Skip to content
Messaging Agents

Restaurants

6 public posts of somebody having an AI agent they text handle restaurants, from 6 people, on Muse and Instinct. 2 went wrong.

A table is the job the vendors demo, and it is also where the failures on this site sit. On 8 September a Muse recommended Bar Agricole and tried for seven minutes to book it; the restaurant had closed two years earlier. Two days before that an Instinct user's Resy account was deactivated and every future reservation cancelled, because his agent had been sweeping one restaurant's availability every ten minutes around the clock on his account, and he published the log and said the platform was right. The successes are the head-to-heads: the same dinner given to Muse and Instinct twice in one week, Muse finishing first both times through its own payment rail, one user going with Instinct anyway because it drove his own browser where his card was saved. Whose account the agent books on, and whose terms it is bound by, is the question this page keeps asking.

The posts, by agent

Muse5

  1. A founder gave Muse and Instinct the same job at the same time: book a restaurant and pay the reservation fee. Muse got through faster; he went with Instinct because it let him use his own browser, so his Link payment details were already filled in, where Muse's in-app browser was not.

    Why it is notable

    The rare side-by-side, and the difference he found was the browser rather than the model: whose browser the agent drives decides whether the person's saved payment details are there. That is the row this site records as the browser approach, and it decided the outcome.

    Posted by a user by Alex Kwon (@startupoppa): x.com, 2026-09-10. Added 2026-09-17.

  2. An investor asked where to go for dinner. Muse recommended Bar Agricole and tried for seven minutes to book it, and could not get through, because the restaurant closed two years ago.

    Why it is notable

    The clearest failure sighting on Muse's record: a recommendation from stale data, followed by seven minutes of persistence against a place that no longer exists. Worth having beside the successes because it shows what the agent does when the world and its sources disagree, which is keep trying.

    Posted by a user by Sheel Mohnot (@pitdesi): x.com, 2026-09-08. Added 2026-09-17.

  3. A founder's day across three agents: Hermes reordered his contact lenses and drafted the family schedule, Instinct found and ordered a replacement showerhead, and Muse booked an anniversary dinner, driving the browser inside the thread.

    Why it is notable

    Three listings from this directory in one person's day, each given a different errand, which is the clearest picture yet of how a heavy user splits the work. Filed on Muse for the booking; the other two are on their listings' records by reference.

    Posted by a user by cbruce.eth (@cbruce_eth): x.com, 2026-09-14. Added 2026-09-17.

  4. Somebody with early access ran Muse across their business and personal admin and reported it paying bills, ordering flowers and booking date nights, describing it as the thing an earlier open-source assistant was supposed to be.

    Why it is notable

    It is the first account of one of these agents doing several unrelated chores for the same person in a week rather than demonstrating one. The three named jobs cross payments, a purchase and a reservation, which are three different kinds of authority to hand over, and no product page separates them.

    Posted by a user by Marshall Haas (@marshal): x.com, 2026-09-08. Added 2026-09-09.

  5. A sceptical user gave Muse and Instinct the same job, researching tickets for a Rosh Hashanah dinner. Muse completed the booking through its payment integration while Instinct was still searching for the RSVP.

    Why it is notable

    It is the only head-to-head anybody has run in public with one prompt and two agents, and the thing that decided it was not reasoning: one had a payment rail wired in natively and the other did not. That is a field on these listings rather than a matter of model quality.

    Posted by a user by Alex Volkov (@altryne): x.com, 2026-09-08. Added 2026-09-09.

Instinct1

  1. iMessage

    A user's Resy account was deactivated and all of his future reservations cancelled after what the platform called behaviour in violation of its terms. The cause was his own agent: he asked Instinct for its activity log and it reported roughly 200 API requests an hour around the clock, an availability sweep of one restaurant every ten minutes since Friday morning at 17 to 19 calls a run, and a two and a half minute burst at the 9am table drop polling every 0.4 seconds.

    Why it is notable

    It is the first recorded case of one of these agents getting its owner banned from a service, and the numbers make it plain that nothing malicious happened: the agent did exactly what it was asked, continuously, at a rate no person could produce. The user's own verdict was that the platform's action was fair. Every listing here that books a table is doing some version of this, and not one of them publishes what it does on your behalf while you are not watching, which is why the activity log is the most useful thing in the story.

    Posted by a user by JC Bahr-de Stefano (@jbahrdestefano): x.com, 2026-09-06. Added 2026-09-09.

All use cases