由哈里·斯蒂宾斯主持，嘉宾包括杰森·莱姆金、罗里·奥德里斯科尔以及MongoDB首席执行官戴夫的播客，深入探讨了科技领域最重大的新闻，内容涵盖了从人工智能竞争到并购（M&A）和风险投资动态等一系列话题。 对话以OpenAI即将达到700亿美元的年化收入（run rate）开场，这标志着其对Anthropic的一次重大反击。虽然Anthropic此前被视为领先者，但OpenAI报告的70%季度环比增长表明了其巨大的重新加速。专家组指出，在人工智能时代，开发者的忠诚度发生了转变，他们会根据性能和成本迅速转向“闪亮的新玩具”。MongoDB首席执行官戴夫强调，对于企业而言，成本和知识产权是关键因素，这使得人工智能模型的快速采用令人印象深刻。即将到来的Anthropic首次公开募股（IPO）将揭示其增长是否受到了OpenAI复苏的影响。 接下来，讨论转向了“Factory与Cognition的惨败（debacle）”，涉及一名曾为Factory提供CRO咨询服务的人员随后加入了其直接竞争对手Cognition。这引发了关于忠诚度的辩论，特别是顾问与全职员工之间的忠诚度。杰森认为，在快节奏的科技世界中，尤其对于非董事会顾问而言，此类举动很常见，而戴夫和罗里则强调，当一个掌握机密信息的人加入竞争对手时，可能会造成声誉损害和被背叛的感觉。维诺德·科斯拉公开在推特上贬低Factory（他的公司曾牵头投资过该公司），进一步激化了这一争议，引发了对投资者行为以及“将声誉损害武器化”的质疑。 Salesforce以20亿美元收购ListenLabs的交易得到了分析，罗里称赞这是利用LLMs彻底改变市场研究（一个亟待颠覆的行业）的公司所取得的强大成果。然而，杰森对一笔20亿美元的收购，即使是创新的人工智能应用程序，如何能显著推动Salesforce这样一家营收接近600亿美元的巨头公司发展表示怀疑。戴夫反驳说，许多人工智能应用公司可能只是“伪装成公司的功能”，暗示有些公司可能会为了分发而出售给更大的平台，特别是当它们缺乏持久的“数据循环”护城河时。此次收购引发了与Intercom达到类似估值所需更长时间的比较，说明了当前人工智能环境下可能实现的快速价值创造。 专家组还讨论了Reflection，一个美国开源模型，认为它可能是美国与中国模型竞争的最佳希望。戴夫和杰森指出，企业对成本效益高、安全且“安全可靠”的美国本土模型有强烈需求，即使这些模型并非处于智能的绝对前沿。罗里在承认其潜力的同时，对新模型已发布的评估指标的可靠性表示怀疑。 Vercel令人印象深刻的增长，其新业务的50%现在由“代理”（agent）驱动（年初时仅为3%），凸显了产品采用方式的重大转变。杰森解释说，代理“选择”供应商是一种强大的力量，通常基于人工智能工作流程中的直接集成和信任，这对公司来说是一个挑战，需要确保他们的产品能够被这些人工智能实体发现并优先选择。 一场法律讨论源于Grox工程师起诉英伟达（NVIDIA），指控其一项110亿美元的许可和雇佣（license and hire）交易，认为这实际上出售了核心技术和关键员工，使普通股东只剩下一个空壳公司。罗里和戴夫强调，这挑战了所有普通股东都应受到平等对待的既定公司法原则。这种结构常被用作规避缓慢的并购监管审批的变通方法，但它对员工构成风险，并可能影响未来的“许可和雇佣”交易。 最后，Meta的“Muse”发布被赞誉为一款成功的产品，与OpenAI“Dots”“不尽如人意”的反应形成鲜明对比。哈里和罗里称赞了Meta在Muse上的执行力，而杰森则给了Dots更多时间，暗示其价值可能在于其持续编码代理能力，而非面向消费者的吸引力。尽管基本面强劲，但Aura首次公开募股（IPO）的失败也得到了讨论，专家组推测这可能是公司与银行家之间在估值上存在分歧所致。

The podcast, hosted by Harry Stebbings with guests Jason Lemkin, Rory O'Driscoll, and Dave, the CEO of MongoDB, delved into the biggest news in tech, covering a range of topics from AI competition to M&A and venture capital dynamics. The conversation kicked off with OpenAI nearing a $70 billion run rate, signaling a significant comeback against Anthropic. While Anthropic was previously seen as leading, OpenAI's reported 70% quarter-on-quarter growth suggests a massive reacceleration. The panel noted a shift in developer loyalty in the AI age, with developers quickly switching to the "shiny new toy" based on performance and cost. Dave, the MongoDB CEO, highlighted that for enterprises, cost and IP rights are crucial factors, making the rapid adoption of AI models impressive. The upcoming Anthropic IPO will reveal if their growth has been impacted by OpenAI's resurgence. Next, the discussion moved to the "Factory and Cognition debacle," involving a CRO advisory to Factory who then joined direct competitor Cognition. This sparked debate about loyalty, particularly concerning advisors versus full-time employees. Jason argued that in the fast-paced tech world, especially for non-board advisors, such moves are common, while Dave and Rory emphasized the potential for reputational damage and the perceived betrayal when someone with confidential insights joins a competitor. The controversy was amplified by Vinod Khosla's public tweet disparaging Factory, an investment in which his firm had led previous rounds, raising questions about investor conduct and "weaponizing" reputational damage. Salesforce's acquisition of ListenLabs for $2 billion was analyzed, with Rory praising it as a strong outcome for a company leveraging LLMs to revolutionize market research – an industry ripe for disruption. Jason, however, expressed skepticism about how a $2 billion acquisition, even of an innovative AI app, would significantly move the needle for a company the size of Salesforce, which is nearing $60 billion in revenue. Dave countered that many AI application companies might be "features masking as companies," implying that some might sell to larger platforms for distribution, especially if they lack a durable "data loop" moat. The acquisition prompted a comparison to Intercom's longer journey to a similar valuation, illustrating the rapid value creation possible in the current AI climate. The panel also discussed Reflection, a US-based open-source model, as potentially America's best hope for competing with Chinese models. Dave and Jason noted the strong demand from enterprises for cost-effective, secure, and "safe" US-sourced models, even if they aren't at the absolute frontier of intelligence. Rory, while acknowledging the potential, expressed skepticism about the reliability of published evaluation metrics for new models. Vercel's impressive growth, with 50% of its new business now driven by "agents" (up from 3% at the start of the year), highlighted a significant shift in how products are being adopted. Jason explained that agents "picking" a vendor is a powerful force, often based on direct integration and trust within the AI's workflow, a challenge for companies to ensure their offerings are discoverable and preferred by these AI entities. A legal discussion arose from Grox engineers suing NVIDIA over an $11 billion license and hire deal, arguing that it effectively sold off core technology and key staff, leaving common shareholders with a hollowed-out company. Rory and Dave highlighted this as a challenge to established corporate law principles where all common shareholders are treated equally. This structure, often used as a workaround for slow M&A regulatory approvals, poses risks for employees and could impact future "license and hire" deals. Finally, Meta's "Muse" launch was lauded as a successful product, contrasting sharply with the "underwhelming" reception of OpenAI's "Dots." Harry and Rory praised Meta's execution with Muse, while Jason gave Dots more time, suggesting its value might be in its persistent coding agent capabilities rather than a consumer-facing appeal. The failure of the Aura IPO, despite strong fundamentals, was also discussed, with the panel speculating that a disagreement on valuation between the company and bankers was likely the cause.