Meta Launches Muse, an AI Agent That Can Book Travel and Make Purchases

Muse can open a browser, fill out forms and continue working after a user closes the app. It can help with a single task, such as arranging travel, or develop a plan around a longer-term goal. The user can interact with it through the Muse app, WhatsApp or the web. Those capabilities move the assistant into a different part of daily life: the steps people normally carry out after deciding what they want to do.

Meta offers an ordinary example of how that might work. Muse could turn a recipe saved on Instagram into a grocery list, help plan a dinner and account for dietary restrictions mentioned earlier. It could also return with a question when a task needs a decision.

The launch also gives Meta a way to connect its AI work with the services people already use. Muse is powered by Muse Spark, a model developed by Meta.

A New Route Into Payments

Shopping is one of the clearest commercial uses. Meta says Muse can prepare a purchase and check out using Link, Stripe’s payment service. Link generates a one-time-use card for an agent transaction so the user’s actual card details are not exposed to Muse. Meta said at launch that Shop Pay and support for 1Password were planned; they should not be presented as launch features.

An agent that researches and completes purchases could change how a customer reaches a merchant. Instead of visiting a site and comparing products personally, the customer may ask an assistant to do the search and bring back a choice.

Meta has also drawn a line around purchases and other sensitive actions: it says Muse asks the user before buying an item or sending an email. Users can decide which services to connect and what access to grant. With email, for instance, someone can allow the agent to read messages without giving it permission to send them. Those controls may be central to whether people are comfortable moving from occasional questions to delegated tasks.

The Question of Access

The more work an assistant undertakes, the more personal information it may need. A calendar helps it find a suitable appointment; email may contain the details needed to manage a booking. Meta says Muse runs inside a dedicated cloud-based virtual machine and keeps connected credentials in secure storage. It also says the agent cannot see users’ passwords or payment methods and that a separate system, called Sentinel, reviews its access to the internet.

Users can review a record of what Muse has done, revoke access to a service and ask it to forget specific information. Meta says conversations and data in a user’s virtual machine are not shared with its advertising systems, and users can opt out of having their interactions used to train its AI models.

Meta has announced an additional encrypted system, called Muse Confidential VM, for later this year. The company says it would use a key held only by the user, preventing even Meta from accessing the contents of the virtual machine. Because that system has yet to launch, it cannot be counted among the protections currently available to Muse users.

Beyond the Phone

Muse’s next setting may be a pair of glasses. At its September 23 Connect event, Meta announced plans to bring the agent to its AI glasses, allowing people to interact with it hands-free. Glasses could make an assistant easier to call upon during the day.

The initial rollout remains limited to the United States, where Meta says Muse is available on iOS, Android and the web. The company describes most use as free, with subscriptions for people who want to do more.

Muse makes a straightforward promise: give an assistant a job and have it come back when the work is done or a decision is needed. Its appeal may lie in the time between those two moments, the comparisons, follow-ups and small administrative steps that can turn a simple errand into an afternoon’s work. The expansion to more devices will show which of those jobs people actually want to hand over.