Meta wants into your inbox, your calendar, your payment methods, and apparently your health records too. That's the real story behind Muse, the personal AI agent the company rolled out this week, and it's a far bigger ask than the shopping-and-email demos suggest.
Strip away the framing and Muse is Meta's entry into a race that's already crowded with agentic assistants promising to act on your behalf rather than just answer your questions. The pitch is familiar: book a flight, sell your car, handle the email backlog, proactively nudge you toward your goals. What's different here is the sheer breadth of access Meta is requesting to make it work — email, payments, health-related services, the works. Most competing agents ask for a slice of your digital life. Muse wants the whole thing. That's either a sign of genuine technical confidence or a company that's decided the only way to compete is to out-ask everyone else on permissions, and I'm not sure which is more concerning.
Trust is the actual product being tested here, not the automation itself. We've had years of chatbots that answer questions; agents that can autonomously spend your money or message your contacts are a different category of risk, and Meta knows it — hence the "secure, privacy-focused" language doing a lot of work in the announcement. But Meta's track record on data handling means it's starting this race from behind, regardless of how good Muse's underlying model is. I'd bet the free tier gets plenty of curious users, and the real test comes when people are asked to actually hand over payment credentials for something more consequential than sending a birthday email.
By the way, it's worth noting who else launched something genuinely different this week, because it's a useful contrast. Google DeepMind's AlphaGenome Atlas — precomputed predictions across roughly 9 billion human DNA variants, free for researchers — is the kind of release that doesn't need a trust campaign. Nobody has to decide whether to let it into their inbox. It just sits there as infrastructure, quietly making genomics research faster for anyone who wants to use it. That's the version of AI progress that compounds without friction, and it's a reminder that not every meaningful release needs a consumer-facing trust negotiation attached to it.
Meanwhile Cognition closing a $2 billion round at a $48 billion valuation tells you where investor conviction currently sits: not with consumer agents asking for your calendar access, but with tools like Devin that automate work developers already do inside contained, permissioned environments. That's a much easier trust problem to solve than "give an app your payment details and hope for the best." Muse might work brilliantly. But the harder question Meta has to answer isn't technical — it's whether people who've spent a decade wary of what Meta does with their data are ready to hand it the keys to their inbox and their wallet at the same time.