Your inbox
Reading it, triaging it, sending email for you, and cancelling what it finds. Below are 6 recorded examples of people doing exactly this with a messaging agent, with what each one shows and where it came from.
6 examples
A user had Muse email every data broker it could find with a removal request, using about two percent of his free weekly allowance, and says it replaces a paid removal service.
Why it is notable
Outbound email to dozens of strangers at once, on the user's behalf, for a job a subscription service charges for. The cost is stated as a share of the free tier, which two sightings now do.
Which surface was used is not in the post, so the channel is blank. Found through a crowd-sourced collection of Muse posts and read at the source; the words are the author's.
Posted by a user by Cuth (@ItsCuthulhu): x.com, 2026-09-11. Added 2026-09-17.
An investor had Muse go through his whole inbox and unsubscribe him from dozens of promotional lists, using about a tenth of the free tier's budget. He adds that Instinct is also good and did not cost a billion dollars, and asks what happens when Google ships its version.
Why it is notable
The only sighting that reports what a job cost against the free tier, which is the number Meta's page does not give. The inbox clean-up is the same job Instinct's users report, and the author says so.
Which surface was used is not in the post, so the channel is blank. Found through a crowd-sourced collection of Muse posts and read at the source; the words are the author's.
Posted by a user by Negligible Capital (@negligible_cap): x.com, 2026-09-10. Added 2026-09-17.
A user living with a severe illness that leaves him without energy used Muse to draft and send three detailed emails about a complex medical matter he could not assemble himself. Posted in reply to Meta's chief AI officer's $1,000 challenge, so it is an entry in the vendor's frame as well as a sighting.
Why it is notable
The accessibility sighting: an agent doing the writing for somebody whose illness takes the capacity to do it. Recorded in his own words and no further, because the condition is his to describe.
Which surface was used is not in the post, so the channel is blank. Found through a crowd-sourced collection of Muse posts and read at the source; the words are the author's.
Posted by a user by Martin (@MartinMoltke): x.com, 2026-09-15. Added 2026-09-17.
A user had 83 missing K-1 tax forms. He had Muse search his email to work out which were genuinely missing, forward the ones already sitting unread to his accountant, then track down whoever sent each prior year's form, email them for this year's, and copy the accountant in. He expects to intervene only on the ones behind portals.
Why it is notable
It is the largest piece of real paperwork any of these agents has been recorded completing, and it is chasing other people rather than filling in a form: eighty-three separate threads that a person would otherwise have to open one at a time. It also shows the access that requires, which is the whole inbox.
A user report, unverified by us. What is checkable is what the product states it can reach, which is on the listing.
Posted by a user by Justin Mateen (@justinmateen): x.com, 2026-09-09. Added 2026-09-09.
- InstinctiMessage
A user in San Francisco has Instinct watch the delivery email that lands in his Gmail Updates tab, where he never reads it. While he was away with family it told him a parcel had arrived, then offered to text the apartment he was staying in to ask whether someone could bring it inside.
Why it is notable
The interesting part is not the notification, it is what the agent said next. Asked whether the parcel had actually arrived, it separated the carrier's claim from the fact, said there was no delivery photo and no porch note, and told him what would settle it. An agent that volunteers the limit of its own evidence is doing something no features page describes, and it is the behaviour that decides whether an unprompted message is worth having.
The same post carries his verdict on the product, which is filed separately under what people say. This entry is the thing that happened; that one is what he made of it.
Posted by a user by Advait Marathe (@AdvaitMarathe): x.com, 2026-09-08. Added 2026-09-09.
A user built himself a swipe-through interface for unfollowing people on Instagram in about two minutes by asking Muse for it, and credits the native Instagram access rather than the model.
Why it is notable
It is the first recorded case of somebody using one of these agents to build a small tool for themselves rather than to run an errand, and what made it possible was not capability but access: the assistant is already inside the account the tool operates on, so there was no integration to authorise and no scraper to write. Every other listing here would need permission it does not have.
Filed as a sighting rather than as a capability claim: nothing on the product's own pages says it builds interfaces, and one person doing it once in two minutes is what we have.
Posted by a user by brett goldstein (@thatguybg): x.com, 2026-09-08. Added 2026-09-09.