Meta Muse UX review: what a synthetic audit found

Published September 29, 2026

Meta Musepersonal AI agentUX auditAI product onboardinghomepage UXsynthetic user testing

Muse is Meta's personal AI agent: you give it a goal and it carries out the task on a cloud computer of its own, from shopping to scheduling. Meta launched it on 8 September 2026. We ran a synthetic UX audit on the homepage, sending a persona suited to the site, someone weighing whether to hand an agent their email and calendar, through the page while it thought out loud.

What our persona found on the Muse homepage

• The page asks for your phone number before it shows you anything. The hero leads with a single field placeholdered "Mobile number or email" above a 408 by 44 pixel Continue button. There is no demo, no capability example and no feature breakdown above the fold. The only inline route to more detail is a "Learn more" link roughly 63 by 13 pixels with no container around it. A visitor deciding whether to trust an agent with their accounts is asked to identify themselves first and find out second. Heuristic: NN/g heuristic 6, recognition rather than recall.

• The SMS disclosure sits below the field it applies to. The line "You may receive SMS notifications from us by using your mobile number." appears only underneath the input, after the point where someone scanning the form has already started typing. It is the only statement on the page about what giving a phone number signs you up for. Heuristic: NN/g heuristic 1, visibility of system status.

• Forty-five unlabelled icons carry the product's scope. Under the heading "Built around your whole life", a nine by five grid of 3D icons runs with no labels, captions, legend or category headers. The spacing is uniform in both directions, roughly 121 pixels between icon centres across and down, with no tint, rule or container grouping one row against another. The copy above names four areas: saving money, improving your health, smart shopping and strengthening your relationships. The reader is left to map forty-five pictures onto four stated areas without a caption to connect them. Heuristic: Gestalt, proximity and grouping.

• The security explanation is written in infrastructure vocabulary. The section on where the agent works reads "Muse Secure VM is a persistent, isolated Linux computer with a full browser that you and your agent can use to complete tasks", and goes on to mention storage, CPU and memory. The reassurance a cautious visitor is looking for is real, but it is expressed in the language of a server spec rather than of the person deciding whether this is safe. Heuristic: NN/g heuristic 2, match between the system and the real world.

One thing Muse does well

The ticket-booking mockup shows the product working rather than describing it. Muse messages that tickets have opened for a film the user is tracking and offers two times. The user replies "Yeah, book for 2" and "Lets do 7:30pm". A Browser card then shows seats being selected on a live seat map, with an "Open browser" button directly beneath it. That button is the thing the body copy promises, a way to step in at any point, and it is shown rather than asserted.

What this audit covered

The run captured three viewport sections of the homepage: the sign-up hero, the capability grid and the cloud-computer section. It did not exercise sign-up, the mobile apps or any signed-in surface, so it says nothing about those.