Jean-Denis Grèze, the CEO of Town, says his company runs with far less middle-management overhead than a startup its size would typically carry — not through a radical org chart redesign, but because AI assistants now absorb a chunk of the coordination work managers used to do. According to Platformer, Grèze describes Town internally as a self-organizing company, built around assistants that sit inside day-to-day workflows rather than living in a separate chat window.

The claim carries more weight than the average founder's future-of-work post because Grèze has actually run engineering organizations before. He previously led engineering at Instacart and co-founded Multi, the multiplayer collaboration startup OpenAI acquired in 2024. That's someone who has dealt with the messy reality of keeping dozens of engineers aligned — and who is now betting a real chunk of that alignment work can be automated.

For AI builders, the interesting part isn't the culture pitch, it's that Town functions as a live test of agentic tooling applied to real internal operations rather than a demo. Whether or not "self-organizing" holds up as a durable structure, the specific jobs Grèze hands to assistants are a useful map of what agent tooling is actually ready for today — and what it isn't.

What "self-organizing" means in practice

Grèze isn't describing a company without managers so much as a company where assistants take over the parts of management that are mostly about moving information around: who needs to know what, who's blocked on whom, and what got decided and why. Instead of a manager holding that context in their head and repeating it across standups and syncs, the assistants hold it and surface it on demand.

The part that doesn't get advertised

None of this removes the need for human judgment. Someone still has to decide what to build, resolve disagreements between teams, and own the outcome when a call goes wrong. What changes is how much of the plumbing around those decisions requires a person's attention. In our estimation, this pattern is most defensible at Town's current size — a self-organizing structure is easier to sustain with a few dozen people who already share context than across an organization spanning multiple offices and functions.

There's also a question the "self-organizing" framing tends to skip over: who's accountable when an assistant routes something to the wrong person, or misses that two teams are duplicating work? Platformer's interview frames Town as Grèze's own experiment in answering that — running the company on its own tooling before pitching it to anyone else.

What it means for AI builders

Town's setup is a reasonably clean signal for anyone building internal agent tooling right now. The tasks Grèze describes assistants handling fall into the same categories that show up in most "AI ops" pitches — status aggregation, meeting follow-up, decision logging — and they map closely to what current models are actually reliable at: retrieving and summarizing context, not making calls that carry accountability.

AiiN's takeaway

Grèze's account is one CEO describing how he runs his own company, not a controlled study. Still, it's a useful data point: a founder with real management experience is betting that assistants can absorb the coordination layer of a company, not just individual tasks within it. Whether that holds at Town's scale — let alone anyone else's — is the part worth watching next.