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

Video summary

Jean-Denis Greze shut down a nearly $1M tax startup to build Town, an AI assistant for email, calendars, and work.

Plaid’s former CTO Jean-Denis Greze explains why he closed an SMB tax startup that earned just under $1 million in its first year and was projected to reach $2.5–$3 million in year two. Despite a working product, annual tax sales and entrenched competitors made venture-scale growth unlikely; after tax season, the founders had roughly a month to pivot. They turned to Town, an AI assistant built around email and calendars, after customers described routine needs such as meeting preparation and drafting replies. Greze calls the product a “token-allocation machine”: cheap AI can handle recurring tasks that would not justify a human assistant, from meeting research to checking restaurant wholesalers for price increases that could cost owners hundreds or thousands of dollars. He argues Town should begin as a flexible product for individuals and SMBs, then tailor automations by role while adding enterprise security. Its edge, he says, is model neutrality and customers’ ownership of their business context. The conversation closes on AI adoption and distribution: most businesses need AI to demonstrate concrete value, while neutral agents could disrupt ad-supported platforms and supply aggregators.

Chapters

1. 0:00Leaving Plaid for SMB tax: under $1M in year one, but incumbents had the edge
2. 6:57Walking away from a working tax business before the October pivot window closed
3. 10:28How Town was born: customer requests led to an email-and-calendar agent
4. 16:16The iPhone moment for AI at work: Town starts with email and calendar
5. 23:22Town as a token-allocation machine: meeting research and restaurant price savings
6. 29:10Horizontal vs vertical: Town’s tailored routines versus lab FDE teams
7. 34:03Why Town over OpenAI and Anthropic: neutral models and portable business context
8. 39:19Why Town over OpenAI and Anthropic: no training, trusted agents, and a broader market
9. 44:32Why AI is underhyped outside the Valley: surfboard inspections and the tinkerer gap
10. 48:54Why AI is underhyped outside the Valley: ChatGPT’s progress and concrete Town demos
11. 52:33Neutral agents and the broken ad covenant: agent traffic surpasses humans and raises ad conflicts
12. 56:36Neutral agents and the broken ad covenant: Muse, Camel Camel Camel, and the decline of supply aggregation