SaaS Stocks Rebound 150% as Enterprise Networks Resist AI Threat

企业网络抵御AI威胁，SaaS股票反弹150% 单体软件平台经受住了自建代码替代方案的冲击，因为深厚的分发渠道、运营商关系和嵌入式网络构成了AI无法取代的结构性护城河。 主持人 John Coogan 和 Jordi Hays 开篇庆祝了 SaaS 龙头企业的市场反弹，指出拥有深厚分发护城河的核心软件平台已挺过最初的 AI 抛售恐慌。他们分析了《华尔街日报》的一项重大调查，该调查揭露了朝鲜 IT 人员如何利用实时 AI 和位于美国的笔记本电脑托管服务渗透美国远程科技岗位，从而为其政府的国家运作提供资金。此外，主持人还探讨了备受关注的 AI 实验室高管离职潮、一起涉及超豪华酒店的网络热议纷争，以及据报道 Anthropic 拟以 60 亿美元收购世界模型实验室 Decart 的谈判。 主持人宣布“SaaS末日论”破灭，指出Palo Alto Networks和CrowdStrike等大型企业软件公司股价强势反弹，其分销网络与极低相对成本构筑了坚固的非代码护城河。 (0:01) 虽然底层大模型重创了Chegg和Canva等单点工具，但Twilio凭借运营商关系和基础设施信任卷土重来，市值升至380亿美元并成为自主AI Agent不可或缺的底层支持。 (5:00) 《华尔街日报》调查揭露数千名朝鲜IT工人利用被盗美国身份、AI实时答题工具及美国本土主机中介，成功渗透远程技术岗位，2024年为朝鲜政权输送了约8亿美元资金。 (15:00) Anthropic据传正就以约60亿美元估值收购实时世界视频模型实验室Decart展开谈判，凸显顶尖AI实验室对环境模拟与世界模型战略价值的重视。 (22:48)

TBPN hosts John Coogan and Jordi Hays argue that market panic over generative AI destroying enterprise software was overblown, pointing to resilient distribution moats among SaaS incumbents even as AI technologies enable unprecedented corporate security threats like North Korea's covert remote workforce. Enterprise Moats Endure While Basic SaaS Collapses (0:01) Coogan and Hays revisit their prediction canceling the "SaaSpocalypse," arguing that the initial market panic wiping out $2 trillion in market capitalization was fundamentally flawed. Software heavyweights have since staged massive recoveries: Palo Alto Networks rose 121% over the past year to a $320 billion valuation—aided by CEO Nikesh Arora purchasing $10 million in personal stock during the selloff—while CrowdStrike jumped 107% to $230 billion. The hosts emphasize that incumbents rely on non-code moats such as enterprise distribution and low relative costs. For instance, an enterprise generating $100 million in revenue pays around $1,000 monthly for Shopify Plus—roughly one-tenth of 1% of revenue—rendering internal attempts to build replacements unviable. Conversely, products competing with baseline large language model capabilities have suffered severe disruptions. Chegg became a prime victim of AI substitution, with its stock declining 99% over five years and revenue falling 39% from $617 million in 2024 to $376 million in 2025 as students adopted free chatbots. Canva is similarly pressured by prompt-to-image generators. In contrast, communications platform Twilio surged 150% over the year to $38 billion, as the hosts argue that autonomous AI agents will inevitably rely on established infrastructure, carrier relations, and spam compliance rather than rebuilding telecom stacks. North Korean Operatives Infiltrate US Remote Payrolls A Wall Street Journal investigation revealed that thousands of North Korean IT workers have secured remote software engineering positions within American companies to generate revenue for Kim Jong-un's regime. According to the US Treasury Department, this operation generated roughly $800 million in 2024 alone, with the North Korean government garnishing up to 90% of workers' pay. Top individual operatives earn up to $300,000 annually by working multiple remote roles simultaneously under stolen American identities. The scheme relies on domestic US accomplices who host corporate laptops in their residences, enabling overseas operatives based in China and Russia to remotely access systems and evade domestic geolocation controls. In one documented arrangement, a US facilitator split a $75,000 salary evenly while performing no technical work. To bypass interview screens at scale, operatives employ real-time AI voice agents and language model scrapers that generate technical answers during live interviews, allowing a single tracked cell to apply to over 1,000 US companies within three months. Big Tech Compensation Cliffs and Strategic Acquisitions Executive turnover at Meta Super Intelligence (MSL) underscores the retention challenges facing leading AI research labs. Multimodal lead Jahuu departed Meta after only one year overseeing initiatives such as Muse Spark, Voice Mode, Muse Image, and Muse Video to launch a new venture named TBD Lab. Jordi Hays notes that after initial nine- to ten-figure compensation packages reach their one-year 25% vesting mark, researchers secure over $100 million in liquid net worth, incentivizing exits toward independent startups. At the same time, consolidation is intensifying around next-generation architectures. Anthropic is reportedly in negotiations to acquire real-time world video model lab Decart for approximately $6 billion, up from a $4 billion valuation the previous quarter. Industry observers regard world models and environment simulations as critical strategic assets and stepping stones toward artificial general intelligence, prompting frontier labs to acquire emerging leaders. Branding Missteps, Influencer Entitlement, and AI Fatigue The hosts critique recent consumer and luxury design developments, starting with Instagram's first wordmark update in a decade. While framed by CEO Adam Mosseri as cleaner, the redesign faced public skepticism for pursuing minimalism just as broader design trends pivot back toward distinctive