Verge test finds Meta's Muse AI agent can handle chores and purchases, with limits

The Verge reporter said Muse, whose avatar was named Marley, offered reminders and could work on several requests in parallel. During a morning test, the reporter asked it to follow up with a landscaper about a backyard renovation bid, buy a replacement steam wand assembly for an espresso machine, and downgrade a security system account to a cheaper plan. The reporter used a snowman-meets-puppy avatar suggested by Meta and watched Marley spin up virtual browser windows while working through the commands.

For the landscaper task, Muse requested access to the reporter's Gmail, confirmed that more time had passed than the landscaper initially estimated for the bid, and drafted an email. The reporter called the draft "rather snippy but wordy" and asked Muse to make it shorter. The reporter then sent the message, writing that hitting send "made me cringe." A follow-up from the landscaper arrived apologetic for the delay, and the reporter replied in a friendlier tone without Muse.

For the espresso machine, Muse found the replacement steam wand assembly and added it to a shopping cart on Breville's website. The reporter completed the order, and the part was expected to arrive in five to seven business days. The Verge described the part as costing about $7 and the task as something the reporter had put off for an unreasonable amount of time.

For the security system downgrade, Muse was able to log into the account after email verification, but the task required a phone call and was not fully completed. The Verge said Muse reported back information the reporter already knew.

The test centered on Meta's pitch that AI agents can triage and handle tedious tasks. Muse performed well enough on simple requests, especially when a credit card transaction was involved, but it did not eliminate the need for the reporter to finish or take over some steps.