Meta is launching Muse, a personal AI agent that it says can carry out multistep tasks such as shopping, sending emails and planning trips. The product will roll out in the US on iOS, Android and the web at muse.ai, with support for Meta’s AI glasses planned for later.
The launch is a notable shift in emphasis: rather than positioning AI primarily as a chatbot or creator tool, Meta is pitching Muse as software that can act. Users give it a goal through the Muse app or WhatsApp, and the agent can open a browser, fill in forms and continue longer-running tasks in the background. Meta says it will return for approval when circumstances change or before actions such as a purchase.
Muse is powered by Meta’s in-house Muse Spark model and is the centerpiece of the company’s broader effort to regain ground against OpenAI, Anthropic, Google and Microsoft.
What changed
The core proposition is autonomous execution for a mass consumer audience. Meta says Muse requires no technical experience and can retain details users share to make suggestions over time. That matters because many current agent products are marketed toward workplace users, developers or technically inclined early adopters.
Meta will offer Muse free to most users, while planning unspecified paid subscriptions for heavier use. It has not detailed the limits of the free tier or the price and capabilities of paid plans.
The distribution advantage is clear. Meta already operates consumer apps used at global scale, and WhatsApp gives Muse a familiar conversational interface. If people can delegate routine tasks without learning a new workflow, Meta could lower one of the main barriers facing agent products: getting users to trust them with consequential work.
The operational challenge is trust
Useful agents need access to sensitive context, accounts and payment flows. Meta says Muse runs in a cloud-based virtual computer isolated from other users’ agents. It says Muse cannot see payment-method passwords, while a separate agent called Sentinel monitors the virtual environment and prevents actions from reaching the internet without approval.
The company also says users can opt out of having their interactions used to train Meta AI models and can ask Muse to forget particular information. A more secure “confidential” virtual-machine option, encrypted so Meta itself cannot access it, is planned for later this year. Support for 1Password and Shop Pay is also planned; Stripe Link is available for secure checkout, according to Meta.
Those safeguards are product claims, not yet evidence of how the system will perform at scale. For operators evaluating consumer-facing agents, the important design pattern is the combination of background execution, explicit approval gates, credential separation and memory controls. Each layer is necessary; none eliminates risks such as mistaken actions, prompt injection or poor handling of personal data.
Why Meta’s position is complicated
Muse enters a crowded field. OpenAI, Anthropic, Microsoft and Google all offer or are developing agent-like tools, with Google also pursuing a broader consumer audience. Meta’s differentiator is likely to be reach and ease of use rather than an entirely new category of capability.
But Meta must overcome an unusually high trust hurdle. The company’s history includes major privacy controversies, and recent AI product missteps have added scrutiny around how it handles user data and safety. An agent that can browse, submit forms and participate in transactions raises the stakes beyond an ordinary chat interface.
For founders and product teams, Muse is another signal that agent UX is moving from demonstrations toward operational systems: systems that need permissions, escalation paths, auditability and clear failure states.
What to watch next
The first question is reliability: how often can Muse complete real-world tasks without incorrect assumptions or excessive approval requests? The second is whether Meta makes its privacy settings understandable and meaningful by default. Finally, pricing and free-tier limits will reveal whether Muse is a broad distribution bet or a subscription-led productivity product.
Meta has the channels to put an agent in front of a huge consumer base. Turning that reach into sustained use will depend less on the novelty of delegation than on whether people feel safe letting it act for them.
