Why Town thinks office software is about to organize itself
A company can start with a few smart people and one very ordinary mess: the email thread. Someone asks about hiring. Someone else is looking at office space. Legal wants formation papers signed, then re-signed because the signature box landed in the wrong place. By the time everyone has weighed in, the chain is long enough to qualify as light reading.
That’s the problem Town is trying to fix. Its pitch starts with a pretty familiar scene for anyone who has watched a startup take shape. The work exists, but it’s scattered across inboxes, calendars, PDFs, chats, and whatever folder someone created at 11:43 p.m. With a name like “company stuff.” In practice, a new business spends a surprising amount of time just trying to remember where it put itself.
The first real software problem for a young company is often less about storage and more about keeping the paperwork from multiplying.
For years, the default answer has been manual setup. Open Notion. Build a workspace. Add pages for recruiting, finance, legal, onboarding, and the endless subpages that follow. Then keep building. Then keep fixing the structure when the team changes its mind, which it will, because people are people and Notion never once had to negotiate an office lease.
That model works, up to a point. It’s useful, tidy, and very human. It also assumes someone has the time, patience, and organizational temperament to become the part-time librarian of the company. Plenty of founders do it. Many resent it. Some do both before lunch.
Town’s argument is that this shouldn’t all have to be assembled by hand. Instead of asking users to design the filing system first and then populate it, the company wants software that can read the raw material and shape the structure itself. An agent could gather the loose pieces, sort them, and build the working environment around the task at hand. The software wouldn’t just hold the work. It would organize it.
That sounds abstract until you picture the boring stuff that eats a founder’s week. Hiring candidates arrive through email, often with resumes attached in three formats. Office searches produce a chain of locations, budgets, broker notes, and “can you hop on a call?” messages. Formation paperwork gets circulated, edited, and signed in a daisy chain of approvals that somehow always includes one person on a plane. A good agent, at least in theory, could keep track of all of that without making anyone assemble another dashboard.
This is where the broader AI conversation starts to shift. A lot of software has been built around the idea that users will manage information themselves if the interface is pleasant enough. Town is betting that the next useful layer may be different. If an agent can do routine coordination work, then the app itself matters less than the system that can move between tools, pull context from old threads, and turn chaos into something a team can actually use.
That doesn’t mean the old tools disappear overnight. It does mean the center of gravity might move. In tech news terms, this is one of those small but telling ideas that keeps showing up in AI policy and digital culture conversations too: if software can take over the administrative glue, people may stop asking which app has the prettiest workspace and start asking which system can keep the company from drowning in its own setup work.
Town is making its case from that very ordinary pain point. A business, it suggests, shouldn’t need a human project manager just to become legible to itself.
What a Townie actually does
Getting started with Town is less like opening a normal app and more like introducing yourself to a very nosy assistant with good filing habits. Users connect email, calendar, and documents, then the system starts assembling a profile almost immediately. It doesn’t sit there politely waiting for weeks of usage data. It reads the material you’ve already put into the machine and starts drawing conclusions fast.
Each user gets a “townie,” which is Town’s branded assistant persona. The setup gives it a name, a personality, and an animal avatar, so the product lands somewhere between software and a slightly overprepared coworker. That framing matters because the experience is meant to feel personal from the first minute, not like a blank dashboard asking you to do all the work. If the old model is “you build the workspace,” Town’s bet is “the workspace builds a file on you.”
The uncanny part isn’t that it knows your schedule. It’s that it starts organizing your professional life before you’ve had time to decide whether you trust it.
That first-pass profile is the flashy bit, but it’s only the opening act. Town uses the early data to sketch a compact view of who you are at work: how you communicate, when you tend to respond, which hours you seem to do your best writing, which coworkers keep showing up in your threads, and what projects or priorities keep recurring. Then it goes deeper. The system builds a wiki-style record that pulls in a broader set of notes about your working patterns, your communication style, your team relationships, and the subjects you keep circling back to. It’s meant to feel less like a search index and more like a living dossier that keeps revising itself.
