Jean-Denis Grez spent a year building an SMB tax tool that worked, with first-year revenue just below $1 million and a plausible $2.5–3 million next year. He still killed it because annual purchasing, weak word of mouth, and incumbents’ distribution made venture-scale growth unlikely.
Jean-Denis says tax customers kept asking for meeting prep and drafted answers, prompting Town to ask why no one had built a great agent for email and calendar. When models became more agentic, that simple workflow focus produced quick product-market fit.
Jean-Denis warns that when the buyer is not the user, teams can end up pleasing the CISO instead of the user. Town focused first on a consumer-grade product and natural growth, then planned to add enterprise admin, security, and compliance features later.
Alex Rampell compares work AI to the iPhone moment, where personal AI magic makes CEOs demand the same at work. Jean-Denis adds that agents need secure access to email, calendar, and systems of record, plus audit and cost controls, so enterprise adoption moves at human speed.
Jean-Denis describes Town as a token-allocation machine: it suggests where spending tokens can improve a user’s work. Cheap tokens make long-tail tasks worth doing, such as prep notes for every meeting or a monthly wholesaler price review that occasionally finds real savings.
Alex says Town started horizontal so the market could reveal verticals instead of presuming them. Jean-Denis explains that Town interviews power users, groups common routines, and seeds templates for roles like restaurant owners or electricians before AI customization takes over.
Jean-Denis argues AI harness capabilities will be similar, so feature leads are hard to sustain. He says software value is shifting from the app to the computer-like agent, and Town bets on AI customizing routines for each user’s role, such as electrician or venture capitalist.
Jean-Denis contrasts Town’s automated routine library with OpenAI-style forward-deployed engineers, who build custom software inside a business. He thinks FDE work looks good for large enterprise deals now, but model progress will make handcrafted deployments less scalable than PLG customization.
Jean-Denis tells mid-market buyers they cannot know which lab will be best over three to five years. Town positions itself as model-neutral, routing across suppliers for intelligence, voice, video, or slides and reducing lock-in, rent hikes, and the risk of backing the wrong provider.
Jean-Denis argues the context and memories AI builds about a business are becoming a core asset, alongside physical and human capital. Town lets customers truly own that layer and says switching should take a day; FDE deployments instead try to make switching harder.
Jean-Denis describes a Town internal CRM where an agent finds a colleague connected to a target customer by querying other employees’ agents. Instead of Slack’s slow human routing, users get immediate answers, with AI trusted to mediate privacy and security for that request.
Jean-Denis cites a surfboard-rental owner who used Town to automate deposit checks. He says AI use appears in odd pockets, but depends on a tinkerer; most businesses lack enough tinkerers, so people treat ChatGPT as better search rather than building time-saving workflows.
Alex Rampell says agent traffic has surpassed human traffic and Amazon’s blocking response breaks the old ad-for-eyeballs covenant. He argues welcoming agents could be the next SEO opportunity, especially for a Walmart-like player whose incentives differ from an advertising business.
Jean-Denis says supply aggregators like Amazon may lose value because their model depended on limited human attention. If neutral agents automate that attention away, blocking them is weaker than building a good personal agent and finding another way to monetize the relationship.
