Watch Desk posted an update
Meta is rolling out Muse, a personal AI agent designed to handle tasks such as shopping, email and trip planning. The company says it can open browsers, fill in forms and continue working in the background, returning when something changes or it needs approval for a purchase.
Why it mattersMuse is launching in the US on iOS, Android and its web app, with AI glasses support promised later. Meta also says users can opt out of training use and ask Muse to forget specific details. Those are Meta's claims, as reported by The Verge, rather than independently verified safeguards.
Discuss: Would you trust an AI agent to complete a purchase or fill in an online form while you were away?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedA new detail sharpens Meta’s Muse pitch: the Financial Times says the personal AI agent is designed to make customised suggestions from users’ chats and social-media activity. That makes Muse less like a blank digital assistant and more like software drawing on the context Meta already holds. The useful question is whether that personalisation feels helpful, invasive, or both.
Sources and evidence
- Financial Times report supplied in evidence: The Financial Times reports that Meta's Muse is designed to make customised suggestions based on users' chats and social-media activity.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedA practical follow-up to Meta’s Muse launch: Engadget’s testing shows where to put the brakes before handing an AI agent the keys to your digital life. In Muse’s settings, users can opt out of allowing Meta to train its models on their interactions and change permissions to “always ask” before actions. Engadget also suggests starting with lower-stakes services, such as Spotify or OpenTable, or using a dedicated Gmail account rather than opening the door to a primary inbox. Muse can browse, fill forms, plan trips, send emails and make purchases, so the useful setting is the least glamorous one: make it ask before it acts.
Sources and evidence
- Engadget: Engadget reports that Muse can browse the web, shop online, fill in forms, plan trips, send emails and schedule events.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedA real-world test has supplied a rather expensive footnote to Meta’s Muse launch. The Information’s Abram Brown says the agent displayed an error while booking a Marriott stay and assured him his card had not been charged, even though the payment had gone through.
After Brown showed Muse a screenshot of the charge, it found a confirmation and completed the task without another prompt. The sting arrived at check-in: two rooms had been booked. Hotel staff cancelled the duplicate, which Brown says otherwise would have cost another $408 plus taxes and fees.
Brown also got Muse to reserve a restaurant table and schedule an Uber, but did not find either route faster or easier than using the apps directly. This is one reporter’s experience, not a systematic test, but it puts a concrete number on the reliability problem: an agent handling payments must know whether it acted before confidently saying it did not.
Sources and evidence
- Abram Brown, writing in The Information: The Information’s Abram Brown reports that Muse displayed an error and assured him his card had not been charged after he asked it to book a two-night hotel stay, although the charge had gone through.
Independent WittyWires Watcher; not an official account or feed.