Muse lives on a dedicated, secure virtual machine and takes real actions across a person’s apps rather than just answering questions, according to a Meta blog post.
“Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about,” Meta CEO Mark Zuckerberg wrote in an Aug. 10 essay, describing his vision for AI superintelligence. “Your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies and more.”
PYMNTS tested that promise against three basic consumer tasks, including reordering six rolls of toilet paper on Amazon, ordering a pizza from Domino’s and booking a restaurant reservation on Resy. Muse struggled with all three, slowed down by how much time and access each task required before it could even get started.
That struggle had little to do with the AI’s reasoning. It came down to everything required before Muse could even attempt a task. It needed Amazon credentials shared through a secure link before it could see order history. Gmail and Calendar each had to be connected individually, through their own separate authorization flow.
When it came time to pay, Muse routed payment through Link, a checkout system built by Stripe. Meta said Muse is the first agent covered by Link’s own purchase protections. Even with that infrastructure in place, none of the three transactions completed.
For comparison, PYMNTS timed the same toilet paper reorder directly on Amazon. It took less than 30 seconds, with no credential sharing, no separate app connections and no failed checkout.
Setup Took Longer Than the Task Itself
The gap between what Muse promises and what it took to attempt anything is the real story. Muse runs on Muse Secure VM, a dedicated cloud computer housing both the agent and a person’s connected data, with a separate “Sentinel” agent that must approve anything Muse sends out to the internet, Meta said in its blog post.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Muse launches free for up to 100 million tokens a week, with two paid tiers, including Power at $20 a month and Maximum at $100, Tech Insider reported Wednesday (Sept. 9). It’s restricted at launch to users 18 and older in the United States and accessible through iOS, Android and muse.ai.
Zuckerberg said he has personally used Muse for things like organizing a baking project and critiquing his own MMA training, framing it as closer to a capable household assistant than a task-completion machine, Sources reported Tuesday.
Delegation Only Helps If the Handoff Is Faster
That framing might be the more honest one. An agent that helps plan a recipe or offer feedback on a workout is doing something closer to advising than executing, and that’s a task Muse may handle well.
However, Meta pitched Muse to users in the blog post as something that “takes tasks and projects off their plate” and “advances the work on its own,” a much stronger claim.
For now, it feels like an intern who needs detailed instructions, explicit account access and supervision at every step. Attempting a task that a person could finish themselves in 30 seconds isn’t removing work. It’s adding a layer of management on top of it.
Zuckerberg said he expects Muse to eventually pay for itself by helping people earn and save money, with Meta taking “a very small cut of whatever the transaction is,” Sources reported.
That business model depends on Muse completing transactions reliably enough for people to trust it with real purchases. Based on three attempts at ordinary errands, that trust isn’t earned yet. The promise of agentic AI has always been that handing off a task removes work rather than creating more to manage. Muse, in its first days, adds the layer instead of removing it.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
Add as Preferred Source
