Why a $1M Startup Wasn’t Worth Building in the Age of AI

在这期播客节目中，Andreessen Horowitz 的合伙人亚历克斯·兰佩尔（Alex Rampell）与 Town（也可拼写为 Towne）的首席执行官兼创始人让·丹尼斯（Jean Denis）进行了交谈，探讨了他的创业历程，以及从面向中小企业的税务产品转型为横向人工智能助理公司的过程。 讨论始于他们的过往交集：在让·丹尼斯（Jean-Denis）担任 Plaid 首席技术官时，兰佩尔曾投资过该公司。离开 Plaid 并休息了一段时间后，让·丹尼斯最初推出了一款由人工智能驱动的 B2B 中小企业税务工具。尽管从技术和产品角度来看行得通，但他们意识到该产品面临着严重的市场开拓局限性。由于报税是一年一度的事件，它无法成为一款日活工具，缺乏口碑传播回路，并且在现有巨头复制该技术之前面临着倒计时般的压力。意识到这不是一家具备风险投资规模的公司后，让·丹尼斯和他的联合创始人决定在一年后将其关停——这是一个由他们对产生影响力和团队成功的核心动力所驱动的艰难转型。 在探索新想法时，他们提出了一个问题：为什么没有人构建出能够基于电子邮件和日历出色工作的智能体？他们发现，没有人围绕会议、会议准备和回复电子邮件等常见的知识工作者工作流构建出优秀的人工智能。利用更新模型的智能体能力，他们非常迅速地实现了产品与市场的契合。 让·丹尼斯解释了 Town 的战略：从一个横向产品起步，专注于面向中小企业和中型市场公司的产品驱动增长（PLG），同时确保数据安全和隐私基础能够扩展到企业级的销售驱动增长。他们刻意避免从自上而下、重合规性的企业需求入手，以保持消费级的产品体验，并坚信企业领导者会要求获得他们在个人体验中感受到的那种神奇效果。 在谈到与 OpenAI 和 Anthropic 等基础模型提供商的竞争时，让·丹尼斯认为 Town 提供了一个中立的智能层。Town 没有被绑定在单一供应商身上，而是将请求路由至当前表现最佳的模型提供商，从而节约资金并降低对单一供应商依赖的风险。此外，Town 确保企业真正拥有自己的数据、上下文和记忆，这代表了人工智能时代的关键资产层。 对话探讨了人工智能如何改变软件格局。让·丹尼斯认为，价值单位已经从应用程序转移到了计算机上，因为人类可以通过人工智能在计算机上做几乎任何事情。Town 充当了一个代币分配机器，能够自动化长尾研究和任务——例如餐厅老板分析批发商的价格变动——而这些任务以前因为过于繁琐而无法手动执行。 最后，他们讨论了为什么人工智能的普及在整个经济中并非均衡分布。虽然硅谷被各种爱动脑筋的修补者环绕，但世界其他地方依赖的是能够立即展现价值的产品体验，而不是一个空白的文本框。Town 通过观察聚合的高级用户行为、提供量身定制的软件例程，以及降低非技术人员有效部署人工智能的门槛，解决了这个问题。兰佩尔和让·丹尼斯在讨论传统供应链聚合和人工摩擦的侵蚀将如何根本性地重塑分销、消费习惯以及零售和广告等行业中结束了对话。

In this podcast episode, Alex Rampell, partner at Andreessen Horowitz, talks with Jean Denis, CEO and founder of Town (also spelled Towne), about his entrepreneurial journey, pivoting from an SMB tax product to a horizontal AI assistant company. The discussion begins with their history: Rampell had invested in Plaid when Jean-Denis was its CTO. After leaving Plaid and taking some time off, Jean-Denis initially launched a B2B AI-powered SMB tax tool. Despite working from a technical and product standpoint, they realized it faced severe go-to-market limitations. Because tax is an annual event, it could not be a daily active use tool, lacked word-of-mouth loops, and faced a ticking clock before incumbents copied the technology. Realizing it was not a venture-scale company, Jean-Denis and his co-founder decided to shut it down after one year—a difficult pivot driven by their core motivations for impact and team success. Exploring new ideas, they asked why nobody had built a great agent working out of email and calendar. They found that no one had built great AI around common knowledge worker workflows like meetings, meeting prep, and answering emails. Leveraging the agentic capabilities of newer models, they achieved product-market fit very quickly. Jean-Denis explains Town’s strategy: starting as a horizontal product with a focus on product-led growth (PLG) for SMBs and mid-market companies, while ensuring data security and privacy foundations can scale up to enterprise sales-led growth. They intentionally avoided starting with top-down compliance-heavy enterprise requirements to preserve a consumer-grade product experience, betting that enterprise leaders will demand the same magic they experience personally. Addressing competition with foundational model providers like OpenAI and Anthropic, Jean-Denis argues that Town offers a neutral intelligence layer. Rather than being tied to one supplier, Town routes requests through whatever model provider is best at the moment, saving money and de-risking vendor lock-in. Furthermore, Town ensures that businesses truly own their data, context, and memories, which represents a critical asset layer in the AI era. The conversation explores how AI changes the software landscape. Jean-Denis argues that the unit of value has shifted from the app to the computer, as a human can do almost anything with a computer using AI. Town acts as a token allocation machine that automates long-tail research and tasks—such as a restaurant owner analyzing wholesaler price changes—that were previously too tedious to execute manually. Finally, they discuss why AI adoption has not been evenly distributed across the economy. While Silicon Valley is surrounded by tinkers, the rest of the world relies on product experiences that demonstrate value immediately rather than a blank text box. Town solves this by observing aggregate power-user behaviors, providing custom-fit software routines, and lowering the barrier for non-tinkers to deploy AI effectively. Rampell and Jean-Denis conclude by discussing how the erosion of traditional supply aggregation and human friction will fundamentally reshape distribution, consumer habits, and industries like retail and advertising.