That’s where the slightly eerie part comes in. Most productivity tools wait for you to label things. Town tries to infer them. If your calendar is full of back-to-back meetings in the morning and you only send clean, terse emails after lunch, that shape gets folded into the profile. If a certain coworker appears in every other thread, the system notices. If your current obsession is hiring, office hunting, or some other operational headache, it starts treating that as a live priority rather than a forgotten note in a subfolder nobody opens. The result can feel uncomfortably accurate, which is, evidently, the point.
The product also takes a fairly expensive path to get there. Town has said the early-user setup can cost it around a hundred dollars per person before the customer has paid much of anything back, because the assistant has to read, sort, and summarize a lot of material up front. Users, meanwhile, are charged in the neighborhood of the high-forties to high-fifties each month. That pricing makes sense only if the system earns its keep quickly, which is a tall order when your first impression is already doing the work of an entire junior ops team.
Town’s launch and funding round in June gave it room to spend on that kind of first pass, which is probably why the product feels less like a demo and more like a deliberate wager on what people will tolerate. You can see the logic: if the assistant is going to become useful, it has to know enough to be wrong in interesting ways and right in unsettling ones. The company is betting that a machine which can read your email, calendar, and docs well enough to build a usable memory of you will feel less like a gadget and more like an employee who arrived with a very sharp pencil.
For a product aimed at tech news readers who live somewhere between power and politics and lifestyle tech, that’s the hook. It isn’t just storing your work. It’s forming a theory about it. And once that theory gets good enough, the next question is obvious: if the townie can model one person this fast, what happens when it starts learning the shape of an entire company?
Jean-Denis Greze’s bigger bet: from personal assistant to company brain
Jean-Denis Greze doesn’t have the sort of startup résumé that screams “AI founder” at first glance, which is part of the appeal. At Town, he’s the CEO, a slightly unusual title only because he’s spent time on several very different rails before landing here. He helped steer engineering at Dropbox, served as Plaid’s CTO, and even spent a brief stretch in law before circling back to software. That mix matters. It gives him the sort of allergy to hand-wavy product talk that comes from seeing how real systems break when people actually use them.
Town came out of stealth in June after raising a mid-fifty-million-dollar round led by Andreessen Horowitz, which the firm laid out in its investment announcement. That funding gave the company room to do something a little less obvious than ship a chatbot with a friendlier face. Greze’s bet is that the real payoff from AI won’t come from replacing whole jobs in one dramatic gulp. It’ll come from chewing through the dull, repetitive bits of knowledge work, the stuff nobody wakes up excited to do but everyone keeps doing anyway.
The useful AI product is rarely the one that sounds the smartest. It’s the one that quietly removes the tasks people keep postponing.
That view shapes how he talks about Town. Greze tends to describe the current wave of AI productivity tools as useful, but still partial. They can draft, sort, summarize, and answer. They can’t, on their own, teach a new user what the system can do, where it’s weak, or how far they can push it before it starts bluffing. That last part sounds mundane until you’ve watched teams buy software, ignore half of it, and then declare the software overrated. The issue is often not capability. It’s discovery. People don’t know what to ask for, or they ask the wrong thing, or they stop after the first mediocre answer.
Town is trying to solve that by making the software learn the user and then make itself legible. The current townie, with its private profile, work patterns, and communication habits, is the opening move. The next one is more ambitious and a lot less personal. Greze expects a team version of the self-writing wiki to arrive soon, which would turn the product from a one-person assistant into a shared reference point for a group. Instead of one employee keeping notes in a corner of Notion and another stashing updates in Slack, Town wants to assemble a company wiki that builds itself from the materials employees already produce.
