a16z Podcast: Why Town’s AI Pivot Beat a $1M Startup Bet

SPEAKER_1 (0:01)

When AI changes the rules of the game, even a promising startup can become the wrong bet overnight. Today we're breaking down an episode of the a16z Podcast, where Alex Rampell talks with Jean-Denis Greze, the CEO and founder of Town. It's a fascinating story about pivoting from a tax product to an AI company built around everyday work. Yeah, and what really struck me about this conversation is how much it's about timing and distribution. Jean-Denis left Plaid and initially pursued SMB tax because it seemed like a natural place for AI to help. But here's the thing—the product worked, but the business model didn't scale fast enough. Right, he said it pretty bluntly: 'It wasn't going to be a venture-scale company.' But the problem wasn't just competition, was it? No, exactly. It was that incumbents could use AI to automate their own tax services before a new entrant could win enough customers. In annual businesses, there's only a tiny window to win, and tax was too discontinuous to become a huge, fast-growing platform. That's a brutal realization.

So that led to a pivotal question: 'Why has no one built a great agent that works out of email and calendar?' Town shifted toward the daily flow of knowledge work—meetings, prep, routine email tasks. Jean-Denis says the goal isn't to sell an abstract assistant, but to find real opportunities where AI saves time or money in ways users can feel immediately. I love how he describes the product as a system that helps people deploy AI better at their jobs, almost like a token allocation machine. Building on that point, they spend a lot of time discussing how AI products actually win.

Jean-Denis makes this great observation: 'The product experience can't just be a text box.' That's such an important insight. The software has to do the work for the user, show value quickly, and earn trust before asking for deeper access to calendars, email, or company data.

He argues that consumer-grade expectations will shape enterprise software, much like the iPhone changed what people expected at work after using it at home. What about the economics piece? That seemed really forward-thinking. Yes! Jean-Denis says, 'The unit of value in software has shifted from the app to the computer.' He envisions a future that's less human-in-the-loop and more automated, with neutral agents choosing the best model for the task while preserving customer ownership of data and context.

And that neutrality matters, especially as model providers change quickly.

The episode closes on this broader idea that AI is diffusing in strange, specific corners of the economy—from restaurant supply analysis to automated poster design. The challenge isn't whether the intelligence exists, but whether companies can make non-tinkerers feel the value that tinkerers already see. As Jean-Denis puts it: 'The best way to predict the future is to observe the present.' Pretty compelling framework for thinking about where AI is heading.