Town's real product is not the assistant. It is the org chart it writes about you.

Town, an AI assistant company that left stealth in June after raising $55 million from Andreessen Horowitz, told Platformer that a team version of its self-writing wiki is "coming out very soon": a company knowledge base that assembles itself from what individual assistants already know ^[1][5]. The consequence for buyers is not another assistant with a personality. It is that the unit of software shifts from the document to the coordination, and coordination is much harder to put on a checklist.

What exists today is narrower. Town sells an assistant, a "townie," that connects to calendar and email, briefs you on your day and your meetings, and takes limited action ^[2]. Onboarding builds a dossier of who you are, what you do and which coworkers you deal with most, then writes a personal wiki covering your profile, communication style and work patterns ^[3]. Platformer's writer reports that within minutes it had worked out he was launching a company and named priorities including studio space, a network deal and a launch episode ^[4].

The competitive claim sits in the comparison, not the demo. Small businesses commonly organize this work in something like Notion, a blank page filled with boxes, tables and databases ^[13]. Platformer's writer set one up, pushed as much as he could to Notion's embedded agents, said it worked well enough, and still came away feeling he had given himself a large new project to maintain ^[14]. That is the wedge: the running cost of a workspace is maintenance, and the pitch is that an agent absorbs it. Replit's Amjad Masad made the wider version of the argument on the same podcast, predicting apps are entering a long twilight as work gets handed to agents ^[15].

Price puts a number on the claim. Most Town users pay $49 to $59 a month ^[7], or roughly $588 to $708 a year per seat ^[8]. Chief executive Jean-Denis Greze, whose title at the company is "mayor" and who spent about seven years as chief technology officer at Plaid, cited a customer who told him it beat her $75 monthly manicure on time saved and was "worth $600 a year for me to not have to do those things" ^[6][9]. He called that product-market fit and said "I live downstream of capitalism" ^[9][10]. It is one anecdote, relayed by the founder, about a consumer seat, not a team deployment.

The team wiki is where the evaluation problem starts. A checklist cannot test whether a system models your company correctly, so the real diligence becomes: connect mail, calendar and chat, wait, read what the thing believes your priorities and reporting lines are, and count the errors. Greze named the failure mode himself, saying that pooling private files too carelessly could put "people's salaries in a spreadsheet by mistake," and "then you've got big egg on face" ^[11]. His forecast, according to Platformer, is that within five years specially trained models will be trusted to enforce company privacy policies, and that small high-trust companies will adopt unified knowledge bases well before large enterprises ^[12].

Watch whether the team version ships with a stated permission model rather than a launch post, and whether anyone publishes a way to correct a wrong inference and see who read it before the fix. Watch retention at roughly $600 to $700 per seat per year ^[8], since the same tools that write your wiki also make the switching cost look like an export. And watch Notion, which already has agents inside the blank page ^[14].