Is Muse Really the Answer to Meta’s AI Woes?

Meta’s Muse, the personal artificial intelligence agent the company launched Sept. 8, has taken Silicon Valley and Wall Street by surprise. Within days of launch the personal AI agent topped the App Store charts, a feat Meta’s earlier AI apps never managed, and Meta shares jumped more than 11% in a single session.

For a company that spent two years being asked what its AI budget was buying, that looks like the consumer AI breakthrough it has been waiting for. Whether the product underneath can hold that position is a separate question.

Muse has more than 2.5 million downloads, CNBC reported, citing Sensor Tower. That outpaces Anthropic’s Claude and xAI’s Grok over comparable launch windows, as well as Meta AI, the standalone app Meta launched last year to little effect. It trails OpenAI’s ChatGPT, which logged 3.1 million downloads in the same stretch.

Muse Has to Be Trusted With the Keys

The bigger risk is what Muse holds while it works. To do its job, Muse connects to email, calendars, payments, health apps and shopping accounts. Meta shipped it despite internal concerns that the agent mismanages access to sensitive data, Reuters reported. Employees testing it during launch week described cases where it disconnected without explanation and uploaded sensitive information without permission, according to internal posts Reuters reviewed.

Meta delayed the launch from April to work on security. Even so, Vishal Shah, its vice president of AI products, told Reuters, “It is impossible to say that there is never going to be a mistake.” The launch came less than two weeks after Meta agreed to an $18 billion multistate settlement over social media harms, TechCrunch noted, Muse also asks users for more of their data than any Meta product before it.

Muse’s phone-calling feature raised another concern. Reuters reported that Meta tested a “human concierge” in which contractors handled some of the calls Muse places and that employees warned it could route sensitive information to call centers. Meta said the test was meant to gather feedback before a public release. The company says users can opt out of having their interactions used for training, and an encrypted version of Muse is due later this year.

Amazon Blocks Muse as Payments Partners Sign On

Getting onto users’ phones was the easy part. Getting into stores is proving harder. Amazon blocked the agent after Meta declined its request to pull it down from the site, Bloomberg reported. Amazon says the agent never disclosed it was accessing the site, doesn’t identify itself while browsing and appears to store customer logins, GeekWire detailed. Meta has said Muse can’t see passwords or payment methods.

Meanwhile, Meta has been lining up payments and commerce partners one integration at a time. Muse launched with Stripe’s Link wallet, and PayPal added its checkout across its merchant network. Shopify CEO Tobias Lütke said Shop Pay would work for Muse on every Shopify store. Flight bookings run through Duffel, a travel API provider, and Ticketmaster and OpenTable plugged in at launch, Skift reported.

OpenAI has already tried that route. It launched Instant Checkout in ChatGPT in September 2025 with Etsy and Shopify and shut it down in March, telling merchants to use their own checkout while it refocused on discovery, TechCrunch reported. Only about a dozen of Shopify’s millions of merchants went live, Modern Retail found. Users researched in the chat and bought somewhere else.

Muse Costs More to Run Than a Chatbot

In July, Meta guided 2026 capital spending to between $130 billion and $145 billion as second-quarter free cash flow fell 91% to $784 million, Sharecast reported. Muse is built to consume more of that per user than anything Meta has shipped. Each agent runs in its own cloud virtual machine with its own browser and keeps working after the app closes, Axios reported. A chatbot answers and stops. Muse holds a machine open for as long as the errand takes.

Muse is free for most users, with paid tiers at $20 and $100 a month. “Those subscription tiers help us cover the compute costs,” Meta Chief AI Officer Alexandr Wang told Axios. Muse carries no ads. Instead, Wang said Meta is exploring commerce revenue instead.

The evidence that consumers will pay is thin. Only about 3% of households pay for any AI service, at a median of $20 a month, the Bank of America Institute found in card data covering roughly 70 million consumers. By comparison, OpenAI said in February that it had more than 50 million paying subscribers against more than 900 million weekly users, a mid-single-digit conversion rate.

Muse’s math is harder on both the cost and revenue sides. Its compute bill per session is higher than a chatbot’s, and Wang said most people should never need to leave the free tier. That leaves the $100 plan and merchant commissions to carry a product whose costs rise with every download the charts are celebrating.

Downloads bought Meta a headline. Retention, merchants and margins will decide whether Muse is a business.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.