--- title: "What Is Meta Muse? The AI Assistant Behind 10 Million Setups" url: "https://goremy.ai/blog/articles/meta-muse-ai-assistant-10-million-users" author: Luis Chavez-Mattos editor: Luis Chavez-Mattos editor_title: Director, Product published: "2026-11-06T12:20:09.111Z" updated: "2026-11-09T18:54:04.958Z" topic: AI Models & Industry description: "Meta's Muse hit 10 million setups in six weeks. Here is who built it, how a team under 200 did it, the polish details, and the cost-to-serve questions." --- # What Is Meta Muse? The AI Assistant Behind 10 Million Setups > Meta's Muse hit 10 million setups in six weeks. Here is who built it, how a team under 200 did it, the polish details, and the cost-to-serve questions. Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-06 · Updated 2026-11-09 ## What is Meta Muse? Muse is Meta's consumer AI assistant, delivered as a phone app with an animated character you name and talk to. People use it for ordinary errands, such as sorting out an account or finding subscriptions to cancel. According to the creator whose coverage this article draws on, about 10 million people set it up within six weeks. It was built by a small team led by Alexander Wang, and it is not the smartest agent on the market. Its pull comes from polish. ## TL;DR - Muse reached **10 million setups in six weeks**, a far faster consumer response than Meta's metaverse push ever got. - Alexander Wang says **fewer than 200 people** worked across the model and the product, with Nat Friedman central to the product work. - The team tested and improved **more than a hundred specific model behaviors** before launch, rather than leaning on raw model intelligence. - Small design details, like an avatar that visibly shows what the agent is doing, **answer the "are you working on it?" question** every delegating user has. - Reality Labs lost **$19.2 billion in 2025 alone**, which is the backdrop for why Muse matters to Meta. - The open problem is **cost to serve**. Muse is expensive to run, and Meta is aiming at 100 million users and then a billion. ## Why does Muse matter after the metaverse bet? Meta's apps reach billions of people, but they mostly run on devices and operating systems that other companies control. Zuckerberg wrote about that frustration in his 2024 letter on open source, where he said he wants more control over the next generation of computing. The ambition goes back at least to the 2014 Oculus acquisition. In 2021 he renamed the company around the metaverse. The mass-market platform he described did not arrive. Reality Labs, the division behind headsets, glasses, and related research, lost $19.2 billion in 2025 alone. That is one year of losses, not the full bill. Muse takes a different route into the same goal. Nobody has to buy into a grand vision. A user arrives on the phone they already own with a practical problem, and if the assistant helps, they come back. The product has to make the first thing it asks of a person worth doing. A big roadmap never gets heard if that first step does not improve someone's day. ## Who built Muse, and why does team size matter? Wang came over from Scale to lead Meta's AI effort. In an October interview with David Senra, he described Zuckerberg approaching him after disappointment with Meta's AI progress and asking what the company should do. Nat Friedman became central to the product work. Wang said fewer than 200 people worked across the model and the product. That is small by Meta standards, and it was a deliberate choice. A larger group tends to settle hard product questions by consensus, which produces compromises. Users feel those compromises as a feature that behaves differently in one tab than in another, or a second set of rules to learn for a related task. A small group with Meta's resources behind it can make sharp calls and own the whole experience. The creator also notes that Zuckerberg stepped back once he had the people he wanted. He recruited broadly but not excessively, funded the work, and let the team run. ## What makes Muse feel polished? The strongest claim in the coverage is that Muse is the most polished consumer AI app available right now. The evidence is specific. - **Visible status.** In the creator's September testing, the assistant (named Marvin, after Douglas Adams) showed it was working by pulling out a little laptop and typing, or scratching its head. No technical status message needed decoding. - **Behavior testing.** Wang described a list of more than a hundred model behaviors that the team tested and improved before launch. A capable model can still be an annoying assistant if it does not know when it needs you and when it does not. - **An expressive avatar.** The character has facial expressions. Meta worked on a model that generates expressions and movements for live conversational avatars and syncs them to speech. - **Consistency beyond the first screen.** The creator says the care continues into secondary tabs and less showy parts of the app, the equivalent of a well-kept hotel bathroom. The first item matters most for anyone building agents. Every time a person delegates a task, they wonder whether the agent is doing it. If they have to stop and investigate, they are supervising software instead of getting on with their day. ## What can Muse actually do? The sources describe one concrete job. The creator used Muse to review recurring charges and cancel $1,285 in annual subscription spending, a chore they had been putting off. The same coverage frames Muse as an everyday assistant, for bills, accounts, and questions about your day, and not as a tool for power users. The sources do not give a full feature list, pricing, or model specifications, so treat those as unknowns. The creator also says plainly that Muse is not the smartest agent available. ## What is the avatar work building toward? The avatar research points at voice. The creator says Meta has anticipated a live voice experience, including a Tamagotchi-like wearable you could hold up and talk to. That product is not in users' hands yet, and the details are unconfirmed in the sources. The expression-generation work was done before the phone app launched, so Meta could start building a continuing relationship with users early. A charming character only helps if the work behind it is good. If the assistant smiles while every interaction confuses you, the polish fails. The character, the conversation, and the task execution have to fit together so asking for help feels natural. ## Is Muse already a platform? No, and the creator says so directly. Ten million setups is a strong early audience, but Meta is looking at billions. Several things stand between the two numbers: | Challenge | What the coverage says | |---|---| | Cost to serve | Muse is expensive, and Meta has to bring cost down at much larger scale | | Network effects | Muse needs to work across assistants and businesses, which Meta has started to do | | Scale | The target is 100 million users and then a billion, partly by expanding geographically | Meta has advantages most builders lack, including an existing business to fund the effort, huge distribution, and compute capacity. Those advantages do not remove the cost question. Serving an assistant that takes real actions for a billion people is a different economic problem from serving 10 million. ## What can builders take from Muse? Three lessons come out of the coverage, and none of them require Meta's budget. 1. **Change something real after a miss.** Meta did not return to the metaverse with the same team and the same assumptions. It brought in new leadership and a new product approach. 2. **Keep the team small enough to decide.** Fewer people means fewer consensus compromises in the final product. 3. **Finish the boring parts.** AI makes first drafts cheap, which tempts people to start the next idea. Retention comes from fixing awkward edges in the one you already have. ## Frequently Asked Questions ### How many people use Meta Muse? According to the coverage, about 10 million people set up Muse within six weeks of launch. That counts setups, not necessarily daily active users, and the sources give no retention figures. ### Who leads the Muse team? Alexander Wang, who came to Meta from Scale, leads the AI effort behind it. Nat Friedman became central to the product work. Wang said fewer than 200 people worked across the model and the product. ### How much did Reality Labs lose in 2025? Reality Labs, Meta's division for headsets, glasses, and related research, lost $19.2 billion in 2025. The same coverage stresses that this is a single year's loss and not the total spent on the metaverse push. ### Is Muse the most capable AI agent? No. The creator says Muse is not the smartest agent on the market. The argument is that its attention to behavior testing, status design, and the avatar makes it feel better to live with than more capable rivals. ### What is Muse's biggest challenge now? Cost to serve. The coverage calls Muse expensive, and Meta needs to cut that cost sharply to reach 100 million users and then a billion. It also needs the assistant to work across other assistants and businesses.