Platform to help [25:19] businesses use AI to grow and transform [25:21] in new ways as well. Meta Enterprise [25:24] will use our strengths that few other [25:25] companies have advanced models, leading [25:28] agents, large scale infrastructure, and [25:29] years of working closely with many [25:31] businesses. Initially, we will focus on [25:33] bringing our full technology stack, [25:35] including the Muse Agent, Meta Business [25:37] Agent, Muse API, Muse Code, and more to [25:39] businesses and developers to help them [25:41] grow. to lead this effort. I'm excited [25:44] that uh Shiron CJ Desai, former guest of [25:48] the show, will join Meta as chief [25:49] enterprise platform officer reporting [25:52] directly to DOC. CJ is an experienced [25:54] enterprise leader with a track record of [25:55] building full stack software and [25:57] delivering results in AI, infra, [25:58] business applications, and security. Uh [26:00] blah blah blah blah blah blah blah blah. [26:04] Um anyways, uh so here's how I think [26:07] this is going to go. I think that they [26:08] created this new business unit. Mhm. [26:10] >> They still have a lot to prove on if [26:12] they can [26:13] >> Sorry, if you scroll down, there's some [26:14] wild comments. [26:16] >> Um [26:16] >> anyway, continue. [26:18] >> Uh anyways, they created this new [26:19] business unit. They're talking about [26:21] apps and agents and all this stuff. I [26:23] think what's really going to happen is [26:24] they're going to do big compute deal. [26:26] >> Mhm. [26:27] >> And then and then it's going to look [26:28] like the this sort of unit is doing it. [26:31] It'll quickly go from let's say 0 to10 [26:33] billion a year of revenue. They can sort [26:36] of use that. They don't have to [26:37] necessarily break out where that revenue [26:39] is coming from, but they can say like [26:40] our enterprise platform is is gone from [26:43] zero to 10 billion in [26:45] >> Yeah, this was like the joke semi [26:47] analysis was was sort of trolling with [26:48] around like oh like all the growth in [26:51] Google Cloud Platform is selling TPUs. [26:54] It was like a little bit more nuanced [26:55] than that. Right. Uh but there's a bunch [26:57] of ways to like wrap up revenue into a [27:01] uh into like an enterprise platform that [27:04] could be tokens, could be doing [27:06] inference, could be selling compute raw [27:09] bare metal deals, could be could be [27:11] actually selling services. Uh we'll be [27:14] interesting to see where it goes. Leave [27:15] us five stars. Apple podcast, Spotify, [27:18] sign up for newsletter tbn.com. We'll [27:19] see you tomorrow for big event. Goodbye. [27:22] >> Can't wait.