Town CEO Jean-Denis Greze on the empty-box problem and why most people still don’t know what to ask AI



Most people still treat powerful AI tools like a slightly smarter Google. They open ChatGPT or Claude, type a question, get an answer, and move on. Very few are using these systems the way the tech world talks about them—as agents that actually *do* things on your behalf.

Jean-Denis Greze, CEO (and self-described “mayor”) of Town, thinks that’s the real bottleneck. In a recent conversation with *Fast Company*, he laid out why the empty chat box is failing ordinary users, why the word “assistant” might be the wrong frame, and why the AI race feels uniquely brutal right now.

Town, the San Francisco startup Greze cofounded with Tony Vincent in late 2024, just raised a $108 million Series B at a $1 billion valuation. Greze previously scaled Plaid’s engineering team from roughly 20 people to 350 as CTO. The company’s bet is straightforward: make AI practical and usable for small businesses and mid-market firms that aren’t already living inside agent workflows.

The empty-box problem

Could Claude or ChatGPT already do most of what Town does? Greze’s answer is essentially yes. The issue isn’t capability. It’s onboarding.

“Most people who use ChatGPT or Claude aren’t using those tools to do what people in tech describe as agentic work,” he said. They treat them as search replacements. Put a mainstream person in front of a blank box and ask them to invent the right prompt while the product’s abilities keep shifting underneath them, and you’ve created a recipe for underuse.

No one has yet figured out how to take regular people and get them to extract the full potential of these systems. That gap is the market Town is chasing—though Greze is candid that distribution could still win and the big platforms could simply absorb the opportunity.

 A small-business use case that stuck

One example that surprised Greze came from a small travel operation at a vacation destination. The business runs almost entirely over email. When rental gear (cars, bikes, etc.) comes back damaged, employees snap photos and email them to a central processor.

Town now runs a daily workflow that pulls those emails and photos, cross-references the CRM, and drafts the follow-up messages. What used to take 25 minutes of tedious drafting every day now takes about five. No traditional SaaS product was ever going to be built for a workflow that specific and small.

 Don’t call it an assistant

Town calls its product a “Townie.” Greze is deliberate about avoiding the terms “assistant” or “chief of staff.”

Those words carry human mental models. What this technology can do today—and what it will do in five years—will likely outgrow those expectations. Most people outside Fortune 500 executive suites don’t even have a clear picture of what a chief of staff actually does.

His preferred analogy is closer to a browser or a device. When the App Store arrived, a meaningful percentage of what people used to do in a web browser shifted to the phone. AI is doing something similar: the share of time we spend interacting with information through browsers and apps will keep declining, while the share we spend talking to an AI will rise. Voice, he believes, is the eventual interface.

 What’s next: live voice and agent-to-agent

Greze is particularly excited about two directions.

First, live voice. Imagine commuting in the morning with a capable agent in your ear: “You got seven emails overnight. Two are urgent. On this one, if you tell me X I can draft the reply. The other requires moving a meeting later today—want me to surface some options?” You can interrupt and steer in real time.

Second, agent-to-agent collaboration inside a company. You could ask, “Who here is close to someone in R&D at Company X?” Your agent quietly consults other agents, discovers that Bob plays tennis with the relevant person, and surfaces a privacy-preserving suggestion: “Bob appears closely connected—want me to ask if he can make an intro?” The network effects inside an organization become usable without forcing people to broadcast their relationships.

 On connecting the sensitive stuff

Should people hand their financial (or medical) accounts over to an AI agent? Greze thinks consumers eventually will, because the upside is too large—flagging high credit card rates, suggesting better options, spotting free money. But trust is non-negotiable. People will only do this with systems they believe are genuinely acting in their interest.

Town’s model reflects that view. Consumer use cases are subsidized by business customers. The main plan is $50 a month—not a $5 experiment. “If you’re not paying for a service, you’re not the customer,” Greze noted. “You’re being monetized in some other way.”

 The psychological cost of the AI race

The hardest part of running Town, Greze said, is the sheer velocity. Software used to enjoy a real early-mover window because copying took time and fixed costs were high. In AI, good ideas get replicated in weeks or a couple of months.

“I’ve never seen anything like it. It’s nerve-racking,” he said. At 47, he is working harder than he ever has. The emotional swing is constant: elation when you ship something before a bigger player, or sudden fear when a tweet from Sam Altman lands and shifts the ground under you.

One personal use case

Outside work, Greze has Town research San Francisco food coverage. Every Monday it looks ahead at the next month and surfaces four classic restaurant tables that are bookable plus four newer spots that have been getting attention in food blogs and newspapers—fuel for his weekly Sunday date night with his wife. He likes good food. He just doesn’t have the time to do the research himself.

The through-line in Greze’s comments is clear: the technology is already capable of more than most people are extracting from it. The unsolved problem is making that capability feel obvious, trustworthy, and useful to people who don’t live inside the AI discourse every day. Town is trying to close that gap—one Townie at a time.

Post a Comment

Previous Post Next Post