Meta’s AI agent is a cute little guy who’s great at spending my money. That is the joke this story wants to make, and it delivers it in full. Muse, Meta’s new personal assistant, is supposed to handle the tedious crap you keep putting off. It does that. It also handles your credit card with surprising enthusiasm.
The reporter tested Muse across four tasks: following up with a landscaper, buying a replacement part for an espresso machine, downgrading a security system account, and arranging junk removal for a fallen tree branch. Each task involved a request, a response, and a result. The pattern became clear fast.
The Landscaper Email
Marley, the avatar name the reporter chose, started by requesting access to Gmail. It confirmed the landscaper had taken longer than expected on the bid and drafted an email. The reporter called it snippy and wordy. The reporter, who normally refuses to let AI draft email, hit send anyway. The cringe was real.
“Make it shorter,” I said.
The Espresso Part
The steam wand assembly for the Breville Barista Express cost about $7. Marley found the part, added it to a cart, and handed the task off to the reporter. The order is done, and the part arrives in five to seven business days. It was the simplest task on paper and the most efficient in practice.
The Security Call
Downgrading the security system account required a phone call. Muse could not make it. The reporter saw a follow-up from the landscaper apologizing for the delay, which softened the tone of the earlier email. The reporter wrote a friendlier reply the old-fashioned way.
The Tree Branch
Arranging junk removal for a fallen tree branch in the backyard took medium effort and a week of procrastination. Marley got the ball rolling, but the task remains unfinished.
The Credit Card Problem
The pattern holds across all four tasks. Muse is most useful when you pile requests on top of each other. The reporter gave Marley one task, then another, then another, and watched virtual browser windows spin up to chase each request. It was dizzying.
The reporter went with a generic cute snowman-meets-puppy avatar that Meta suggested at setup. Marley tapped cheerfully on a keyboard as it worked its way through the commands. The tasks were mostly pebble-in-the-shoe annoyances, the kind of thing you put off for weeks, months, or years.
What the Reporter Learned
The reporter’s own rules collapsed under the strain of actually using Muse. The AI drafted an email the reporter normally would not let a robot touch. The reporter broke that rule for testing purposes and regretted it.
Muse is better at spending money than at making phone calls. The phone-call problem is a weird future to consider. There is a path where robots talk to robots before any human enters the chat. There is also a short-term path where humans in call centers talk to other companies’ robots before reaching another human. Either way, the end state is that the robots do business without any intervention at all.
The Verdict
Muse worked better than the reporter expected, especially when there was a credit card transaction to be made. The email draft was a cringe moment. The espresso part was a smooth win. The security call was a reminder of what AI still cannot do. The tree branch is still waiting.
The reporter’s honest reaction is amusement. The AI agent that was supposed to handle the tedious crap ended up excelling at spending money. That is the joke, and it lands.
The story leaves one open question. Will Muse ever learn to make a phone call on its own? The reporter thinks so. The logical next frontier for these AI agents is the phone call. The weird future soon will be here, and the robots will talk to each other without any human in the loop at all.
Source material: “Meta’s AI agent is a cute little guy who’s great at spending my money,” The Verge.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.