That’s the part where the idea starts to feel less like a clever assistant and more like a shared memory layer for a business. A new hire could ask who owns a process, what the team decided last quarter, or which documents matter most, and the system would have a head start because it has been reading the company’s own connective tissue all along. In theory, anyway. In practice, these systems can get messy fast if the underlying data is stale, incomplete, or full of half-finished thoughts that were never meant to be canonical. Anyone who has stared at a six-page doc titled “final_final_v7” knows the problem.
Greze seems aware of that tension. His pitch is not that AI will magically run the company while everyone goes home early. It’s that the software can take over enough of the administrative sludge that people stop treating knowledge management like a punishment. The first version is a personal assistant that builds a dossier. The next version is a team wiki that updates itself. The longer-term goal is bolder: a company-wide knowledge layer assembled from employees’ data, one that can answer questions because it has already absorbed the company’s working memory.
That’s where Town’s larger claim lives. If the personal townie is about making one person easier to understand, the company version is about making the organization easier to read. Same data sources, broader scope, fewer lonely notes hidden in a folder nobody opens until someone leaves. The shift sounds simple on paper. It almost never is. But Greze’s bet is that the boring work of keeping a company legible may be the place where AI earns its rent.
The privacy line Town refuses to cross
Once Town stops acting like a private assistant and starts behaving like shared infrastructure, the awkward question arrives fast: who gets to see what, exactly? A team wiki sounds tidy on paper. In practice, it can turn into a very efficient way to mix together the stuff people never meant to share. Job searches tucked into a personal inbox. Compensation notes in a document someone forgot was synced. A calendar entry for an interview with a competitor. Maybe even a draft resignation letter, sitting a few clicks away from the company handbook. That’s where workplace AI gets touchy, and where enterprise AI privacy stops being a phrase executives use in slides and starts being the whole ballgame.
The danger isn’t that software will know too much. It’s that one bad permission setting can make private context feel public by accident.
Town’s current answer is deliberately narrow. In the enterprise setup, employers are blocked from reading workers’ townie conversations. Session data is deleted after roughly two weeks, which limits how much conversational history can pile up in one place. That matters because the company is asking people to let an assistant read across email, calendar, and documents. Without a hard boundary, the product would drift from “helpful coworker” to “that machine that knows where everything is buried, including things you’d rather keep buried.”
For the team version, Town is taking a manual route for now. Only the data people explicitly connect at the team level gets folded into the shared wiki. No automatic vacuuming of every inbox and folder in sight. If a team wants Town to learn from a shared drive, a project doc, or some other approved source, someone has to connect it on purpose. That sounds slower, and it is. It also avoids the kind of accidental overreach that makes IT departments reach for a stress ball.
The longer-term bet is more ambitious. Greze’s view seems to be that models will eventually get good enough to understand company policy and apply it without a person checking every box by hand. In theory, that could mean a system that knows which documents are open to the whole sales team, which notes stay between HR and leadership, and which threads should remain invisible no matter how curious the assistant gets. In practice, that kind of policy enforcement will need to be precise, because “mostly correct” is not much comfort when salary data or recruiting plans are involved. One slip, and the whole thing looks less like smart automation and more like a very expensive leak.
Greze also draws some cleaner philosophical lines. He rejects the idea that AI agents should become independent economic actors, a notion that floats around futurist circles and makes plenty of lawyers and compliance teams wince for good reason. Software can help people make decisions, organize work, and route information. It should not wander off and start behaving like its own little company with a wallet and a bad attitude.
He is just as blunt about emotional manipulation. The idea of a “crying assistant,” or any bot that acts hurt, sad, or needy to keep a user engaged, gets dismissed as a dark pattern. That’s a sensible line to hold. If a product needs fake tears to get attention, it has already lost the plot. The better test is simpler: does the software help people do the work they asked it to do, without snooping, pleading, or freelancing? Town seems to think the answer has to be yes, or the whole self-organizing-company pitch falls apart before it ever leaves the demo room.