heritage aesthetics. Meanwhile, a bespoke Porsche 911 collaboration between Singer and Louis Vuitton featuring handbag-leather exterior accents was widely panned online as tacky "AI slop IRL." Tensions around content creators also surfaced in an incident where ultra-luxury hotel brand Aman turned away YouTube reviewer Ryan Walker from a $4,663 single-night booking. Although Walker claimed his reservation was canceled without warning, reporting confirmed the hotel sent notice the previous morning, and observers supported Aman's enforcement of guest privacy rules against influencer filming. The hosts conclude by noting growing public "AI fatigue," where relentless tech discourse faces growing consumer apathy even as capital expenditure and competition remain intense. While core enterprise software has proven far more resilient against AI obsolescence than panic-driven markets anticipated, the technology is rapidly transforming corporate security risks and engineering mobility, leaving companies to grapple with covert payroll infiltration and retention challenges as compensation packages mature.

节目主持人 John Coogan 与 Jordi Hays 认为，关于“代码生成将摧毁传统软件业”的恐慌被严重夸大，具备深厚分发网络与基础设施壁垒的 SaaS 企业正在强劲反弹；与此同时，AI 技术的扩散不仅加速了大模型实验室的人才流动与巨额并购，还被朝鲜技术人员用于建立伪装渗透美国企业的远程黑工网络。 SaaS护城河重估与基础工具分化 (0:01) 主持人指出，此前由于市场担忧“氛围编码”（vibe coding）能让企业自行低成本构建定制软件，导致 SaaS 板块蒸发了高达 2 万亿美元市值，但近期的复苏证明了非代码护城河的稳固性。Palo Alto Networks（市值达 3200 亿美元）与 CrowdStrike（市值达 2300 亿美元）过去一年股价分别上涨 121% 与 107%，其 CEO 甚至在恐慌期斥资 1000 万美元增持并获得 1600 万美元账面收益。以 Shopify 为例，年营收近 1 亿美元的企业每月仅需向 Shopify Plus 支付约 1000 美元（仅占营收约千分之一），任何理性创始人都不会耗费工程资源去自研替代系统。 然而，缺乏底层壁垒的轻量级工具确实遭到了基础大模型的毁灭性打击。Chegg 因 ChatGPT 和 Gemini 免费提供作业辅导，过去五年股价下跌 99%，年收入从 2024 年的 6.17 亿美元下滑 39% 至 2025 年的 3.76 亿美元，市值缩水至 8000 万至 9000 万美元；Canva 也受到 Grok 等直接生成完整信息图表工具的挤压。相反，拥有运营商网络关系、反垃圾邮件过滤体系及多年信誉的 Twilio 在低谷后飙升 150% 至 380 亿美元市值，成为自主 AI 智能体与现实电信网络通信必不可少的“智能体基础设施”。 朝鲜黑工利用AI与中介远程渗透美企 《华尔街日报》发布的纪录片《潜伏：朝鲜的美国秘密劳工》揭露，朝鲜 IT 劳工正系统性渗透美国主流企业的远程软件开发岗位，为平壤当局聚敛巨额资金。根据 FBI 的估计，有数千名朝鲜 IT 人员正活跃在这些远程职位上。调查显示，一个受监控的朝鲜工作小组在三个月内向超过 1000 家美国公司投递了申请；在远程技术面试中，他们借助实时 AI 语音代理和大模型工具，在面试官提出专业问题时即时生成精准答案供候选人朗读。 该黑色产业链高度依赖美国本土的“协助者”。这些协助者在自己家中架设和代管企业配发的笔记本电脑，使身在中国和俄罗斯等非引渡国家的朝鲜人员能够远程接入，借此伪造美国本土的网络流量。纪录片中一名协助者仅凭代管设备便与朝鲜劳工平分 75000 美元年薪。黑工们还利用被盗的美国公民身份绕过背景调查，一人同时兼职多份全职工作。顶级朝鲜 IT 劳工年收入可达 30 万美元，而美国财政部估计朝鲜政府抽走了其高达 90% 的收入，仅在 2024 年一年就获取了近 8 亿美元资金。 顶尖AI实验室人才流失与世界模型并购 在前沿 AI 实验室动态方面，Meta 超级智能实验室（MSL）多模态负责人 Jahuu 在入职仅一年后便宣布离职，与核心成员 Mark 和 Alex 联合创办“TBD Lab”。Jahuu 此前在 MSL 主导了 Muse Spark、Voice Mode、Muse Image 及 Muse Video 等核心模型。主持人分析指出，这反映了头部科技企业面临的薪酬与人才困境：在经历高额的九位数签约争夺战后，研究人员在工作第一年兑现 25% 股权后便可获得超过 1 亿美元的流动资产，促使大量顶尖学者选择离开大公司官僚体系重新创业。 与此同时，前沿大模型的竞争正迅速向环境模拟与物理世界模型拓展。据报道，Anthropic 正就以约 60 亿美元估值收购实时世界视频模型实验室 Decart 展开谈判，而 Decart 在上一个季度的估值仅为 40 亿美元。行业观察人士指出，这一潜在交易表明顶尖实验室正在将世界模型视为与推理优化同等关键的战略资产；李飞飞与 Oliver Cameron 等研究者也认为，实时环境模拟与世界模型是通往通用人工智能（AGI）的核心基石。 品牌重塑争议、隐私冲突与AI叙事疲态 节目还探讨了消费文化与舆论环境的演变。Instagram 时隔十年推出的首个极简文字标识更新被批评为略显滞后，因为大众审美趋势正从极简主义转向复古与繁复设计；Singer 与 Louis Vuitton 联名改装的保时捷 911 则因过度堆砌手袋皮革元素，被网络舆论批评为“现实版的 AI 废料”。此外，超奢酒店安缦（Aman）取消并驱逐自费评测博主 Ryan Walker 价值 4663 美元预订的事件引发了两极分化的争论：尽管社交媒体受众声讨酒店，但商业读者普遍支持安缦维护住客隐私的举措，认为高端客户支付高额溢价的初衷正是规避镜头打扰。 针对当前科技圈对 AI 的狂热讨论，节目援引 Bucco Capital 的观点指出，过度高频地谈论 AI 正变得令人厌倦，技术终将像早期互联网一样融入日常生活背景。主持人认同 AI 最终会演变为常规基础设施，但强调在当前巨大的经济利益与激烈竞争下，行业的关注焦点短期内依然难以转移。 无论是 SaaS 企业依赖传统分发渠道抵御技术冲击，还是黑工网络与大模型实验室对人才和资本的追逐，均表明人工智能正在从最初的概念颠覆期迈入复杂的现实博弈阶段；尽管公众对泛滥的 AI 营销逐渐产生审美疲劳，但在地缘安全、资本运作及底层基建层面的争夺仍未出现放缓迹象。

## Background 该视频由科技脱口秀频道 TBPN 于 2026 年 8 月 20 日发布在 YouTube 平台，由 John Coogan 和 Jordi Hays 共同主持。根据简介，本期为 30 分钟的精简版节目《Diet TBPN》，该节目被《纽约时报》誉为“硅谷的新宠”，致力于快速提炼与解读当日科技行业最受关注的重磅新闻与核心趋势。 ## TL;DR 节目首先宣告“SaaS 末日”终结并展开“胜利巡游”，深入剖析了 SaaS 板块在蒸发 2 万亿美元后凭借结构性护城河实现的分化与强势反弹，对比了安全巨头的暴涨与 Chegg 的溃败。随后，主播聚焦 Meta 超级智能实验室（MSL）在开出天价薪酬一年后的核心人才“兑现离职潮”，并详细拆解了《华尔街日报》关于朝鲜 IT 幽灵劳工利用 AI 工具与美国本地跳板渗透远程工位的惊人调查。最后，节目还探讨了奢华酒店安缦与网红的纠纷公关战、Anthropic 传闻以 60 亿美元收购世界模型 Decart 的行业影响，以及对“谈论 AI 已然过时”论调的深度反思。 ## Atmosphere & style 整期节目散发着典型的硅谷科技局内人闲聊氛围，语调轻快幽默、节奏密集且充满辛辣讽刺。两位主持人配合默契、插科打诨，在严谨剖析万亿美元级资本博弈与跨国黑产内幕的同时，穿插着诸如“男孩子的零花钱数学”与对“现实版 AI 垃圾”豪车的毒舌吐槽。整体听感既有对宏观科技趋势洞若观火的自豪与锐度，又带着像在硅谷茶水间听硬核八卦般的畅快与松弛感。 ## Bulletin - [00:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1s) SaaS 末日论宣告破灭：全行业曾因“氛围编程”恐慌蒸发超 2 万亿美元市值，如今拥有销售渠道、客户低占比成本与不可替代分发网络的公司强劲反弹，网络安全巨头股价大涨逾 100%，而单纯依赖代码的单点工具如 Chegg 市值暴跌 99%。 - [15:19](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=919s) 华尔街日报披露朝鲜 IT 幽灵劳工渗透美国：数千名朝鲜技术人员通过 AI 语音实时提词面试作弊、盗用身份并雇佣美国中介托管硬件跳板，2024 年为平壤政权输送近 8 亿美元外汇。 - [08:32](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=512s) Meta 超级智能实验室（MSL）面临兑现离职潮：顶薪挖角的核心研究员在满一年兑现首笔巨额股权（约 25% 悬崖期）后纷纷出走创立新公司，高额报酬机制演变为创业者的第一桶金。 - [22:48](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1368s) Anthropic 传拟以 60 亿美元估值收购世界模型初创 Decart：估值较上一季度跃升 50%，凸显前沿大模型实验室正将物理世界模型与极速推理优化作为通往 AGI 的核心基础设施。 ## Key takeaways - [00:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1s) “SaaS 末日”恐慌证伪：市场曾认定大模型一键生成应用会让所有软件估值归零，但 Shopify、Spotify、Roblox 等企业依托分发网络与生态构筑了极高迁移成本，代码行数并非唯一壁垒。 - [04:22](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=262s) 网络安全板块迎来狂欢：Palo Alto 过去一年上涨 121% 且 CEO 亲自千万美元抄底，CrowdStrike 上涨 107%，证明底层刚需防御能力不受生成式 AI 冲击。 - [04:22](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=262s) 标准化内容工具遭遇毁灭性打击：Chegg 五年市值蒸发 99%，2025 年营收骤降近四成，验证了缺乏深层护城河的知识搬运类 SaaS 极易被基础大模型全面替代。 - [04:22](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=262s) Twilio 逆袭并重塑为“Agent 基础设施”：依靠跨全球电信运营商协议、路由与反垃圾合规网络，Twilio 股价反弹 150% 至 380 亿美元，成为 AI 智能体连接物理世界的底层通道。 - [08:32](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=512s) 顶级 AI 科学家的“球星化”与天价薪酬困境：科技巨头为顶尖研究员开出九位数天价薪酬包，但在一年悬崖期兑现后，财富自由促使顶尖人才加速流失并自立门户。 - [12:37](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=757s) 品牌视觉趋势的分水岭：Instagram 更换使用十年的字标因过于扁平极简遭到审美反弹，而 Singer 与 LV 联名款保时捷因堆砌符号被讽刺为“现实版 AI 垃圾”。 - [15:19](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=919s) 黑客与 AI 工具的黑产合流：朝鲜技术团队在技术面试中实时使用 AI 语音助手提词生成代码，三个月内向千余家美企批量渗透求职。 - [15:19](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=919s) 远程办公漏洞滋生本土中介黑产：美国本土协从人员代收企业笔记本作为硬件跳板并参与五五分成，朝鲜政权强制扣缴劳工高达 90% 的工资作为外汇收入。 - [20:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1201s) 安缦取消网红预订引发两极公关战：事件折射出奢华酒店维护客户私密性与自媒体博主依托冲突流量变现之间的商业矛盾，不同平台受众评价彻底割裂。 - [22:48](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1368s) 世界物理模型成为 AGI 必争之地：Anthropic 洽购 Decart 反映出顶级实验室将实时视频生成与推理优化视为与大语言模型并列的核心战略支柱。 - [25:42](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1542s) 技术扩散与万亿资本角逐：尽管有观点认为过度高谈阔论 AI 正在变得“令人尴尬”，但底层基础设施重构与全球产业主导权争夺正处于资金规模最大的攻坚期，远未到翻篇时刻。 ## Detailed walkthrough ### [00:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1s) 宣告“SaaS末日”终结：从2万亿美元恐慌到理性回归 节目一开场，John Coogan 与 Jordi Hays 便兴奋地表示要来一次“胜利巡游”（victory lap）。Jordi 提到自己前一天翻看各大 SaaS 公司的股价走势图，发现大量标的涨势惊人，于是发短信问 John：“我们是什么时候宣布‘SaaS 末日’（SaaSpocalypse）取消来着？”他们翻出了今年 2 月在 Substack 上发表的专栏文章，早在当时市场极度恐慌时，他们就断定所谓的技术崩塌“不会发生”。尽管当时许多人认为这是言之过早的自大判断，彼时整个软件行业弥漫着末日情绪，大批传统 SaaS 公司被粗暴地打上“不过就是一堆代码”的标签，市场认定在 AI 时代任何人都可以通过“氛围编程”（vibe coding）轻易复制一切。 回看这场风波，SaaS 行业的遭遇极其惨烈——在最严重的抛售潮中，整个科技软件板块足足蒸发了 2 万亿美元的市值。当时的狂热悲观逻辑认为，既然所有公司都能在一夜之间借助大语言模型（LLM）“氛围编程”出一套自己的专属 CRM，那么任何基于庞大代码库构建的公司估值都理应归零。虽然从长期来看，软件代码作为单体护城河的价值确实大打折扣，点对点解决方案（point solutions）也面临越来越激烈的竞争，但现实远比这种简化论复杂得多。大量被错杀的 SaaS 巨头展现出了根本无法用“代码行数”来衡量的深层实力，市场在恐慌中显然把“婴儿和洗澡水一起倒掉了”。 这些被忽视的隐形护城河，包括数以千计的业务拓展代表（BDR）构建的销售网络，以及极低的企业收入占比成本优势。以 Shopify 为例，很多年营收接近 1 亿美元的电商巨头，其每月在 Shopify Plus 上的支出大约只有 1,000 美元左右，仅占其总成本的极小比例。在商业现实中，即便今天的 AI 模型再强大，你也必须雇用多名工程师日以继夜地进行“氛围编程”，才可能维持一套勉强可用的同类电商基础设施。半年前，TBPN 圈定了六家核心标的作为这场 SaaS 辩论的案例研究：Google、Meta、Spotify、Shopify、Roblox 和 Salesforce。事实证明，DoorDash、Spotify、Roblox 等平台的核心壁垒在于不可替代的网络效应与分发管道——用户未来确实可能收听 AI 生成的音乐，但大概率依然是在 Spotify 上收听；玩家或许能一键生成游戏，但依然离不开 Roblox 的基础设施与庞大玩家生态。 ```mermaid flowchart TD accTitle: SaaS企业在AI冲击下的护城河分化 accDescr: 对比单纯依赖软件代码的脆弱企业与具备结构性护城河的抗跌企业 subgraph Vulnerable ["高脆弱性：纯代码与知识搬运"] A["单点工具 / 纯代码库"] --> B["被基础大模型直接替代\n（如作业搜题、简单修图）"] end subgraph Resilient ["强抗跌性：结构性隐形护城河"] C["销售与渠道网络\n（成千上万 BDR 销售大军）"] D["网络效应与分发平台\n（如 Spotify、Roblox 玩家生态）"] E["极高性价比与重构成本\n（如 Shopify 仅占客户支出极小比例）"] F["物理网络协议与运营商绑定\n（如 Twilio 通信牌照与防垃圾机制）"] end ``` - **SaaS 行业蒸发市值**：在“SaaS 末日”抛售潮的高峰期，全行业软件公司累计损失超 2 万亿美元市值。 - **Shopify 典型案例**：年营收近 1 亿美元的企业，使用 Shopify Plus 的月服务费仅约 1,000 美元，企业绝无动力自研替代。 - **早期反转观察名单**：TBPN 于 2 月份锁定的六大标志性抗跌公司为 Google、Meta、Spotify、Shopify、Roblox 和 Salesforce。 --- ### [04:22](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=262s) SaaS的胜者与真正受害者：网络护城河 vs 基础LLM降维打击 随着恐慌消退，市场上涌现出一批势不可挡的 SaaS 赢家，其中网络安全领域表现尤为强劲。网络攻击的泛滥让安全防线变得前所未有的关键：Palo Alto Networks 过去一年股价暴涨 121%，市值飙升至 3,200 亿美元；CrowdStrike 也在过去 12 个月内大涨 107%，市值达到 2,300 亿美元。两人特别调侃了 Palo Alto 的首席执行官 Nikesh Arora——在 TBPN 宣布“SaaS 末日终结”的一个月后，这位 CEO 亲自抄底，掏出 1,000 万美元自有资金买入自家股票，如今该笔投资已暴赚至 2,600 万美元，公司股价更是创下 315 美元的历史新高。Jordi 笑着打趣：“下次见到他得调侃一下，怎么当时只买了 1,000 万？如果你真的笃定，至少应该加 10 倍杠杆啊！”这笔净赚的 1,600 万美元被两人戏称为高管的“男孩子的零花钱数学”（boy math）。 然而，这场 AI 革命并非皆大欢喜，确实存在一批被基础大模型直接“降维打击”而彻底丧失生机的牺牲品，其中最具代表性的便是 Chegg。Chegg 的商业模式在新冠疫情期间盛极一时，但在 ChatGPT、Gemini 等大模型能够免费提供作业解答、逐步演算辅导后，其核心价值瞬间瓦解。Chegg 的股价在过去五年内暴跌了 99%，市值缩水至仅剩 8,000 万到 9,000 万美元左右；其 2025 年营收仅为 3.76 亿美元，较 2024 年的 6.17 亿美元同比骤降了 39%，陷入了只能靠不断裁员维持微薄利润的收缩螺旋。此外，Canva 近期也因图像生成技术（如 ChatGPT 图像生成、Nano Banana Pro、Grok 新的图表生成能力）能够“一键式”（one-shot）直出高水准海报与信息图表，而被媒体报道正面临类似的被替代危机。 与此形成鲜明对比的是底层工具类企业 Twilio 的强势逆袭。Twilio 过去一年股价上涨了 150%，市值已恢复至 380 亿美元。很多人未曾意识到它的高价值，误以为它还停留在过去五六十亿美元的估值低谷。两人分析指出，Twilio 绝不仅仅是一段可以被“氛围编程”的代码，它背后是数十年积累的全球电信运营商关系网络、路由通道以及极其繁琐的垃圾短信拦截合规协议。在 AI Agent 大行其道的今天，与其将自己定位成传统 SaaS，不如直接顺应潮流自称“Agent 基础设施”（agentic infrastructure）——当 AI Agent 需要跨网络发送短信与物理世界交互时，Twilio 就是它们必须从货架上取下的现成基础设施。 - **网络安全板块涨幅**：Palo Alto Networks 过去 12 个月上涨 121%（市值 3,200 亿美元）；CrowdStrike 上涨 107%（市值 2,300 亿美元）。 - **高管抄底神操作**：Palo Alto CEO Nikesh Arora 自掏腰包 1,000 万美元抄底，短短数月浮盈至 2,600 万美元。 - **Chegg 灾难性业绩**：股价五年暴跌 99%，当前市值仅 8,000 万至 9,000 万美元；2025 年营收从 2024 年的 6.17 亿美元萎缩 39% 至 3.76 亿美元。 - **Twilio 强劲反弹**：市值回升至 380 亿美元，过去一年大涨 150%，成为 AI 智能体调用的通信基础设施底层。 --- ### [08:32](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=512s) Meta MSL人才流失潮：顶薪挖角一年后的“兑现离职” 节目随后把话题切向了人工智能实验室的核心人员流动，重点聚焦在 Meta 超级智能实验室（MSL，Meta Superintelligence Lab）的高管离职潮。John 与 Jordi 翻出了一年前引爆整个行业关注的“人员交易卡”（traded card）——大约一年前的八月，顶尖研究员从 OpenAI 轰轰烈烈地跳槽至 Meta 的 MSL，被视为 Meta 在全行业大举抢夺顶尖 AI 智囊的标志性事件。然而仅仅一年之后，该研究员便在社交媒体上正式宣布离职创业。 在离职声明中，该研究员写道：“我将离开 Meta 创办一家新公司。在 Mark 和 Alex 身边参与建设 TBD 实验室的经历令人深感鼓舞与充实。我为我们多模态团队在 Muse Spark、Voice Mode（语音模式）、Muse Image 和 Muse Video 上取得的成就感到无比自豪，更为成就这一切的团队感到骄傲。