Let an AI agent go to town with your credit card? You must be nuts. But that’s the pitch — and the appeal — of Muse, the new agentic assistant from Meta that’s turned into an unlikely AI hit. I’ve been playing with Muse for the past few weeks and, until a couple of days ago, I wasn’t too impressed. To me, it looked like yet another chatty AI assistant, all talk, unable to do the simplest things without fumbling the ball or burning through usage limits. And, of course, trusting Muse entails trusting Facebook and Instagram parent Meta, the maker of those creepy, camera-equipped Meta glasses. But after trying a few agentic AI mainstays — asking “Mica” (you can name your Muse assistant) to book a restaurant reservation (yawn), triage your Gmail, or check my morning commute (it referred me to New York City’s official subway website — thanks?) — I decided to go shopping. That’s when things got interesting. Welcome to another edition of Prompt Mode, your weekly AI newsletter. I’m your host, Ben Patterson. Each week on Prompt Mode , I’ll be serving up analysis of the AI trends that matter to everyday users like you and me. Stay tuned for practical AI tips, hands-on experiences with the latest AI tools, and–you guessed it–prompts to help you get the most out of your AI assistants. Thanks for reading, and if you like what you see, just sign up right here . Available on iOS, Android, Mac, and on the web (I used the iOS app), Muse isn’t the first AI agent to shop on your behalf — Amazon’s Alexa for Shopping can do that, of course — but it is among the first to support Link by Stripe, a one-click digital wallet supported by thousands of online merchants. For those retailers who don’t take Stripe, Link by Stripe can create a single-use virtual card that’s “scoped” for a specific purchase, meaning Muse can use your Stripe wallet pretty much anywhere. While Muse can use Link for Stripe to purchase items on your behalf, it can’t see the details of the payment methods (including credit card and bank account numbers) you’ve entered into your Stripe account. It’s also not supposed to make any purchases without your final, explicit approval. Now, assuming you trust Muse with your credit cards (a big ask), what makes it better than just shopping on Amazon — which, by the way, has blocked Muse from accessing its website ? I learned the answer when I started using Muse to look for theater tickets. Muse, as it turns out, is a tireless online shopper, running in its own virtual cloud-based machine and willing to scour multiple websites looking for deals. As a Manhattan dweller who loves theater but can’t afford Broadway tickets every night, I got the idea of using Muse to scoop up bargains for the hottest tickets in town. I started with the opera — specifically, Verdi’s Macbeth at the Met Opera, which just got a rave review from the New York Times (meaning, time to act fast). “How much are tickets to Verdi’s Macbeth at the Met,” I asked Muse, which searched the web and came up with the same results you’d get from ChatGPT or Claude, then advised me to check the official Met Opera website. “Can you check the official website for me,” I pressed, and then it spawned a web browser and began scrolling and clicking, eventually finding me a $71 seat (including a $10 Met service charge) for a weekday evening performance. Muse then logged me into the site (the Muse app can store your credentials securely without letting Muse see them) and teed up the ticket for checkout. All I had to do was say “yes” to Muse, and I’d be going to the opera. Ultimately I backed out, asking Muse to cancel the transaction and instead check availability for another hot ticket: The Holes , a limited-run Off-Broadway play that also just got a good review from the New York Times . I asked Muse to find whatever tickets are left, and it navigated to the production website, clicking on each and every performance (it’s only running for about four more weeks) before pinpointing precisely which ones still had seats, and for how much (turns out $92.41 is the best I could do). Again, I could have asked Muse to snap up the tickets if I wanted to. Finally, the big one. “What about Hamilton tickets? Cheapest options in the next 30 days?” Muse chewed on this one for several minutes, browsing this site and that, before finding one: a rear-mezzanine seat in row M for a Saturday matinee, $222 all-in. Did I want it? Not this time, but I could have asked Muse something like, “Get me two Hamilton tickets tonight for the lowest price,” and Muse could make it so, all while I’m making a tuna sandwich. Of course, all this agentic shopping burns up tokens, and while Meta seems to be generous with usage limits for now (I’ve only used 11 percent of my weekly allotment on the free plan), there’s no telling when Meta may choose to tighten its purse strings. (You can up your Muse usage allowance with $20- and $100-a-month subscription plans.) Still, I can see the appeal of having Muse seek out discount tickets and other deals that aren’t of the usual, one-click Amazon variety, all while you’re busy doing something else. I can also imagine the trainwrecks once everyone’s shopping with Muse-style AI agents, each climbing over the other to snag that last ticket. More in AI this week Speaking of Meta’s creepy AI glasses, the company just announced a new pair without cameras , as well as a Tagamochi-style “Muse Charm” that you can wear at all times. Hmm. [Meta, Engadget] In what’s being called a world’s first, Australian officials are accusing OpenAI models of hacking its government health website , with Prime Minister Anthony Albanese adding that there “will obviously be legal consequences.” [BBC News] The successor to the notoriously chatty Claude Opus 5, the updated Opus 5.5 model seems less prone to rambling about “loose ends” and other dangling threads. [PCWorld] Could Google, OpenAI, and Anthropic actually set aside their differences to establish an AI safety standards body ? It may indeed be in the works, sources say. [The Information] Everyone hates AI, right? Not according to this Gallup poll , which found that a majority of global respondents “said they felt curious, happy, or excited about A.I,” as opposed to the gloomy outlook of the “worried West.” [New York Times] Gemini-powered features like the Magic Pointer and Ramber are the stars of the Googlebook show, but it was a lower-profile Linux addition that got my attention. [PCWorld] Prompt of the week: The “step zero” prompt Ever feel paralyzed in the face of a task, either big or small? When your brain’s on overload, even a small chore can feel daunting, and breaking a bigger task into smaller ones might make things worse. If you’re feeling truly stuck, try the “step zero” prompt . Rather than giving you a lengthy task list, this prompt serves up a simple, mechanical first step, like “Turn on your computer” or “Walk into the kitchen,” plus a few basic follow-ups. The idea is just to get you started, which sometimes can be the toughest part. That’s all for now! Thanks for reading the latest issue of Prompt Mode. Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.
Pcworld
Original source
