Instinct, Town and the new wave of personal AI assistants: what they are and whether you should try them

If you follow technology news, you will have seen two names come up a lot over the summer. Instinct, a San Francisco startup whose product almost nobody can use yet, raised money at a $2.5 billion valuation in late August. Town, another San Francisco company, raised $55 million in June and has been spreading through the inboxes of Silicon Valley executives, mostly by word of mouth. Both are described as "personal AI assistants" or "consumer AI agents", and both are already being talked about as if they were the next thing after ChatGPT.

What is different about these tools

For the last three years, most people's experience of AI has been a chat bot. You type a question, you get an answer, and then you go and do something with that answer yourself. ChatGPT, Claude, Gemini and the rest are extremely good at this, but the work of actually sending the email, booking the flight or moving the meeting still sits with you.

The new assistants are built around a different idea. You connect them to your accounts (email, calendar, messaging apps, sometimes your files and your phone), you talk to them the way you would talk to a human assistant, and they go and do the thing. Reply to that email. Find a plumber and book them. Cancel the subscription. Prepare a briefing on the person I am meeting at 2pm. The pitch is a ‘chief of staff’ for people who would never have had one.

The reason this is happening now, rather than in 2024, is partly that the underlying models have become good enough at multi-step tasks to be trusted with them.

And also a wave of enthusiasm that started with a hobbyist project. In late 2025 a developer called Peter Steinberger released an open-source personal assistant, eventually named OpenClaw, that ran on your own computer and could be reached through WhatsApp or Telegram. It became a phenomenon among technical people, and Steinberger has since joined OpenAI to work on agents there. OpenClaw was never something a normal non-technical person would install, but it proved there was an appetite, and Instinct and Town are two of the attempts to package the same idea for everyone else.

Instinct

Instinct is the one with the eye-watering numbers. According to Forbes, the operating company was only registered in California in April this year, by a 23-year-old former Sierra researcher called Noah Shinn. It went from a $100 million valuation to over $2.5 billion in a matter of weeks, through a $75 million Series A led by Kleiner Perkins and then a $250 million Series B co-led by Index Ventures and Benchmark. The product has been in private beta since February. There are two ways in: an invitation from someone who already has access, or the public waiting list on the company's website, which the company says it is working through as it adds computing capacity. Nobody has said how long the queue is, and early on the invitations were circulating mainly among investors and their friends. It is free for now.

What it does, as far as anyone outside the beta can tell, is act as an assistant you reach by text message or phone call. You connect it to Gmail, your calendar, your messaging apps and, on your phone, potentially your location, audio and screen. Early users have posted examples of it drafting replies, arranging travel, coordinating home services and following up on conversations that had gone cold. Shinn has said one user relied on it through the process of buying a house. The website is deliberately bare, and the company has said very little in public beyond confirming the funding.

The attention has come with scrutiny. TechCrunch reported that Instinct's terms of use give the company broad rights to retain user data, potentially for training its models, and that one early user found copies of her emails were still being stored after she had disconnected her Google account. Shinn responded that the company was taking the concerns seriously and had shipped updates to reduce the risk of the assistant acting outside what users intended. Forbes has also reported that the business models under consideration include either a paid subscription or an advertising model that would draw on what the assistant sees of your communications and spending. Nothing has been decided.

So the honest summary of Instinct today is: a very capable-sounding product that almost nobody can access yet, from a company that is a few months old, with a valuation that reflects investor excitement rather than any evidence of a working business. That is not a criticism, it is just where things stand today.

Town

Town is the more approachable of the two, and the one you can actually sign up for today. It was founded in late 2024 by Jean-Denis Grèze, previously chief technology officer at Plaid, and Tony Vincent, previously a director of applied AI at Google. Its first product was actually a tax and compliance service for small businesses; the personal assistant came later. In June it raised a $55 million Series A led by Andreessen Horowitz, and by July Inc was describing it as Silicon Valley's new favourite tool. As this is written, it is reported to be closing a further $90 million at a valuation of around $1 billion, co-led by Forerunner and Menlo Ventures. Nothing is confirmed by the company, but the numbers give a sense of how much money is chasing this category at the moment.

The product design is where Town has made its bet. Your assistant is a "Townie": you give it a name and a character (Grèze's is a silver fox called Ivy, a Fortune journalist's is a rotund furry creature called Algernon), and the whole thing is meant to feel closer to a friendly helper than a piece of enterprise software. Underneath the cuteness it is a fairly serious work tool. It connects to email, calendar, Slack, Zoom, Google Drive, meeting note apps and a long list of other services. It drafts in your writing style, triages your inbox, prepares briefings on the people you are about to meet, and runs "routines", which are recurring jobs that trigger on a schedule or an event: a morning summary at 7am, a weekly client recap, a note whenever a particular kind of email arrives. There is an iOS app for approving actions from your phone, and the company's own copy leans heavily on the phrase "with your oversight".

Town is also clear about what it costs. The pricing page lists a free tier limited to 30 chats a month, then paid plans at $15, $49, $99 and $199 a month, each with a monthly allowance of credits that you consume as the assistant does work. Heavy users will need to watch the credit meter, and third-party breakdowns put the realistic annual cost of the $49 Pro plan at somewhere between $600 and $750 once overages are included. That is not cheap for an individual, but it is of course considerably less than even the cheapest human assistant.