随着时间推移，我越来越被一个对人类未来至关重要、却在很大程度上尚未得到充分探索的问题所吸引。” 面对这一高调告别，Jordi 犀利地指出了 Meta 等大厂在招募顶级人才上面临的系统性困境：MSL 实际上是在与时间赛跑。大约一年前，Meta 不惜开出九位数乃至十位数的惊人天价薪酬包，从各大竞争对手手中大肆掠夺顶级研究员。然而，这类天价薪酬机制通常包含极高的首年兑现比例，当这些专家在一年后顺利跨过“悬崖期”（cliff），拿到相当于总额 25% 的巨额套现——“银行账户里凭空多出 1 亿美元”时，很多人会瞬间进入“我彻底财富自由了，该去干点真正想做的事”的状态。对于这些身价过亿的顶级科学家而言，哪怕只拿走那座“巨大金山”的四分之一，也足以支撑他们毫无后顾之忧地辞职去启动自己的全新探险。 ### [10:10](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=610s) MSL 人才离职潮与 TBPN 首张走红全网的“球星卡” 在讨论 Meta 旗下超级智能实验室（MSL）的人才流动时，主持人们指出，许多顶尖研究员在拿到巨额薪酬或股权套现后，看着银行账户里的庞大数字，往往会重新审视人生目标，思考自己究竟快不快乐、眼下的工作是否有成就感，或者是否该自立门户去创立一家公司。针对刚刚离职的高管 Jia-Bin Huang（黄嘉斌），大家调侃无论他拿到了多丰厚的离职大礼包，显然都已经做好了冒险的准备。黄嘉斌在离职声明中表示自己“越来越被一个对人类未来至关重要的难题所吸引”，主持人们打趣猜想，说不定他是想创办一个全新的社交网络来挑战老东家，或者终于参透了如何打造“安全、对齐的社交媒体”。他们预测，在接下来的一到两个月内，MSL 还会出现更多核心技术人员出走创业的案例。 话题随后转向了 TBPN 幕后制作人 Tyler 制作的第一张球星交易卡（Trading Card）。这张卡片最初只是 Tyler 随手拼出来的创意，背景甚至直接用了一块棒球场，上面甚至连 TBPN 自己的品牌 Logo 都没有加，却意外成为席卷全网的爆款。原始推文在 X 平台上斩获了超过 3 万个点赞，被第三方无水印账号搬运到 Instagram 后更是突破了 10 万点赞，彻底打破了圈层壁垒。 这波出圈风暴不仅让《纽约时报》的每日播客《The Daily》在一两期节目中剪辑并引用了 TBPN 对扎克伯格“亲自给员工做汤”等幕后八卦的深度爆料，甚至还引来了法国电视台派摄像组专程赴美采访。主持人们回忆起当时接受法国媒体采访的滑稽场面：对方记者反复追问“这人到底赚了多少钱”，而他们只能坦诚相言不仅不清楚具体数字，即便知道也不能透露，但留下一句大实话——在硅谷顶级 AI 人才争夺战中，“薪水根本没有上限”（There are no salary caps）。 - **全网破圈传播**：Tyler 制作的球员卡风格图在 X 平台获得 30,000+ 点赞，在 Instagram 搬运帖上斩获约 100,000 点赞。 - **主流媒体引用**：节目关于 Meta 挖角与扎克伯格留人内幕的讨论被《纽约时报》播客《The Daily》收录，并接受了法国国家级电视团队的入场专访。 --- ### [12:37](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=757s) Instagram 十年来首换字体 Logo 与 LV 联名保时捷的“现实版 AI 垃圾”争议 Instagram 负责人 Adam Mosseri 刚刚宣布更换应用顶部的品牌字标（Wordmark），这是该平台整整十年来首次对该标识进行调整。官方给出的理由是希望带来更简洁、现代的视觉风格，同时向经典工艺致敬。然而，互联网大众对这次改版并不买账。主持人们分析指出，虽然左侧使用了十年的老版字体确实略显陈旧，但它极具辨识度与标志性。随着如今品牌设计界正逐渐从五年前巴黎世家（Balenciaga）等奢侈品牌掀起的“极简无衬线风”回调向“极繁主义”（Maximalism），Instagram 此时跟进扁平极简反而显得有些落后于时代潮流。 紧接着，两人将视线转向了一款在社交媒体上引发众怒的奢华联名——保时捷改装厂 Singer 与路易威登（Louis Vuitton）合作打造的联名款保时捷 911。这款车被大量汽车爱好者和网友痛批为“史上最俗气的改装车”。车身采用了类似 LV 经典手袋的配色，引擎盖上生硬地绑着皮质手袋绑带，车轮甚至选用了突兀的亮蓝色，内部内饰则铺满了大面积皮革印花。 主持人们边看实车视频边无情吐槽，直言这台车简直就是“现实中的 AI 废料”（AI slop IRL），看起来就像山寨厂商未经授权搞出的劣质联名。社交网络上的评论区更是骂声一片，充斥着“穷奢极欲的下一阶段”（next level gluttony）以及“机械增压级工业垃圾”（aspirated slop）等尖锐批评。尽管视频中展示的同款头盔和部分挡把细节尚存设计感，但整体堆砌的土豪审美让绝大多数车迷完全无法接受。 --- ### [15:19](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=919s) 华尔街日报重磅调查：朝鲜 IT 幽灵劳工如何利用 AI 与美国跳板渗透远程工位 《华尔街日报》耗时一年多推出了一部长达 30 分钟的深度调查纪录片《渗透：朝鲜的秘密美国劳工大军》（*Infiltrated: North Korea’s Secret US Workforce*），揭露了朝鲜政权如何建立起一套庞大而精密的跨境替身网络，借助 AI、被盗用的美国公民身份以及美国本土协从人员，让其远程技术人员混入美国公司的研发团队，从而将数以亿计的美元外汇输送回平壤。联邦调查局（FBI）证实，目前有数千名朝鲜 IT 人员正活跃在美国各行各业的求职市场中。记者团队获取了一个朝鲜劳工行动小组被泄露的浏览器历史记录、电子邮件、日程表以及屏幕录像，全面还原了这一黑产链条的惊人内幕。 ```mermaid flowchart TD accTitle: 朝鲜远程 IT 渗透流程图 accDescr: 描述朝鲜工程师如何通过 AI、假身份及美国本土中介渗透美国企业并抽水薪酬的全过程。 subgraph 招聘欺诈 A["批量投递求职信<br/>(单组3个月投递1000+公司)"] --> B["技术面试作弊<br/>(AI语音助手实时提词)"] B --> C["盗用美国公民身份<br/>(一人分饰多角通过背调)"] end subgraph 硬件跳板 C --> D["企业邮寄办公笔记本电脑"] D --> E["美国本土中介(Facilitators)<br/>代收代管设备运行"] end subgraph 远程操控与回款 E --> F["朝鲜工程师自非引渡国远程登录<br/>(中俄等国接入美国家庭网络)"] F --> G["薪资到账并分账<br/>(50/50 分成或高额抽佣)"] G --> H["平壤政权抽缴高达90%外汇<br/>(2024年总计吸金近8亿美元)"] end ``` 根据调查，该行动小组在短短三个月内就利用 AI 技术向超过 1,000 家美国企业疯狂投递简历。在技术面试环节，他们会在后台常驻 AI 语音助手实时监听考官问题，并以极快的速度生成代码范例或技术解答，让求职者照本宣科轻松通关。更有甚者直接雇佣美国本地人充当“白手套”（Frontmen）：本土傀儡只需接收雇主邮寄来的办公笔记本电脑，插上网线充当跳板，完全不需要写一行代码；而藏匿在中俄等非引渡国家的朝鲜工程师则通过远程桌面接入这些主机，营造出员工在美国本土办公的虚假 IP 轨迹。 主持人们深入剖析了这一骗局为何在新冠疫情与远程办公普及后野蛮生长。涉案的美国本土“协助者”（Facilitators）以被动收入为诱饵参与犯罪，例如一份年薪 75,000 美元的软件工程岗位，美方协助者与朝鲜黑客各分一半薪水。美国财政部估计，部分朝鲜技术人员年收入可达 30 万美元，而朝鲜政府会强制扣押其海外 IT 劳工多达 90% 的工资收入。仅在 2024 年一年，这一跨国影子外包网络就为平壤政权带来了近 8 亿美元的资金进账，规模之庞大令人不寒而栗。 - **申请规模**：单个受监控的朝鲜行动小组在 3 个月内利用自动化与 AI 工具申请了 1,000 多家公司。 - **面试作弊手法**：技术面试全程挂载 AI Voice Agent 实时检索提词，甚至由美国本地人代面后交接。 - **利益分成模式**：美国本土中介协助代管企业笔记本，单份 7.5 万美元年薪工作双方五五分账（50/50）。 - **政权抽水规模**：美国财政部测算，朝鲜劳工收入的 90% 被政府直接没收，2024 年仅此渠道就创收近 8 亿美元。 --- ### [19:07](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1147s) 奢华酒店安缦单方取消网红订单：4,663 美元引发的社媒争端 节目的最后部分提及了《华尔街日报》报道的另一起热点争议事件：顶级奢华酒店品牌安缦（Aman）单方面取消了一名 YouTube 奢华酒店评测博主 Ryan Walker 价值 4,663 美元的客房预订，随后双方在网络上爆发了激烈的公关骂战。 这起事件迅速演变为探讨商业实体与自媒体网红之间日益恶化的紧张关系的典型案例。尽管安缦新店开业被誉为全年度最受瞩目的奢华酒店盛事，但主持人们在翻阅了详细报道后戏谑表示，虽然大家都喜欢看豪门翻车，但在这场具体风波中，无论是从契约精神还是言行举止来看，各路证据似乎都对这位试图以此制造话题的博主非常不利。 ### [20:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1201s) 奢华酒店博主与安缦的冲突：《华尔街日报》与 YouTube 评论区的两极分化 网络奢华酒店评测博主沃克（Walker）因入住安缦酒店（Aman）受阻而引发轩然大波。Walker 一直以来标榜自己的商业模式是纯粹的“独立与透明”——他自费支付昂贵的房费，不接受酒店赞助或公关免单，类似于知名汽车评测人道格·德穆罗（Doug DeMuro）的评测策略。此前，他曾自掏 26,000 美元体验四季酒店（Four Seasons）游艇行程并公开给出严厉差评。而在最近的一次安缦酒店开业首周体验中，酒店前台却告知他并没有预订记录，事件甚至升级到酒店报警将他护送离开。 随着事件发酵，双方各执一词的细节逐渐浮出水面。Walker 在离开时声称自己漏看了一封邮件，该邮件在前一天发送，因酒店开业首周接待能力受限而提前取消了他的订单并致歉。然而，Walker 一开始向观众声称邮件是在前一天深夜才送达，直到随后接受《华尔街日报》（The Wall Street Journal）求证时，才改口承认邮件其实在前一天上午就已经收到，但他并未在后续的视频或直播中修正这一时间线。安缦发言人则表示不对具体预订置评，但指出 Walker 绕过了主入口，明显有故意找茬炒作的嫌疑。 两派舆论在不同平台上呈现出鲜明的对立景象： - **YouTube 评论区**：涌入了 5,000 多条倾向于 Walker 的留言，大量自称是安缦忠实客户和旅游业内人士的网友对酒店的处理方式表示愤慨，纷纷宣称“再也不会预订安缦的任何物业”。 - **《华尔街日报》评论区**：高赞读者则对酒店的做法拍手称快，留下“安缦，我这就来”、“任何封禁网红博主的酒店都是在给普通客人提供最棒的服务”等评价，批评者直言 Walker 是个“完全在演戏的自大狂（complete tool）”，认为奢华度假的核心价值本就是远离四处可见的镜头与打扰。 ### [22:48](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1368s) 传 Anthropic 拟以 60 亿美元收购世界模型独角兽 Decart 根据最新市场传闻，前沿 AI 巨头 Anthropic 正在深入接洽，计划以高达 60 亿美元的估值收购专注于世界视频模型与推理优化的 AI 实验室 Decart。这一估值相较于 Decart 仅仅一个季度前的 40 亿美元估值有了大幅跃升。节目主播回忆起与 Decart 创始人迪恩（Dean）的多次交流，赞赏其对自家技术的极端自信——Dean 甚至在没有任何事先彩排或录播测试的情况下，直接在直播访谈中实时演示模型推理和视频生成效果，展现出令人惊叹的工程水准。 行业分析师认为，若该笔交易达成，Decart 将成为首家被前沿基础大模型实验室收购的世界视频模型公司。这标志着各大前沿实验室正将物理世界模型（World Models）与极致的推理优化技术，视为与大语言模型并驾齐驱的核心战略支柱。尽管目前这类技术更多停留在令人惊艳的原型演示阶段，尚未迎来类似生成式动画或“吉卜力时刻”的破圈杀手级应用，但结合此前与奥利弗·卡梅隆（Oliver Cameron）以及李飞飞（Fei-Fei Li）等学界与工业界领袖的探讨，世界模型已被公认为是通往通用人工智能（AGI）不可或缺的关键拼图。 ```mermaid flowchart TD accTitle: Anthropic收购Decart战略路径 accDescr: 展示Decart的核心技术如何融入Anthropic的AGI战略版图 subgraph DecartCore["Decart 核心技术壁垒"] A["世界物理模型<br/>(World Models)"] B["极速推理优化<br/>(Inference Optimization)"] end subgraph FrontierIntegration["前沿模型能力整合"] C["视频生成与多模态交互"] D["B2B 与企业级代码智能"] end subgraph StrategicGoal["终极战略目标"] E["AGI 全球计算基础设施"] end A --> C B --> D C --> E D --> E ``` ### [24:24](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1464s) 社交符号学的演变：从经典“笑哭”到“泪流成河”的流行代际 节目随后转入一段轻松有趣的社交观察，探讨了当代人使用表情符号（Emoji）的微妙习惯演变。两位主播讨论了当前主流的三种情绪表达符号：传统的正向笑哭脸（😂）、倾斜侧躺的笑哭脸（🤣），以及泪流成河的大哭脸（😭）。知名财经博主乔·魏森塔尔（Joe Weisenthal）最近在社交平台上公开放弃了传统笑哭，转向倾斜的“笑得打滚”表情，引发了关于 Emoji 代沟的讨论。 主播分享了自己的演变轨迹：泪水直喷的大哭符号（😭）已经成为 2025 年全球范围内最流行的多功能表达，既能传达强烈的幽默感，又带有自嘲与无语的夸张情绪。许多人在经历了长达十多年只使用微笑或拒绝任何流泪符号的时期后，如今也开始频繁混用 iMessage 的“哈哈”点赞与各种流泪 Emoji。大家甚至打趣讨论起是否该在团队工作群中尝试冷门且充满反差感的“猫咪笑哭”表情，看看里面是否藏着某种不可替代的“Alpha 表达红利”。 ### [25:42](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1542s) “谈论 AI 已经过时了？”：技术漫化与万亿资本角逐的终局反思 节目的尾声引用了投资机构 Buco Capital 发布的一段辛辣博文：“我觉得我们已经非常接近这样一个临界点：继续过分关心或整天高谈阔论 AI 已经变得有点令人尴尬了。翻篇吧，到底谁在乎呢？”这句犀利的吐槽引发了关于科技扩散（Diffusion）深层次规律的探讨。 主播分析认为，这种论调在技术发展史上并不罕见。正如今天的互联网一样，大众早已不再高谈阔论“互联网”这个底层技术本身，而是自然地讨论在网络上发生的具体事件、产品与公司——当一项革命性技术真正全面渗透到社会肌理中时，底层的概念就会“隐形化”。然而，现在断言可以“翻篇”还为时过早，因为眼下正处于全球有史以来资金规模最大的产业军备竞赛之中，不仅牵扯到底层基础架构的重构，更关系到未来全球经济主导权的重新分配。在万亿美元资本与国力竞争的决战时刻，这远非一句随意的调侃所能轻易终结。 ## Notable quotes > [00:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1s) 我们是什么时候宣布‘SaaS 末日’取消来着？ — Jordi Hays > [04:22](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=262s) 下次见到他得调侃一下，怎么当时只买了 1,000 万？如果你真的笃定，至少应该加 10 倍杠杆啊！ — Jordi Hays > [08:32](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=512s) 我将离开 Meta 创办一家新公司。在 Mark 和 Alex 身边参与建设 TBD 实验室的经历令人深感鼓舞与充实。我为我们多模态团队在 Muse Spark、Voice Mode、Muse Image 和 Muse Video 上取得的成就感到无比自豪，更为成就这一切的团队感到骄傲。随着时间推移，我越来越被一个对人类未来至关重要、却在很大程度上尚未得到充分探索的问题所吸引。 — 黄嘉斌（Jia-Bin Huang） > [10:10](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=610s) 薪水根本没有上限 — John Coogan & Jordi Hays > [12:37](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=757s) 现实中的 AI 废料 — John Coogan & Jordi Hays > [12:37](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=757s) 穷奢极欲的下一阶段 — 网络评论 > [12:37](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=757s) 机械增压级工业垃圾 — 网络评论 > [20:01](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1201s) 任何封禁网红博主的酒店都是在给普通客人提供最棒的服务 — 《华尔街日报》读者评论 > [25:42](https://www.youtube.com/watch?v=aZN8Jy0akYg&t=1542s) 我觉得我们已经非常接近这样一个临界点：继续过分关心或整天高谈阔论 AI 已经变得有点令人尴尬了。翻篇吧，到底谁在乎呢？ — Buco Capital ## Mentioned John Coogan, Jordi Hays, Nikesh Arora, Jia-Bin Huang（黄嘉斌）, Tyler, Adam Mosseri, Ryan Walker, Doug DeMuro, Dean, Oliver Cameron, Fei-Fei Li（李飞飞）, Joe Weisenthal, Mark Zuckerberg, TBPN, Shopify, Google, Meta, Spotify, Roblox, Salesforce, DoorDash, Palo Alto Networks, CrowdStrike, Chegg, OpenAI, Canva, Twilio, Meta 超级智能实验室（MSL）, FBI, 《华尔街日报》（The Wall Street Journal）, Singer, Louis Vuitton（路易威登）, 保时捷 911, 安缦酒店（Aman）, 四季酒店（Four Seasons）, Anthropic, Decart, Buco Capital, 《渗透：朝鲜的秘密美国劳工大军》（Infiltrated: North Korea’s Secret US Workforce）, The Daily, ChatGPT, Gemini, Muse Spark, Voice Mode, Muse Image, Muse Video ## 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 [Source](https://www.youtube.com/watch?v=aZN8Jy0akYg)

[0:01] Jordi Hays: 我们是什么时候宣布‘SaaS 末日’取消来着？ [4:22] Jordi Hays: 下次见到他得调侃一下，怎么当时只买了 1,000 万？如果你真的笃定，至少应该加 10 倍杠杆啊！ [8:32] 黄嘉斌（Jia-Bin Huang）: 我将离开 Meta 创办一家新公司。在 Mark 和 Alex 身边参与建设 TBD 实验室的经历令人深感鼓舞与充实。我为我们多模态团队在 Muse Spark、Voice Mode、Muse Image 和 Muse Video 上取得的成就感到无比自豪，更为成就这一切的团队感到骄傲。随着时间推移，我越来越被一个对人类未来至关重要、却在很大程度上尚未得到充分探索的问题所吸引。 [10:10] John Coogan & Jordi Hays: 薪水根本没有上限 [12:37] John Coogan & Jordi Hays: 现实中的 AI 废料 [12:37] 网络评论: 穷奢极欲的下一阶段 [12:37] 网络评论: 机械增压级工业垃圾 [20:01] 《华尔街日报》读者评论: 任何封禁网红博主的酒店都是在给普通客人提供最棒的服务 [25:42] Buco Capital: 我觉得我们已经非常接近这样一个临界点：继续过分关心或整天高谈阔论 AI 已经变得有点令人尴尬了。翻篇吧，到底谁在乎呢？

[1:00] vibe code [5:00] John Coogan: agentic infrastructure [8:00] Jordi Hays: operating up against a clock [12:00] Adam Mosseri: cleaner and more modern [12:00] John Coogan: AI slop IRL [12:00] John Coogan: tacky unauthorized-looking collab [12:00] aspirated slop [12:00] next-level gluttony [16:46] Infiltrated: North Korea's Secret US Workforce [24:24] Any hotel that bans influencers is doing regular guests a great service. [22:48] Ghibli moment [24:24] John Coogan: LARPing as a cat [24:24] Bucco Capital: caring about AI or talking about it a lot is a bit embarrassing [24:24] Bucco Capital: I think we're very close to the point where caring about AI or talking about it a lot is a bit embarrassing. Move on already. Who cares?

