Meta Unleashes Muse

Meta has entered the next phase of the agentic AI race with Muse, a personal AI agent designed not merely to answer questions but to execute tasks. Available initially in the U.S., Muse can send emails, book travel, complete forms, shop online and work toward longer-term user goals.

Muse operates through a dedicated Muse Secure VM, where the agent and connected user data are isolated. Users can interact with it through the Muse app or WhatsApp, while the agent can continue working in the background after the user closes the application.

Payments represent one of its most significant capabilities. Through an integration with Stripe’s Link, Muse can complete purchases at more than one million Link-enabled businesses. For other merchants, Link can generate a single-use virtual card tied to an approved transaction, keeping underlying card credentials hidden.

Muse also moves AI deeper into personal finance. Users can connect financial information and allow the agent to use relevant data when completing approved tasks, potentially turning AI from an advisory interface into a transactional financial assistant.

Security becomes critical when an AI can act autonomously. Meta says Muse requires approval for sensitive actions such as sending emails or making purchases, maintains an audit trail, separates credentials from the agent, and uses an independent Sentinel agent to approve internet interactions.

Strategically, Muse demonstrates Meta’s ambition to move beyond social networking and conversational AI into personal superintelligence and agentic commerce. Its enormous consumer ecosystem could provide a powerful distribution advantage against competing AI assistants.

The bigger shift is profound: AI is evolving from “tell me what to do” to “do it for me.” But giving agents access to email, financial accounts, browsers and payments dramatically raises the stakes around privacy, authentication, fraud and accountability. The winner in personal AI may therefore be determined not only by intelligence, but by who users trust enough to let AI act—and spend—on their behalf.