Meta sees Muse as more than a personal AI assistant. Mark Zuckerberg has described a future where people’s agents communicate and transact with one another, helping users shop, save time and pursue personal goals. That vision could create a new source of revenue for Meta, but the company still has to show that the service can scale safely and affordably.
Meta’s vision for a network of agents
On Meta’s third-quarter earnings call, Zuckerberg said he wants billions of people to use Muse. In his proposed network, personal agents could work with friends’ and family members’ agents, and communicate with businesses’ agents when users want to buy something.
The idea shifts some online activity from people navigating websites themselves to software agents handling tasks on their behalf. Zuckerberg said these agents could help with relationships, saving time or money, building a business and creating new things. He expects Muse could eventually generate more revenue per user than Instagram and Facebook, which currently bring in most of Meta’s money.
That is a large ambition for a product Meta released to the general public last month. Zuckerberg predicted Muse would be highly profitable in a few years, but the earnings call offered little detail about how the proposed network would work in practice.
Early usage and shopping offer clues
Zuckerberg said Muse had been downloaded and set up by more than 10 million users in the US and Canada. He called it one of Meta’s most successful launches ever. A quarter of people who initially downloaded Muse use it daily, according to his account; he did not provide specific retention figures.
Shopping is the opportunity he highlighted most clearly. Zuckerberg said a quarter of Muse users shop through the platform and spend more than $200 per purchase on average. Meta earns a fee on commercial transactions made through Muse, giving the company a potential way to make money if users and businesses conduct more shopping through agents.
In theory, an agent can search for products and make recommendations at any time, rather than being limited by a person’s availability. That could mean more transactions on the platform. But the potential does not establish how much shopping agents will actually handle, or how often users will trust them to complete a purchase.
Trust and business participation are unresolved
For an agent to act on someone’s behalf, users need confidence that it will follow their instructions and protect their information. The source article describes reports of agents leaking content from private conversations. It also recounts a YouTuber’s account that his Muse agent sent a stranger to his home during a Facebook Marketplace transaction.
Zuckerberg has said Muse was “built for privacy and security from the ground up.” Even so, the reported incidents raise practical questions about how the system behaves when it makes arrangements or handles sensitive information. A Canadian bank executive told Bloomberg the bank would consider blocking Muse from accessing customers’ financial data.
Businesses also need a reason to take part. Meta expects businesses to cover transaction fees while Muse is likely to remain free for consumers. Susan Li, Meta’s chief financial officer, said that “everyone is just going to have to adapt” when an investor asked about friction in getting businesses on board. That response expresses confidence in the direction of travel, but it does not explain how reluctant businesses will be persuaded to participate.
Computing costs could limit the plan
Muse’s reach depends on Meta’s ability to run a service that Zuckerberg called compute-intensive. He said the company has not launched it internationally because of compute efficiency. Meta plans to spend between $135 billion and $145 billion in 2026, double what it spent last year, as it builds out its AI capabilities.
Zuckerberg told investors Meta plans to “10x” the cost-efficiency of its compute next year, without explaining how. A network serving billions of agents would require substantial computing resources, and the article notes that running GPUs at that scale is expensive and energy-intensive. Lower costs are therefore central to the business case: if the service cannot become cheaper to operate, its expected profitability may be harder to reach.
Muse has attracted early users in the two countries where it is available, and its shopping activity gives Meta a possible revenue model. But a larger network rests on several unsettled conditions: users must trust agents, businesses must accept the arrangement, and Meta must control the cost of running the service. The earnings call presented a vision and some initial figures, while leaving those questions open.