Winning the assistant race in weeks, not years: Town's Jean-Denis Greze on the work-agent bet
Key takeaways from the In Depth interview with Town co-founder and CEO Jean-Denis GrezeDealroom has a profile for this one. Try Dealroom → (October 2026):
Agents are the new window pane. Greze says Town is not building another app but the interface through which people will interact with most content — a browser or device for AI, which is why the stakes now feel like a zeitgeist.
Work beats personal on economics. AI is a way to use power to create intelligence; in personal life there is a limit to how much intelligence you want to spend on, but in work, every dollar applied to revenue-generating tasks invites more. The business side also has more tolerance for sharing data, which lets AI break down the data silos that exist today.
Two assistants per person. Privacy and data ownership push people to run two interfaces — one personal, one work. Employers do not want preparation data or notes leaving with an employee, and work and personal versions need different APIs, integrations and end-user requirements even for the same use case, like travel.
Why the moment arrived. Anthropic's November and December model releases made autonomous white-collar work possible — the trigger for OpenClaw's takeoff — and the June–July improvement in browser and computer use let agents operate tools such as Figma rather than only well-shaped APIs; accuracy moved from 70–80% to 90–95% with far less disruptive failures, and agents turned proactive.
Copy cycles collapsed to weeks. Competitors can dissect a winning feature and ship their own version in weeks to months, where it used to take years to decades. Town launched in June, felt untouchable through July, and now races rivals it did not expect; Greze says a product builder must be right four or five times over 18 months, with network-effect moats compressed from years to a quarter or two of clean air.
Never bet on a single lab. Building on top of models from companies that also compete with you is the norm: who has the best frontier model in one, two or five years is unknowable, so committing to a product that only supports its own models is the wrong bet — the smartest move is to stay multi-model.
Multiplayer agents need sharing rules. Town's bet on work agents over personal agents rests on agents working well among many people: Townies hand off tasks and ask each other questions, and the product is about getting agents to share the right information — the design question of what one agent chooses to reveal to another's.
Trust and guardrails. Greze argues trust is built by design: users approved 95% of agent actions, which is why Town started rolling its guardrails back, and he expects people to increasingly delegate decision-making to AI as assistants prove themselves.
Read more: In Depth — First Round Capital