Meta AI has launched the Muse AI agent, a personal assistant built to handle everyday tasks such as online shopping, sending emails and planning a trip, and shipped a companion app that puts it directly on iPhone. The launch is documented on Meta’s own site, and it is the company’s most direct attempt yet to put an autonomous agent in front of ordinary users rather than developers.
What the Muse AI agent can do
The distinction Meta is drawing is between an assistant that answers questions and one that acts. Once given a goal, Meta says Muse can work on its own: opening a browser, filling out forms and negotiating on a user’s behalf, rather than stopping to hand instructions back to the person who asked. The company’s own examples include selling a car and booking a plane ticket, tasks that involve several steps, real money and other parties, not just a single query and a single answer.
Wired reports that Muse is built to compete directly with agents called OpenClaw and Instinct, and that Meta has built privacy controls into the product because handing an agent access to forms, payments and negotiations raises the stakes on what it can see and do compared with a chatbot that only talks. 9to5Mac notes that Muse’s presentation looks similar to Grok Bot, the agent xAI ships inside its own apps: a dedicated persona with its own app icon, rather than a feature added to an existing assistant. For now the standalone app is iPhone only, sitting alongside rather than replacing Meta’s existing Meta AI app.
A pitch built on two different arguments
The Verge frames Muse as part of a multi-billion-dollar strategy overhaul at Meta, aimed at catching up to OpenAI, Anthropic and Google after a rocky run in the AI race.
Wired’s framing is different: its coverage centres on trust, on whether people will hand an agent the ability to act on their behalf, spend their money and fill in forms without supervision. Both things can be true at once, but they point at different risks. Catching up to rivals is a product and engineering problem Meta can solve with better models and more compute. Getting people to trust an agent with a car sale or a travel booking is a behavioural one, and it does not move at the pace of a release cycle.
That second problem is the one that has slowed every agentic AI launch so far, including OpenAI’s GPT-6 Astra upgrade to ChatGPT and Codex, which added similar autonomous task handling months before Muse arrived. An agent that fills out a form incorrectly, negotiates a bad price or misreads a step in a purchase is a different kind of failure than a chatbot giving a wrong answer, because by the time anyone notices, the agent has already acted.
What to watch
Muse is launching as an iPhone app first, with no confirmed timeline for an Android release. Whether Meta widens access, and how much it lets Muse act unsupervised before asking for confirmation on a purchase or a negotiation, will say more about how much the company trusts its own agent than any launch statement does.
Image: Ka Kit Pang via Openverse, licensed under CC BY-SA 4.0.
