Meta’s AI agent is a cute little guy who’s great at spending my money

In a hands-on test reported by The Verge, Meta’s Muse agent — personified as a cheerful avatar called Marley — was set to work on a string of everyday chores the reporter had been avoiding. With permission to access the user’s Gmail and accounts, Marley checked the status of a landscaping bid and drafted a follow-up message that the user then trimmed and sent. That task was completed with minimal effort from the reviewer.

For a small online purchase, Marley located the replacement steam wand assembly for the reviewer’s Breville espresso machine, added the item to a shopping cart, and prepared the order; the user had to finish the payment step themselves. The part was scheduled to arrive in five to seven business days. When asked to downgrade a security-system plan, Muse was able to access the account and report that a phone call was required, but it could not complete the phone-call step — a capability the reviewer notes is reportedly under development.

Marley was also helpful in researching junk-removal options for a fallen tree branch, demonstrating the agent’s ability to open browser windows, compare services, and advance the logistics for the task, though some user interaction remained necessary. Across these examples the reviewer found Muse effective at reducing friction for routine, noncritical chores, especially those that involve web browsing and drafting messages, while tasks that depend on real-time voice calls or final human approval still required manual involvement.

The Verge’s experience shows that contemporary AI agents can meaningfully reduce the effort involved in small household administrative tasks, particularly when granted account access and the ability to interact with web pages. At the same time, the review highlights practical limitations around autonomous transactions and phone interactions, pointing to the likely next areas of development for such personal assistants.