[0:01] We got to take a victory lap. You want [0:03] to start taking a lap, Jordy? You want [0:04] to take a a lap while I tell everyone [0:06] about our SAS apocalypse victory lap? [0:07] >> I was just appreciating some various [0:10] software as a service companies [0:13] >> charts yesterday. Bunch of companies up [0:16] like crazy [0:17] >> and I was thinking I texted John I was [0:19] like when did we cancel the SAS [0:21] apocalypse? [0:21] >> Yep. [0:22] >> You pulled up our original Substack that [0:25] we sent back in February. [0:27] >> We just decided at that time [0:30] It's not happening. [0:31] >> It's not happening. Uh we canled it [0:34] >> and uh there was a lot of [0:36] >> maybe too early to take a victory lap, [0:38] but in hindsight, but there was there [0:42] were some good arguments. There were [0:43] some interesting arguments and there was [0:45] a lot of fear, but there were a lot of [0:46] companies that were getting thrown in [0:48] the SAS bucket as just like a pure pile [0:51] of code. You could vibe code it. Uh and [0:54] while that thesis might play out over a [0:56] few years, it's it was a little bit too [0:58] soon. it seemed a little bit too [0:59] aggressive and so uh we wanted to [1:01] revisit the SAS apocalypse and the [1:03] cancellation of the SA SAS apocalypse [1:05] see where things are now. So, just to [1:06] set the stage, the SAS apocalypse was a [1:09] rough rough go. $2 trillion of market [1:12] cap lost across uh the the the SAS [1:15] apocalypse, the major selloff of [1:17] technology, software companies broadly. [1:20] Uh two trillion dollars wiped out, gone. [1:23] Moment of silence, moment of silence. [1:25] But a lot of it's come back. It appeared [1:26] only logical at the time that every [1:29] company would be vibe coding their own [1:31] CRM uh and this would happen imminently [1:34] and that any company built on a big pile [1:37] of code would go to zero. Of course, the [1:39] core thesis still holds over the long [1:42] term, but it's a lot messier in reality. [1:44] So, yes, having a huge monolithic piece [1:48] of software is less of a moat today than [1:50] it was a decade ago. That's for sure. [1:52] Competition is increasing, especially [1:54] for point solutions. But many of those [1:57] SAS companies that were so beaten up in [1:59] the SAS apocalypse were revealed to have [2:03] sources of strength that didn't fit [2:05] neatly into the lots of lines of code [2:07] written bucket. So babies were [2:10] thrown,000 [2:12] business development representatives. [2:14] >> Yeah. [2:15] >> Source of strength. [2:16] >> That's big. also just uh what it another [2:19] thing that's very valuable is what [2:22] percent of revenue are you claiming from [2:25] your customers. So if you are going to a [2:28] customer and you're saying I'm taking [2:30] 30% cut, you're probably at more risk [2:33] than someone who's saying I'm an IT [2:35] solution and you're going to spend [2:38] onetenth of 1% of your [2:39] >> Yeah. Shopify is the best example, [2:41] right? [2:42] Yeah. A lot of a lot of e-commerce [2:44] entrepreneurs ask them like what's your [2:47] biggest expense? [2:48] >> Yeah. [2:49] >> None of them will say Shopify. I think [2:51] even a brand that is like a Dale [2:53] >> Shopify pro uh Shopify Plus uh for a [2:57] business that I know very intimately uh [2:59] I think is around $1,000 a month in cost [3:03] and the business is doing almost 100 [3:04] million a year and even even with [3:08] >> even with how good the models are today [3:10] you would need multiple people basically [3:12] vibe coding around the clock. [3:14] >> Yeah. [3:15] >> To have a product that was comparable. [3:17] >> Yeah. A lot of babies were thrown out [3:18] with the bath water. Uh the the the most [3:21] ridiculous one was I think Door Dash, [3:23] but there were lots of people coming for [3:25] Spotify and a whole bunch of different [3:27] platforms that should be very enduring [3:30] because their source of strength is a [3:31] network effect or something like that. [3:33] So six months ago, we we identified six [3:36] companies that we wanted to use as case [3:38] studies for the SAS apocalypse. It was [3:40] Google, Meta, your favorite company, uh [3:43] Spotify, Shopify, Roblox, and Salesforce [3:46] as evidence that largecale software. [3:48] have a minute to talk about [3:50] >> meta. [3:51] >> I mean, business is doing great and it [3:54] always should have been [3:55] >> Don't get me started. [3:56] >> It never should have been Don't get me [3:57] started. [3:58] >> It never should have been beat up. Uh [4:00] same with same with Google. Uh and then [4:02] Spotify and Shopify were the interesting [4:04] ones. Spotify, of course. Uh there is a [4:06] world where you're listening to AI [4:08] music, but there's also a world where [4:09] you're listening to AI music on Spotify. [4:12] Same thing for Roblox. Yes, you'll be [4:14] able to vibe code uh a game, but having [4:16] all the Roblox network infrastructure [4:18] distribution that that should be [4:20] valuable in the in the future. You know, [4:22] now you have a long list of unstoppable [4:24] SAS companies, companies that can't just [4:26] be immediately spun up and replaced by [4:28] an AI app. Uh there are AI winners now [4:31] and it's a lot of who you'd expect. Uh [4:33] cyber security is more important than [4:35] ever. You mentioned Crowdstrike, but [4:36] Palo Alto Networks is also on a tear. Uh [4:38] Palo Alto Networks over the past year is [4:41] up 121%. It's a $320 billion company. [4:45] Crowd Strikes a $230 billion company up [4:48] 107% this year over the past 12 months. [4:52] Palo Alto CEO bought the dip five months [4:54] later. He must have been reading us [4:56] because a month after we canceled the [4:57] SAS apocalypse, he was like, I I think I [5:00] like this Palo Alto Network stock. [5:02] >> He of course is the CEO of Palo Alto [5:04] Networks. So, but he put $10 million of [5:07] his own money into the company. It's now [5:09] worth $26 million. Nikesh Aurora [5:12] invested as investors question whether [5:14] AI could disrupt cyber security. [5:16] PaloAlto has reached an all-time high. [5:18] >> 315. [5:18] >> Got to give him uh some trouble next [5:20] time we see him. Why only 10? [5:22] >> Oh, 10. [5:24] >> Does that really move the needle? [5:25] >> Yeah, he should have been 10x levered. [5:27] >> Yeah, [5:27] >> if you really believed. Come on. But, [5:29] you know, he has 16 million of basically [5:32] play money now. I mean, it's boy math. [5:34] you know, you make you make 16 million [5:35] trading your own stock when you're the [5:37] CEO. You got to spend that on something [5:39] fun. There are SAS apocalypse victims [5:42] that have not come back, are probably [5:44] not coming back, need to completely [5:46] reinvent the business because they have [5:48] been uh made obsolete by just base level [5:52] LLM capabilities. CHEG is the canonical [5:55] example of SAS apocalypse victim that [5:56] has not made a comeback. Uh the stock is [5:58] down 99% [6:01] over the past five years. It was hot [6:03] during COVID. Uh, and of course, when it [6:06] comes to looking up answers to homework, [6:09] uh, the basic free edition of Chat GPT [6:12] gets you there. Gemini, whatever you [6:14] want to use is going to answer those [6:16] questions and hold your hand alongside [6:19] your homework while you're doing it. [6:21] >> So, they had 376 million of revenue in [6:24] 2025, but that was a 39% decrease [6:28] year-over-year from 617 in 2024. Yeah. [6:32] >> Uh it's now roughly an 80 uh or $90 [6:35] million market cap. [6:36] >> Yeah. So very shrinking business, very [6:39] difficult. You have to continue cutting [6:41] every year to make any money, pull any [6:44] profit out of that business. The newer [6:45] company that is more uh in the headlines [6:48] these days is the information is [6:49] reporting that Canva is slipping into a [6:51] similar situation because a lot of Canva [6:53] designs can be oneshot by image models [6:56] like Chhat GPT images, Nano Banana Pro, [6:59] uh Grock, imagine I saw the new Gro [7:01] image uh examples and and a few of them [7:04] were infographics. And then there's also [7:06] tools companies like Twilio is doing [7:08] incredibly well. I don't know if you've [7:10] been tracking this. It's up 150% over [7:12] the past year. Another story where huge [7:15] boom during coow [7:17] >> selloff [7:18] >> $ 38 billion company [7:20] >> y [7:20] >> I if you asked me what Twilio was worth [7:24] uh before I just checked this I would [7:25] have said I don't know five billion six [7:27] billion something like that [7:28] >> and and and that's where it was a couple [7:30] years ago. Um, but it's done really, [7:32] really well. And I think a lot of that [7:34] is that it is difficult to go and vibe [7:37] code all of the interactions that you [7:39] need to actually send text messages [7:40] across the network. [7:42] >> It's a gentic infrastructure, John. If [7:44] you historically were a SAS company, [7:46] pivot to just being calling yourself a [7:49] gentic infrastructure. [7:51] >> It's it's more like Yeah. I mean, it's [7:53] funny. It's it's more like it's it's [7:55] infrastructure that will be pulled off [7:58] the shelf by the agent. And so uh I mean [8:01] we use Twilio for our app where we want [8:04] to be able to interface with an [8:06] application via text message and we use [8:09] Twilio for that even though we could go [8:11] and vibe code that all of a sudden [8:12] you're dealing with the different mobile [8:15] carriers and the cell networks and are [8:17] they going to flag you as spam and [8:19] Twilio has all the the huge decades of [8:21] relationships built out in a real [8:23] network there that val that even though [8:25] it's just a tool and it's just [8:27] consumption software at the end of the [8:29] It has this moat and so it's been doing [8:31] really really well. [8:32] >> We have to talk about [8:35] >> Yahoo. [8:37] >> Oh yeah. Traded. [8:38] >> Uh he is uh this was uh took us back in [8:43] time as well. Let's try to pull up the [8:46] original traded card that kicked it all [8:48] off. August of last year. [8:51] >> I sent it to [8:52] >> uh Yahoo. You uh went from OpenAI to [8:57] MSL. uh was one of the first [8:59] high-profile [9:01] uh uh exits in that whole saga. One year [9:04] later, he just announced this morning, [9:06] he said, "I'm leaving Meta to start a [9:08] new company, building the TBD lab [9:10] alongside Mark." And Alex has been [9:12] deeply inspiring and fulfilling. I'm [9:14] proud of what our multimodal team [9:16] accomplish across Muse Spark, Voice [9:18] Mode, Muse Image, and Muse Video, and [9:20] even prouder of the team that made it [9:22] possible. Over time, I've felt [9:24] >> increasingly drawn to a problem that [9:26] will matter deeply to humanity's future, [9:29] yet remains largely underexplored. [9:32] >> I forget what day it was, maybe it was [9:33] Monday or Tuesday when um I was uh [9:37] feeling a little spicy, but I was just [9:39] saying that uh MSL is is basically [9:43] operating up against a clock, which is [9:45] that they did this massive talent rate [9:48] across all these companies uh about a [9:51] year ago. Yeah. [9:52] >> And when you're paying these people nine [9:54] figures, 10 figures, uh many of those [9:58] people are going to basically spend a [10:00] year and hit a point where they're like, [10:01] "Okay, I have $100 million in the bank [10:05] and I'm kind of good now." [10:07] >> 25% of a huge pile of gold is still a [10:10] huge pile of gold sometimes. [10:12] >> Exactly. [10:12] >> And a lot of those people are just going [10:14] to basically see the number in their [10:16] bank account and be like, "Am I happy [10:17] doing what I'm doing? Is this [10:18] fulfilling? Or do I want to go build a [10:21] company?" maybe they've always wanted to [10:22] build a company and so [10:24] >> in the case of Jahu [10:26] um [10:27] >> whatever package he had clearly he's [10:30] down to go take a risk [10:32] >> I don't think this will be the last of [10:34] the exits that we see out of MSL over [10:37] the next even one or two months [10:39] >> what if he starts a social network that [10:41] would be risky what if he's like I'm [10:43] coming for it all [10:44] >> I mean he says I've felt increasingly [10:46] drawn to a problem that will matter [10:48] deeply to humanity's future so maybe [10:51] He's figured out how to make aligned [10:52] social media. Social media safety. [10:54] >> I like that. Yeah. Uh well, we should [10:56] let Tyler pat himself on the back for [10:59] this card. [10:59] >> Reels that make you go like this. [11:03] >> This card was the first TBPN trading [11:06] card and Tyler whipped it up uh by [11:09] himself and it doesn't even have our [11:12] brand on it and it also has a literal [11:15] baseball field in the background. But [11:18] this video or this this image went so [11:21] viral. The original post got 30,000 [11:24] likes on Axe was [11:26] >> it did around 100k on Instagram. Random [11:29] account that again wasn't ours. [11:30] >> It was crazy. It actually broke [11:32] containment [11:33] >> for not watermarking it. And it was the [11:35] f it was the first it was the first [11:37] moment where we had been talking about a [11:40] story that really broke through to the [11:42] mainstream because of this trading card. [11:44] We wound up covering [11:46] >> on French television. [11:47] >> French television. That's where I was [11:48] going. That's where I was going. So, uh, [11:50] but because of this trading card, we [11:52] talked about this a whole a this story [11:55] over a course of weeks. It was very [11:57] interesting and dramatic. There were a [11:58] whole bunch of scoops that came out [12:00] around uh Mark Zuckerberg making people [12:02] soup and stuff. It was it was a lot of [12:04] really entertaining stuff. We were [12:05] featured in the New York Times daily [12:07] podcast, I think once or twice for [12:09] talking about it. They clipped us and [12:10] included our coverage in their telling [12:13] of the story for their broader audience. [12:15] And then French television sent out a [12:17] bunch of reporters and cameras to come [12:18] and interview us, which was very funny [12:20] because they kept asking us like exactly [12:22] how much money does does did this guy [12:24] make and we're like look, we don't know [12:26] exactly how much he makes and even if we [12:27] did, we probably wouldn't want to share [12:29] that. But uh it was a very fair. I did [12:31] tell them there's there's no salary [12:33] caps. [12:34] >> That's true. There are no salary caps. [12:37] >> The head of Instagram, Adam, just posted [12:39] a new word mark after 10 years [12:42] >> of leaving it unchanged. Adam says the [12:45] word mark at the top of the app hasn't [12:47] changed in 10 years. So, it was time for [12:50] a refresh. Cleaner and more modern with [12:53] references to the original [12:55] >> and the simplicity and craft that's [12:57] always made it Instagram. People don't [12:59] like it. I don't have necessarily super [13:02] strong opinions on it personally [13:03] immediately. I do think the original [13:06] Instagram word mark here on the left [13:09] >> had started to feel extremely dated. [13:12] >> Mhm. [13:13] >> But it's still iconic. I feel like we're [13:16] headed back to this sort of maximalism [13:19] and branding, right? like however I [13:22] don't know how many years ago it was [13:23] like five years ago all the big fashion [13:26] houses started going from their like [13:27] historical word marks updating them and [13:30] making them like much more simple and [13:32] just [13:32] >> Yeah. Balenciaga. [13:33] >> Yeah. Yeah. And there's like 10 10 [13:35] different examples. [13:36] >> Yeah. [13:37] >> And so this this move almost feels a [13:39] little bit like lagging in some ways [13:41] where I actually like that the Instagram [13:43] logo was like it it it did feel dated [13:45] but it it was so distinct. [13:47] >> Yeah. I sent a different treatment that [13:50] I thought they should go with. [13:51] >> I mean, it's not too late. So, [13:53] >> I think that would speak to me [13:56] personally a little bit better. Uh, but [13:59] uh, you know, their their treatment, [14:01] it's a choice. It's clean. Louis Vuitton [14:04] collaborated with Singer and created [14:06] this insane Singer X Louis Vuitton [14:10] collaboration for a 911 that looks like [14:13] a handbag. They actually they're calling [14:15] it the most tacky car. [14:18] >> I was I didn't know where you were going [14:20] to go. I knew I saw this all over the [14:22] all over the feed. I knew we were going [14:24] to be talking about it, but I didn't [14:25] know where you'd sit. [14:27] >> Yeah. So, [14:28] >> why is it tacky? It looks pretty [14:29] >> Let's play this video. [14:31] >> This isn't that tacky. [14:33] >> I like the color. [14:35] >> You don't like the color? [14:37] >> The blue wheels are tacky. That this [14:40] part looks nice. [14:41] >> Well, okay. [14:42] The bag [14:44] >> interior is just brutal. [14:45] >> The stick is pretty cool. [14:47] >> The hood with the straps. Extreme. [14:49] >> This part's sort of crazy. [14:51] >> Yeah, this part's sort of [14:52] >> helmets. I I don't hate the helmet. [14:55] >> There's another version that doesn't [14:56] have blue wheels. It's a little more [14:58] subdued, but [15:00] >> it it feels like AI slop [15:02] >> IRL. This is like to me like a tacky [15:05] unauthorized collab. [15:07] >> I know exactly the kind of person that [15:09] would would buy this. [15:10] >> Look at that. People are saying next [15:12] level gluttony, [15:15] aspirated slop. [15:18] >> People do not like it. New bombshell [15:20] reporting from the Wall Street Journal [15:22] reveals that North Korea has built a [15:25] secret workforce inside American [15:28] companies using stolen identities, [15:31] AI, and accompllices in the United [15:33] States to cheat its way into remote jobs [15:36] and funnel hundreds of millions of [15:38] dollars back to Kim Jong-un's regime. Uh [15:41] it's a fascinating story. The Wall [15:42] Street Journal article is posted uh it's [15:46] a 30inute documentary. They spent over a [15:48] year investigating this. The FBI says [15:50] that there are thousands of North Korean [15:53] IT workers applying for jobs across [15:55] America. After a year-long [15:57] investigation, the Journal obtained a [15:59] trove of leaked browser histories, [16:01] emails, calendars, and screen recordings [16:03] from one cell of workers, providing a [16:06] remarkable look at how the operation [16:08] actually works. The North Koreans apply [16:10] for jobs at enormous scale. one cell [16:12] tracked by the journal applied to more [16:14] than a thousand companies over just [16:16] three months using AI at nearly every [16:18] step and in the documentary they show a [16:21] product that is basically clearly they [16:23] will leave a voice agent running while [16:25] they're being interviewed in a technical [16:27] interview give the answer when asked [16:30] about a particular technology if they [16:31] are proficient they will simply have AI [16:33] look up the answer give the answer to [16:36] the interviewer get the get the job and [16:39] in some cases they're using real [16:41] Americans as frontmen. Basically, they [16:43] will do the interview and then they will [16:46] ship a remote laptop to say, "Hey, [16:48] you're a new you're our new remote [16:50] worker. You are qualified to work on our [16:52] software engineering team at this small [16:54] company. We'll send you a laptop and [16:56] then you just collect a check, send half [16:59] of it to the North Koreans, and you [17:01] don't have to do any work. So, it's like [17:02] passive income, free money for you, the [17:04] American." [17:05] >> So, you said they're real Americans, but [17:07] they're not patriots. Yeah, that's what [17:09] I'm hearing. Well, I I the the the [17:10] documentary is pretty gut-wrenching. [17:12] Like the the the the individual who they [17:14] talked to who uh who was participating [17:17] in this obviously uh is probably going [17:20] to see some uh some legal consequences [17:23] for this, but uh it did seem like he had [17:25] a very very tough go and got into a very [17:28] rough situation to be in that situation. [17:30] The operation also relies on help from [17:32] inside the United States. North Korean [17:34] IT workers pay American facilitators, [17:36] they're called facilitators, to host [17:38] company laptops in the United States, [17:40] allowing the actual workers overseas to [17:42] remotely connect