Loading page…
A hands-on test found that connected accounts helped the agent finish everyday tasks, while outdated listings and retailer restrictions still demanded human help.
Listen to this story
CNN’s test of Meta’s Muse, launched September 8 in the US, found an agent that can turn moving and trip requests into schedules, emails and documents—but does not reliably carry tasks through to completion. Outdated restaurant listings undermined recommendations, while Amazon blocked browsing and Target prevented automated checkout; some purchases also ran into seller requirements. Meta’s security checks and cloud-computer setup do not solve those limits.
With Google apps connected, Muse emailed a friend and turned restaurant suggestions into a Google Doc with alternatives; CNN found ChatGPT and Gemini sometimes more detailed.
Meta says its Sentinel agent approves internet access or requests permission; Muse cannot see passwords or payment methods and checks before sending email or making purchases.
Meta built Muse to take care of tasks, not just suggest how to do them. In CNN’s hands-on test, the agent sent an email and organized trip ideas. It also recommended long-closed venues and hit barriers while shopping. Acting on a request is one thing; getting the next step right is another.
Muse launched on September 8 as a personal AI agent designed to work across connected services. It is available in the US through a standalone app and WhatsApp. CNN tested how much everyday work it could finish.
For an upcoming move, the CNN reviewer listed what was left to pack, gave a deadline and asked to work in short sessions. Muse returned a day-by-day schedule with time allotted for each session.
After the reviewer connected Google apps, Muse emailed a friend about a planned trip. It then turned the friend’s restaurant suggestions into a Google Doc with alternative options. CNN also found that ChatGPT and Gemini sometimes gave more detailed responses on similar tasks.
Asked for date-night ideas, Muse suggested two nearby places that had been closed for years. The reviewer would have needed to check those suggestions before making a reservation. This was an accuracy failure, separate from the website restrictions that complicated shopping.
Amazon blocked Muse from browsing its store, though the agent could find product links through search engines. At Target, Muse could browse, but the retailer blocked automated checkout clicks. The reviewer had to enter personal details manually. One restriction limited browsing; the other stopped autonomous checkout.
A shampoo purchase stalled for another reason: the seller required a professional license number. Muse inspected the site and explained the obstacle. It could identify why the task stopped, even though it could not finish it.
Muse’s Ideas tab described an agent that could message Facebook Marketplace sellers and negotiate a price. In CNN’s attempt, Muse said it could not contact sellers directly and offered to draft a message instead. A Meta spokesperson maintained that the description was accurate and said Muse can negotiate within user-set parameters. Why the attempt fell short remains unclear.
Meta says Muse runs on a dedicated cloud computer. A separate Sentinel agent approves internet access or asks the user for permission. Muse cannot see passwords or payment methods, Meta says, and checks before sending email or making purchases. Those safeguards do not fix an outdated listing or override a retailer’s website rules.
Sensor Tower data cited by CNN put Muse above 2.5 million downloads since launch and at the top of the iPhone App Store’s free-app chart. The test leaves a question that downloads cannot answer: how often will users have to check or finish the agent’s work?
Loading discussion...
Join the conversation
Why would you give stores—or shoppers—the final say?
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.