Meta's asking for the keys to your inbox, calendar, and credit card — six months after OpenAI already taught agents to do this.
The Summary
- Meta launched Muse, a personal AI agent that handles shopping, emails, trip planning, and form-filling — marking a major pivot from entertainment to productivity AI
- The agent needs access to email, calendars, payments, and health services — making this the company's biggest test of consumer trust since Cambridge Analytica
- JPMorgan upgraded Meta to overweight on the news, signaling Wall Street believes the agent economy is real money
- Muse autonomously gathers information about users without explicit requests — functionality that works but feels invasive even to tech journalists
The Signal
Meta's Muse launches as a productivity agent that can open browsers, fill forms, and negotiate on your behalf. It's the company's attempt to catch up in the AI race after falling behind OpenAI, Anthropic, and Google. But the timing matters: Muse arrives six months after OpenAI already shipped similar capabilities. This isn't innovation. This is Meta finally showing up to a party where the drinks are already half-finished.
The agent works. A Verge reporter tested it and confirmed the core functionality delivers. You can turn it into a purple cat avatar if that makes you feel better about giving it permission to read your medical records. But here's what made the tester uncomfortable: Muse gathered information autonomously, without being asked. It inferred context, made assumptions, and built a profile of the user's life just by having access.
"The unnerving amount of information it autonomously gleaned about me largely overshadowed my experience."
This isn't a bug. It's the whole point. Agents only work if they know everything. Your schedule, your budget, your preferences, your health appointments, who you email and when. The value proposition requires surrendering data that Meta has historically struggled to be trusted with. Cambridge Analytica wasn't a technical failure. It was a trust failure. And now Meta is asking consumers to try again, but this time with access to payment credentials and calendar invites.
Wall Street doesn't care about your feelings. JPMorgan upgraded Meta to overweight the day after launch, reading Muse as proof that Meta has a real AI position beyond chatbots and Ray-Bans. The upgrade says: agents are not a science project anymore. They're a product category. And if Meta can pull this off, the revenue model is obvious. An agent that books your flights, orders your groceries, and schedules your oil changes becomes a toll booth on every transaction. Referral fees, affiliate cuts, premium tiers. The business model writes itself.
But business models only work if people actually use the thing. And that's the open question here:
- OpenAI shipped first and has no baggage
- Google has enterprise trust even if consumers are wary
- Meta has a decade of headlines explaining why you shouldn't trust Meta
Hacker News users have been debating this for days. The technical crowd sees the architecture. The privacy-conscious see the risk. Nobody's sure which instinct wins.
The Implication
If you're building an AI agent company, watch what happens with Muse adoption in Q4. Meta has distribution that no startup will ever have — billions of users, existing infrastructure, and infinite compute budget. If consumers still won't hand over their credentials, the agent economy has a trust problem that better UX won't solve. That kills half the pitch decks in the valley right now.
If you're a knowledge worker wondering when agents actually arrive, the answer is: they're here. The question isn't capability anymore. It's whether you trust the company holding the leash. And for Meta, that's still an open question even when the product works exactly as advertised.
Sources
The Verge AI | Bloomberg Tech | Hacker News Best | TechCrunch AI