to them while appearing [17:44] to be employees working domestically. [17:47] Also, the North Korean workers, they go [17:49] to non-extradition countries. So, [17:51] they'll go to China and Russia and a few [17:53] other countries. And then from there, [17:55] they will be remoting into an American [17:57] laptop. So it just looks like okay there [18:00] like the web traffic is coming from it's [18:02] not coming from North Korea but it's [18:04] coming from a country where if the FBI [18:05] says hey can you send us this person [18:07] they're a criminal that country says no [18:10] way uh we're not we're not sending them [18:11] to you. So for one job paying $75,000 a [18:14] year. He said he and the North Koreans [18:16] split split the salary 50/50. Uh the [18:19] workers uh also use stolen American [18:22] identities to pass employment [18:23] screenings, often juggling multiple [18:25] identities simultaneously and holding [18:27] down several jobs under each name. This [18:29] all boommed during COVID and the remote [18:31] work boom. Um and the scale of the [18:34] operation is enormous. Some North Korean [18:36] IT workers earn as much as $300,000 a [18:39] year. The Treasury Department says North [18:40] Korean says the North Korean government [18:42] can seize as much as 90% of the wages [18:45] earned by its overseas IT workers. and [18:48] recently estimated that these operations [18:50] generated nearly $800 million in 2024 [18:53] alone. Pretty pretty big. So, the [18:55] journal's 30inute documentary titled [18:58] Infiltrated: North Korea's Secret US [19:00] Workforce follows the operation in [19:02] detail. It's a fascinating watch and I [19:05] highly recommend that you check it out. [19:06] So, go take a look. A new luxury hotel [19:08] canled a YouTuber's $4,663 [19:13] stay. Then came the viral feud, reports [19:17] the Wall Street Journal. A dispute [19:19] between Aman, which we've talked about a [19:20] lot on the show, and content creator [19:22] Ryan Walker illustrates the growing [19:25] tension between commercial enterprises [19:27] and influencers. It's a fascinating [19:29] story. And somebody's got to stand up [19:32] for the Amand. Somebody has uh No, [19:36] honestly, I don't I don't think the Aman [19:38] actually needs that much standing up [19:41] for. [19:41] >> No. The Wall Street Journal broke it [19:43] down in great detail and it looks bad [19:45] for this guy. Anyway, let's go through [19:48] it. The Wall Street Journal reports that [19:51] uh it was supposed to be the hottest [19:53] hotel opening of the year. Let's play a [19:55] little bit of his video. [19:56] >> Don't know me. I'm a luxury hotel [19:58] reviewer here on YouTube. I specialize [20:01] in very honest reviews of the world's [20:03] best properties. [20:05] I'm not going to give away too much as [20:06] to what happened. I want you to watch. [20:08] >> And this is a cool thesis. It's like he [20:09] pays for the hotels himself. He doesn't [20:12] get paid by the hotel for the review. [20:16] The hotel doesn't pay for his stay and [20:18] consequently he can be more independent. [20:21] This was the Doug Demiro strategy. [20:22] >> Excuse me, Mr. Walker. We don't have [20:24] your reservation. [20:25] >> Long story short, that video sparked a [20:28] bunch of, you know, there's thousands of [20:32] comments on there. People trashing [20:34] >> the properties, you know, basically [20:37] fully taking his side. Yeah. [20:38] >> He says, "I'm honest and transparent. [20:41] I'll tell you exactly what to expect and [20:42] why." He had previously paid $26,000 for [20:45] a Four Seasons yacht trip that he panned [20:47] in May. [20:48] >> And sorry for the spoiler. He's not [20:50] honest or transparent. [20:52] >> George is so upset. Staff recognize him, [20:55] inform him that he has no reservation, [20:57] and according to Walker, eventually call [20:58] the police to escort him away. Sounds [21:00] terrible. As Walker drives off, he finds [21:03] an email he says he missed, sent the day [21:05] before. apologizes for canceling his [21:07] stay due to scaled back capacity during [21:09] opening week. Not looking closely at the [21:11] time stamp, he tells viewers the email [21:13] arrived late the previous night. When [21:15] later pressed by the journal, Walker [21:17] revised his timeline, saying it arrived [21:20] the previous morning. Walker did not [21:22] amend the timeline in his video or in a [21:24] subsequent live stream. The internet [21:26] quickly took Walker's side. Viewers of [21:28] his YouTube channel, including people [21:29] who identified as Aman loyalists and [21:32] travel industry professionals, expressed [21:34] disbelief, will never book another Aman [21:37] property again, read one of the more [21:39] more than 5,000 comments, many of which [21:41] expressed similar sentiments. Aman [21:43] spokesperson said the company does not [21:45] comment on bookings. The Aman [21:46] spokesperson said Walker bypassed the [21:49] main entrance, ending he was looking for [21:52] trouble access. My theory is he knows [21:55] >> exactly what he was doing. M. [21:58] >> So he comes during the first week, books [22:00] one night, knows that he's shouldn't be [22:02] filming, still decides to film. [22:05] >> Yeah. [22:06] >> And is effectively looking for trouble [22:08] the entire time. [22:10] >> And the crowd over on the Wall Street [22:11] Journal loves it. The top rated comment, [22:13] "I'm on, here I come." Any hotel that [22:16] bans influencers is doing regular guests [22:19] a great service. [22:20] >> Yeah. Yeah. I mean, that's a big thing [22:21] is that people go to certain places to [22:24] not have cameras all over the place. And [22:26] uh it's getting rarer and rarer and uh [22:29] if if uh [22:30] >> it's so it's so it's so actually [22:31] fascinating to look at the difference. [22:33] YouTube comments are like wow I'm never [22:35] going to go to the Aman and then Wall [22:37] Street Journal is like [22:38] >> I'm looking now [22:39] >> Walker is a complete tool says Ray [22:42] Edward says Edward says seems Mr. Walker [22:45] struggles with the truth. Anthropic is [22:48] apparently in talks to buyart [22:52] for6 billion [22:54] >> image generation. [22:56] >> We've had a bunch of awesome [22:57] conversations with with Dean over at [22:59] Dart. He's always he's very very early. [23:03] He was willing to do [23:04] >> live demos of their product like in [23:07] interviews completely un [23:10] >> you know we didn't even test with him [23:11] before the show. He was just ripping [23:14] them live. So, always had a ton of [23:15] confidence in the product. Um, and [23:18] they've been just cooking. [23:19] >> They were valued at 4 billion a quarter [23:21] ago. Uh, himu has some extra context [23:24] here. Says, "Decart could be one of the [23:26] first AI labs focused on world video [23:28] models to be acquired by a Frontier AI [23:31] lab. I think this also could mark an [23:33] initial phase where Frontier Labs start [23:35] treating world models as a major [23:37] strategic capability alongside inference [23:40] optimization as a major offering from [23:43] Decart. Uh yeah, interesting to imagine [23:47] how this actually links to the core [23:51] thesis of you know B2B and enterprise [23:54] and and coding but uh certainly [23:57] certainly an amazing technology that [23:59] feels like on the verge of uh of a [24:01] breakout. Uh the demos are amazing but [24:04] we haven't had the like Giblly moment [24:06] for world models yet. like it's very [24:09] much a prototype demo video. Go and see [24:12] what it's looks like. But we've talked [24:14] to a lot of people Oliver Cameron and [24:16] Feay Lee about world models. There's a [24:18] lot of optimism about uh how this all [24:21] plugs into the AGI pursuits of the [24:24] Frontier Labs. Are you a standard crying [24:28] face emoji guy, a tilted crying face [24:31] emoji guy, or a tear streaming down your [24:35] face emoji guy? I'm uh normally [24:40] >> Okay, I want to actually pull this up so [24:42] I can see more closely. [24:43] >> Which one are you? Joe Weisenthal has [24:46] recently transitioned to a tilted crying [24:50] face emoji. [24:51] >> I actually I don't I don't dabble in the [24:54] tilted. I do I do the tears. [24:55] >> I do the tears and the straight on. [24:57] >> Oh, wait. Do you do the streaming down [24:58] the face? [24:59] >> Yeah. Yeah. Yeah. [25:00] >> Oh, I I need to mix that one in. It's [25:02] the most popular emoji of 2025. [25:04] >> Yeah. I went through maybe a decade [25:07] where I I wouldn't I wouldn't touch any [25:10] of the crying. [25:11] >> Oh, any of the crying ones. You do a [25:12] smiley, [25:13] >> but I've been I've been laughing a lot. [25:15] Okay. The last couple years, mainly [25:17] because we've been doing this show. You [25:19] got a good time. [25:20] >> Emoji. I like that in iMessage you can [25:22] do the haha, but then you can also throw [25:24] the crying emoji. But I got to [25:26] experiment with the tear streaming [25:27] emoji. [25:28] >> You ever go with the cat versions? [25:29] >> No. I never LAR as the cat on. [25:32] >> Never go Never go cat. But expect a cat [25:35] emoji from me soon, Tyler, in our group [25:38] mass in our group chat because uh might [25:40] be underrated. There might be some alpha [25:41] there. [25:42] >> Well, John, I'm going to read a buco [25:44] capital bloke post that I think applies [25:47] to this exact moment right now. [25:49] >> He says, I think we're very close to the [25:51] point where caring about AI or talking [25:53] about it a lot is a bit embarrassing. [25:55] Move on already. Who cares? [25:57] >> Move on already. I mean, there is a [25:59] point where like we don't talk about the [26:01] internet anymore. or we talk about [26:03] what's happening on the internet. We [26:04] talk about particular internet [26:06] companies. Uh this stuff does diffuse. [26:09] If it diffuses fully, you stop talking [26:11] about the the underlying technology. But [26:14] there's a horse race on Buco. There's a [26:16] lot of money on the line. The entire [26:18] global economy potentially. But [26:20] >> that's right. He's just horsing around. [26:22] >> Thank you for tuning in to TVPN. Sign up [26:25] for our newsletter at tbpn.com. Leave us [26:26] five stars on Apple Podcasts and [26:29] Spotify.