Meta spent years talking about an agent that would do more than answer questions. On Tuesday it shipped one. Muse is a personal assistant that sends email, books travel, builds grocery lists from recipe videos, and can check out with Stripe. It is free for most people. Power users pay $20 a month. The ones who hand it their whole week pay $100.
The agent runs inside a dedicated virtual machine in Meta’s cloud, with a browser the user can see. It connects to Facebook and Instagram, and to Google Workspace, OpenTable, Spotify, Ticketmaster, and Apple Health. If a service has a public API, Muse can wire itself in with credentials you provide. If it does not, it uses the browser. Vishal Shah, Meta’s vice president of AI products, told reporters the system can also write software of its own. “It isn’t really limited in what it can do,” he said.
That is the pitch and the problem. Less than two weeks after Meta agreed to an $18 billion multistate settlement over social media’s harms, the company is asking people to give it email, calendars, payments, and health data. There is no advertising inside Muse, according to chief AI officer Alexandr Wang. Commerce is on the table.
You can name the agent, give it an avatar, and decide how much it may touch. You can message it in a dedicated app, on the web at muse.ai, or in WhatsApp. Support for Meta’s glasses is promised “soon,” which in this company usually means the next Connect.
The comparison set is not ChatGPT. It is the messy category of computer-using agents that have sprung up this year — OpenClaw, Instinct, and a dozen smaller tools that drive a browser like a temp with a company card. Meta’s advantage is distribution: billions of people already keep a Meta app on the first home screen. Its disadvantage is the same fact.
Wang said most people should never need to pay. The subscriptions exist to cover compute for the heavy users. That is a familiar sentence in consumer AI. The unfamiliar part is a social network asking to operate your life rather than occupy it.
