"光锥"播客的主持人深入探讨了新兴的垂直领域AI Agent（人工智能代理），预测它们将孕育出许多市值超过3000亿美元的公司。他们将其与SaaS（软件即服务）公司的崛起相提并论，SaaS公司在过去二十年中一直主导着硅谷的融资，占风险投资的40%以上，并催生了300多家独角兽企业。SaaS的成功建立在诸如XML HTTP request（Ajax）等技术进步的基础上，这些技术实现了富互联网应用程序，并促进了从桌面软件到基于Web的服务的转变。 该播客探讨了大型语言模型（LLM）的出现如何带来类似的范式转变，从而实现根本性的新能力。主持人将云和移动技术带来的第一波机遇分为三类：显而易见的大众消费产品（例如，文档、电子邮件）、意想不到的大众消费创意（例如，Uber、Airbnb）以及B2B SaaS。初创公司在第一类中基本上输给了谷歌和Facebook等巨头，在第二类中通过在巨头最初没有竞争的领域进行创新而取得了成功，并在第三类中占据了主导地位。没有一个统一的“SaaS领域的微软”，从而导致了专业化SaaS公司的蓬勃发展。 他们认为，LLM也会出现类似的情况。虽然像通用AI助手这样显而易见的大众消费应用很可能被科技巨头所主导，但初创公司的潜力在于开发专注于特定行业和任务的垂直领域AI Agent。正如早期的Web应用最初很笨拙一样，早期的AI应用可能会面临诸如“幻觉”之类的挑战，但它们会像SaaS一样迅速发展。 主持人随后讨论了为什么现有企业未能渗透到B2B SaaS领域，他们认为原因在于难以在众多领域构建专门的解决方案。B2B SaaS需要在利基领域拥有深厚的专业知识，这使得大型公司难以与专注于特定领域的初创公司竞争。此外，传统的企业软件通常用户体验较差。相比之下，垂直领域SaaS公司可以通过专注于特定需求来提供卓越的体验。 AI带来的范式转变在于，它不仅有可能取代软件，还有可能取代操作软件所需的人员。初创公司可能会更有效率，所需的员工更少，尤其是对于重复性任务而言。这可能会导致只有极少员工的公司出现。 主持人讨论了垂直领域AI Agent的真实案例，例如调查领域的Outset和QA测试领域的Momentic。虽然旧的QA公司需要与现有团队合作，但AI Agent现在可以完全取代这些角色。同样，AI Agent Nico正在进行招聘人员通常会做的全部技术筛选。Capital AI的聊天机器人甚至减少了所需的开发者关系团队的数量。 该团队还讨论了来自YC公司的案例。一个例子是Salient，这是一个用于自动催收汽车贷款的AI解决方案。过去，这些任务必须在高流失率的呼叫中心完成。现在由Salient完成，并且正在与大型银行合作。 他们强调，垂直领域AI仍处于早期阶段，解决方案必须经过高度定制才能满足客户的需求。

The hosts of "The Lightcone" podcast delve into the burgeoning world of vertical AI agents, predicting they will be the source of numerous $300 billion+ companies. They draw parallels to the rise of SaaS companies, which have dominated Silicon Valley funding for the past two decades, representing over 40% of venture capital and spawning over 300 unicorns. The success of SaaS was predicated on advancements like XML HTTP request (Ajax), which enabled rich internet applications and facilitated the shift from desktop software to web-based services. The podcast explores how the advent of Large Language Models (LLMs) presents a similar paradigm shift, enabling fundamentally new capabilities. The hosts categorize the initial wave of opportunities arising from cloud and mobile technologies into three groups: obvious mass consumer products (e.g., docs, email), unexpected mass consumer ideas (e.g., Uber, Airbnb), and B2B SaaS. Startups largely lost out in the first category to incumbents like Google and Facebook, found success in the second by innovating in areas where incumbents didn't compete initially, and dominated the third category. There is no singular "Microsoft of SaaS", leading to a proliferation of specialized SaaS companies. They suggest a parallel scenario will emerge with LLMs. While obvious mass consumer applications like general-purpose AI assistants will likely be dominated by tech giants, the potential for startups lies in developing vertical AI agents focused on specific industries and tasks. Just as early web apps were initially clunky, early AI applications may face challenges like "hallucinations," but they will evolve rapidly, much like SaaS did. The hosts then discuss why incumbents didn't penetrate the B2B SaaS space, citing the difficulty of building specialized solutions across numerous domains. B2B SaaS requires deep expertise in niche areas, making it challenging for large corporations to compete with focused startups. In addition, legacy enterprise software often suffers from poor user experience. In contrast, vertical SaaS companies can deliver superior experiences by focusing on specific needs. A paradigm shift with AI is the potential to supplant not only software but also the personnel needed to operate it. Startups are likely to be more efficient and require fewer employees, especially for repetitive tasks. This could lead to companies with minimal staff. The hosts discuss real-world examples of vertical AI agents, such as Outset in the survey space, and Momentic in QA testing. While older QA companies needed to work with the existing team, AI agents can now fully replace those roles. Similarly, Nico, an AI agent, is doing the entire technical screen that recruiters would normally do. Capital AI's chatbot has even reduced the number of dev rel teams required. The team also discusses examples from YC companies. One example is a Salient, an AI solution for automating the calling for auto loan payments. In the past, these tasks would have to be done in call centers with high turn over. This is now being done by Salient, and is now working with big banks. They emphasize that vertical AI is still in its early stages, and solutions must be heavily tailored to meet customer needs.