The AI personal assistant just went mainstream. What that unlocks for the way you run your own health.

For most of the last two years, the ‘AI personal assistant’ has been a demo. A slick launch video, a waitlist, a founder on stage talking about agents. Meanwhile the actual work — the inbox at 700 unread, the calendar with three overlapping calls, the wearable app that pinged you sixteen times about nothing — kept happening to you by hand. That gap has been the interesting part of the story. It is starting to close.

The reason this month is worth writing about is a small San Francisco company called Town, which came out of beta at the beginning of June and has more than quadrupled its users since — not through a launch spectacle but by word of mouth between people who found it genuinely useful. It raised a $55 million Series A led by Justine Moore at Andreessen Horowitz in June 2026, and its founders are Jean-Denis Greze, formerly chief technology officer at Plaid, and Tony Vincent, previously director of applied AI Product at Google. The product itself is an AI assistant you name, give a form to — a bunny, a wombat, a pink axolotl — and then connect to your email, calendar, Slack, Drive and meeting notes. It then quietly starts drafting, briefing and booking on your behalf, and lets you press send.

why a bunny with your inbox is actually the story

Tony Vincent is old enough to remember the Tamagotchi — the 1990s digital pet whose entire job was to be looked after. That reference is not decoration. The reason Town spread beyond the technical class is that it inverted the direction of intimacy: instead of asking you to learn its interface, it asked you to name it, give it a shape, and put it on your shoulder. The result is that a plumber outside Sydney can hand a wombat named Eddie access to his accounting software and go on a Mediterranean cruise, and it feels less like operating a large language model and more like leaving instructions with a house-sitter.

This is the piece the health stack has been getting wrong. Most personal-health AI products have been built by technical people for technical users, and they read like it — dashboards, thresholds, alerts. The people who actually need the help most are not looking for another dashboard. They are looking for something to notice things on their behalf and hand back a sentence they can act on. Town’s taste is the interesting bit. The bunny is the wrapper.

what a Townie actually does, translated to a wellness stack

Once a user grants Town access to several surfaces, Justine Moore at a16z pointed out the specific behaviour worth stealing: the assistant ‘ingests all of your context, and then it suggests things for you.’ That is a subtle but important reversal. Most AI tools sit and wait for a prompt. A well-tuned assistant reads your surfaces and comes back with a small, quiet menu of things it noticed that you could do next. When your surfaces are health surfaces — wearable exports, lab PDFs, symptom notes, a journal — the same behaviour is powerful in a very different way.

In the reference article, Kirsten Green at Forerunner Ventures uses her Townie, Maxie, to prep meetings, review deals, and monitor 76 portfolio companies. Translate that pattern to the body you happen to live in: prep the next appointment (what actually changed since the last one), review this week’s data (did the sleep drift meaningfully or is that noise), and monitor the small handful of things you actually care about (resting heart rate, cycle regularity, one supplement you are testing). None of that requires exotic technology. It requires a system that reads your files and shows up with a short brief.

“The dream is not an assistant that answers when you ask. It is one that reads what you already collect and hands you the two sentences that make the week make sense.”

the ‘human at the edge’ principle — steal this exactly

The best sentence in the piece belongs to Shlanda Burch, director of people and culture at Madison Reed, who described her team’s workflow as ‘human at the edge’ rather than ‘human in the loop.’ It sounds like a small distinction. It is the entire ballgame for health.

‘Human in the loop’ means the AI does a step, you approve, it does the next step, you approve. It is slow, it is exhausting, and it quietly trains you to rubber-stamp. ‘Human at the edge’ means the AI does the whole loop end-to-end and hands you the finished thing to sign or send. Your judgement lives at the boundary where the work meets the outside world. That is the only defensible shape for anything that touches your body: a machine can compile your week, but you decide what leaves your hands — the message to your GP, the tweak to a supplement, the decision to keep training or rest.

the taste question, which is really the slop question

Vincent said the quiet thing out loud in the piece: he has ‘a strong aversion to AI slop’ and thinks taste is what most of the current AI product wave is missing. That is exactly the risk for health. Sloppy AI output looks like a confident paragraph, sounds like a specialist, and quietly averages your case into the median case. The failure mode is not that the AI refuses to help. It is that it helps a generic person while you are the one reading it.

The remedy is not to avoid the tools. It is to insist that the tools you keep in the stack are the ones that write like they know you specifically — your history, your medications, your priors — and that flag their own uncertainty when the population in the study did not include people like you. If a tool writes a health paragraph that could have been written about anyone, it is slop with better posture. That is a rejection criterion for anything you install.

what this changes for the 3-Layer Stack

Nothing in the underlying architecture. Everything in the ceiling.

The Research Layer is still where you turn a worry into a decent question and read the actual literature. The Ledger is still the file you own with your data, your history and your notes. The Protocol is still the small set of experiments you run on yourself and review. What changes is the assistant that sits across those three: it is now realistic, in mid-2026, for that assistant to read all three, produce a weekly brief, draft the message you were going to send, remember what you decided last month, and keep the whole thing quiet the rest of the time.

That is the interesting unlock. Not a new layer — a new operator across the layers.

what to do this week — whether or not you touch Town

1. Pick one health surface you keep in more than one place — the wearable app, the journal, the lab PDFs — and put a copy in a folder you own. The assistant, whichever one you eventually use, needs a single place to read from.
2. Write the one-page brief you wish landed on Monday morning by hand, once. Two sentences per surface: what changed, what to watch. That is your prompt template.
3. Decide the one edge where a human always stays: nothing goes to a clinician, into your body, or onto a shared calendar without your explicit send. Write that rule down.
4. Pick one assistant — the built-in one in your daily chat tool is fine to start — and give it the folder plus the template. Ask for the Monday brief. Judge it not by whether it impressed you but by whether it made your week two sentences quieter.
5. Review after four weeks. Keep it if the briefs got sharper. Retire it if it started producing confident paragraphs about a generic person. That is the taste test.

“The real question is not whether the personal AI assistant works. It clearly does. The question is whether you will build the small habit — single folder, one-page brief, human at the edge — that lets you use it on the surface that matters most, which is your own body.”

closing

The Inc. piece is about venture capitalists, plumbers and T-shirt printers naming a cartoon animal and handing it their inbox. That is the fun bit. The important bit is that a consumer-grade personal AI assistant is now real enough that you can build a stack around it without waiting another year. If you do, the interesting one to build first is the one that reads what you already collect about your own body. The tool will keep changing. The habit — single folder, weekly brief, human at the edge — is the part that stays yours.