Meta Moves Its Muse Agent From Launch to Scale
On Meta’s third-quarter earnings call, Mark Zuckerberg said the company is entering “a new phase and a more operational phase,” centered on scaling Muse, its personal agent. He said Muse now has clear product-market fit, meaning users want the product enough to keep using it. The Engadget report of the call was published October 21, 2026.
Zuckerberg said more than 10 million people have started using Muse since its early September rollout. He said Meta expects to grow the service to serve billions of people. Muse is currently available only in the United States and Canada. Meta previewed it on its smart glasses at the Connect event in September, and Zuckerberg said it will arrive on the glasses the following month.
On monetization, Zuckerberg said he expects Muse to eventually generate more revenue per user than Facebook and Instagram. He put the average revenue per American user across Meta’s family of apps at several hundred dollars a year. Meta is already working on ways to earn from the agent. One plan is commerce tools that would let Meta charge businesses fees on transactions the agent facilitates, which works like a commission on sales the agent helps complete. Shopping is the most common use case, according to Zuckerberg, with people spending an average of $200 per purchase through Muse.
Subscriptions are the second leg. Meta plans to offer AI subscriptions to people and business owners who use Muse more heavily. Zuckerberg also said Muse is significantly more expensive for Meta to operate than its other services. He tied the agent to the glasses business as well, saying Muse is central to driving AI subscription sales on the glasses, and that Meta does not currently make a large profit on the glasses hardware.
Yahoo Finance reported that Muse had 2.182 million daily active users on October 5, 2026, and that the agent brings back 73% of new users, meaning nearly three in four new users return. Benzinga reported that Meta’s quarterly revenue was $60.8 billion, above analyst estimates of $59.5 billion, while earnings of $6.18 per share fell short of estimates. A Reuters preview published October 20 said the quarter’s results would focus on AI spending and adoption of Muse, with Zuckerberg looking to build a new revenue stream from it.
The product has moved quickly. Bloomberg reported on September 8, 2026 that Meta had announced Muse as an AI agent for personal tasks and organization. In a Facebook video published September 21, Zuckerberg framed its primary purpose as helping users complete tasks, manage work, and stay productive. CNBC reported on September 29 that Meta had launched Muse for Small Business, a specialized version for small businesses. Earlier, in July, Meta had introduced Muse Image, the first image generation model from Meta Superintelligence Labs, available in Meta AI. The New York Times reported on October 9 that Meta had delayed releasing the Muse app for months over safety concerns, before new competition pushed Zuckerberg to release it.
In my view, the central question on this call is not the user count but the unit economics, meaning whether revenue per user covers what each user costs to serve. A consumer product that costs more to run than Meta’s existing services has to earn its keep through transactions and subscriptions rather than the engagement-driven advertising that funds the feed apps. Zuckerberg’s own numbers imply that Muse must monetize well above the current per-user average to justify its cost. Whether a conversational agent can take a meaningful cut of purchases, and whether heavy users will pay a recurring fee, are open questions that the coming quarters will test.
The broader context here is the glasses. Muse gives the hardware a reason to be used daily, and a subscription layer gives Meta a way to earn from a device it does not yet sell at a profit. For developers and enterprise teams watching the agent space, the more immediate signal is the shift in Meta’s language from building a product to operating one at scale. Expanding from two countries to a global base, while managing safety review and inference cost per session (the compute bill for each conversation), is a considerably harder engineering and governance problem than the launch itself. If Meta can make the agent a dependable way to get routine tasks done, the commerce and subscription revenue Zuckerberg described would follow from that utility, and that is the outcome the company is now betting on.