The case for trying one

The appeal is easy to see. Most knowledge workers spend a meaningful share of their week on coordination rather than on the work itself: finding a time, chasing a reply, summarising a call, looking up who someone is before a meeting. An assistant that handles even half of that, and does it in your own voice, is worth something. Early users of both products describe the difference between an AI that talks about your day and one that actually runs it.

There is also a compounding effect. The longer one of these tools watches how you work, the better it gets at anticipating you, which is why investors keep saying that the ‘moat’ in this category is the accumulated context rather than the underlying AI model. For an individual user that translates into a tool that improves with use, which is actually quite rarer in software.

And the entry cost is low. Town's free tier is enough to find out whether the idea suits you, and when Instinct opens up, it is free too, at least for now.

The case for caution

The obvious one is access. To do its job, a personal assistant needs to read your email, see your calendar, know who you are talking to and, in Instinct's case, potentially see your screen and hear your surroundings. That is a great deal more than you have ever given a piece of software, and it is worth being very clear-eyed about the fact that a small startup now holds it. The Instinct data retention episode is a reminder that the terms of use matter, and that "disconnect" does not always mean "delete".

The second is that these products act, and actions can go wrong. A chatbot that gives you a bad answer costs you a few minutes. An assistant that sends the wrong email to the wrong client, or books the wrong flight, costs a lot more. Both companies have built approval steps for this reason. The sensible stance is to treat these tools as very junior staff for the first few weeks: useful, quick, and not yet trusted with anything irreversible.

The third is the business model. Town has one, and it is the familiar subscription-plus-usage model. Instinct does not yet, and one of the options on the table is advertising informed by what the assistant can see of your life. That would be a fundamentally different product from the one early users are enjoying for free, and anyone getting attached to it now should keep that in mind.

Finally, there is the question of whether the incumbents simply do this themselves. Google, Microsoft and Apple already sit on the email, calendar and phone that these assistants need to connect to, and OpenAI has hired the person who started the whole wave. Grèze's response, when Fortune put this to him, was that Superhuman competes directly with Gmail and has not been locked out, and that Notion built a large business on top of Google Docs. That is a fair point, but it is also the argument every startup in a platform's shadow makes, and some of them turn out to be right.

What this looks like if you run a small business

Most of the coverage of these tools is written by and for people in large technology companies, but the case for them is arguably stronger in a small business, where the owner or a small team of managers are the coordination layer for everything. Nobody in a 15-person firm has an assistant. The person who runs sales also chases invoices, books the van in for its service and replies to the supplier at 10pm. That is exactly the kind of fragmented, low-stakes, high-volume work these assistants are built for, and at $15 to $49 a month, a subscription is a rounding error against the hours involved.

The risks are also more concentrated. In a small business the owner's inbox usually contains everything: customer data, staff matters, bank correspondence, the accountant, the lawyer. Connecting it to a third party is a bigger decision than connecting a product manager's inbox at a large company, where sensitive material lives elsewhere and there is an IT department keeping an eye on things.

A small business has less margin for the assistant getting something wrong with a customer, because each customer matters more. The practical response is to be very deliberate about the scope: connect a shared or secondary mailbox rather than the owner's personal one, keep the assistant on tasks where a mistake is embarrassing rather than costly, and keep approvals switched on for anything that leaves the building.

Two examples of what using these tools might look like:

• A small marketing agency owner sets up a routine that, every morning, pulls together the day's meetings with a short note on each external attendee (who they are, what was last discussed, any open items in the email thread) and flags emails from clients that have gone unanswered for more than two days. Nothing is sent without approval. The owner still writes every reply, but stops losing half an hour each morning working out what the day is and stops letting client emails slip through. That is a modest, repeatable saving, and it is the kind of thing an assistant can do reliably today.
• The second is a trades or services business, a plumbing firm or a small cleaning company, that lives on scheduling. The assistant sits on the bookings inbox, drafts confirmation and reminder messages in the firm's usual wording for a human to approve and send, moves appointments when a customer asks and the calendar allows, and produces a Friday summary of the week's jobs, cancellations and anything still unconfirmed. It does not quote prices, does not commit to a time without a person checking, and does not touch payments. That leaves the interesting decisions with the owner and takes away the part of the job that was mostly typing.

The businesses that get value from these tools in the next year will be the ones that treat them as a way to remove a specific, well understood chunk of admin, and then widen the remit as trust builds, rather than the ones that hand over the whole operation on day one.

Should you try one?

If you are an individual who spends a lot of time coordinating, and you are comfortable connecting a work email account to a young company, Town is the reasonable place to start. Use the free tier, connect one account rather than everything at once, keep the approval steps switched on, and see after a fortnight whether it has saved you more time than it has cost you in checking its work. Many people will find the answer is yes for a narrow set of tasks and not yet for the rest.

If you are doing this for a business rather than for yourself, remember that the data these tools need is not only yours. It belongs to your clients, your staff and your suppliers, and your obligations to them do not change because the software is clever. Read the terms of service properly and check what the data protection rules in your jurisdiction say about handing communications to a third party before you connect anything. This is a category where being six months late costs very little and being six months early can cost quite a lot if it goes wrong.

As for Instinct, the practical advice is simple: unless you know someone with an invitation, you can join the waiting list and wait. By the time your turn comes, it will have a business model, a clearer privacy position and either a much larger or a much smaller amount of enthusiasm around it.

The broader point is that this is almost certainly what the next phase of consumer AI looks like: less about a clever answer in a chat window and more about software that has your permission to act. That shift is going to happen whether it is Instinct, Town, or something from a company you already pay. Getting a feel for it now, carefully, is not a bad use of an afternoon.