Meta's Muse agent steals the spotlight from OpenAI and Anthropic, but runs into the wall of advertising trust

By TechCrunch · September 27, 2026.

At its annual Connect —the event where CEO Mark Zuckerberg sets out the company's AI roadmap— Meta unveiled Muse, its new agent designed to handle users' everyday tasks. It also showed a wearable device with Tamagotchi aesthetics (the 1990s virtual pet game) that, as Meta itself insists, is intended for adults only.

The launch coincided with a week of model announcements from OpenAI and Anthropic focused on coding and enterprise tools, as journalists Kirsten Korosec, Sean O'Kane and Anthony Ha discussed on TechCrunch's Equity podcast. While the rest of the AI frontier races toward the enterprise, Meta is betting, against the current, on the end consumer.

Sean O'Kane tested Muse over the past few weeks. One of the first things he did with it was follow one of the agent's own suggestions: tracking down unclaimed funds in his name —a task almost nobody gets around to— and he found real money, with a check already in the mail. But he qualifies it himself: it was "more of a party trick" than a reason to open the app again every day.

O'Kane describes Muse, in his own account on the podcast, as a consumer version of the technology from OpenClaw, an agent startup Meta reportedly bought and integrated months ago. One detail caught his attention: on startup, Muse did not automatically pull his data from Threads, Instagram or Facebook, as he expected; it treated him "like a stranger." With use, however, he began to notice that the agent keeps pushing to bring that information into the system.

That is where the real obstacle appears, and it is not technical. For an agent to be useful beyond the one-off stroke of luck —canceling unused subscriptions, spotting duplicate charges, proactively managing finances, something firms like Rocket Money already offer— it needs deep access to sensitive data: the credit card, the email account and all that other personal information. And there, in O'Kane's words, "the trust wall" with Meta comes into play: its business is selling advertising, and the more it knows about a user, the more precise —and interesting, by the company's own argument— the ads it sells them. O'Kane compares the situation with Apple's new Siri, which in his view inspires more trust precisely because Apple's business does not depend on advertising.

As industry context, the strategic divergence has business logic behind it. As Ha speculated on the podcast itself, OpenAI and Anthropic are pushing toward the enterprise in part because both are weighing IPO plans on a near horizon and need to show revenue at an unprecedented scale to justify sky-high valuations; this is one journalist's live read, not dates confirmed by the companies. Meta does not face the same urgency: it is already profitable thanks to advertising and has a distribution base no AI startup can match —Facebook, Instagram, WhatsApp and Threads together account for billions of active users. For Meta, betting on the consumer means playing on the ground where it already wins.

Our take is that this episode matters more than its lighthearted tone suggests, because it exposes the structural conflict of interest that will decide who wins the race for the "agent that runs your life." It is not a question of which model reasons better —in fact Muse does not even compete on those terms— but of which company earns permission to touch your finances, your email and your private conversations. A company whose business model depends on selling advertising attention starts that race with a credibility handicap that no product launch resolves on its own, and Meta's record on handling personal data —widely scrutinized over the past decade— does not help close that gap.

In the short term, it is reasonable to expect Meta to win attention and curious users —it has the distribution to do it— but converting that curiosity into the kind of deep trust required to hand your financial life to an agent will prove far harder. In the long term, however, the underlying promise —an assistant that, unprompted, tracks down forgotten money, cancels unnecessary spending and sorts out everyday paperwork— is exactly the kind of daily friction that AI-driven abundance should be steadily eliminating, freeing up time and attention for what really matters to each person. The obstacle is not that the technology does not work: O'Kane's case shows that it does. The obstacle is that, as long as the economic incentive of whoever builds the agent remains selling more ads, every user retains a legitimate reason to doubt whom that help actually serves.