Meta’s AI Agent Can Buy Things and Send Emails Without You Doing the Work

Meta is betting that the next big AI battle won't be about who gives the smartest answers, but who can actually get things done.

Muse is designed to act as a digital operator, rather than simply a chatbot. Users can give it a goal and allow it to navigate websites, fill out forms, book travel, send emails and make purchases. It can even continue working after the user closes the app and return when it needs approval.

That changes the competitive landscape. If consumers start asking an AI agent to shop, book flights, negotiate bills or manage appointments, the traditional website and app could increasingly become the backend rather than the destination. That creates an obvious threat to businesses that depend on consumers directly visiting their platforms.

The bigger question is trust. Muse can connect to email, shopping and other services, while Meta has built a separate AI system called Sentinel to monitor Muse's actions. Meta says sensitive actions such as purchases and sending emails require user approval, while credentials are stored separately.

But the technology is arriving with real concerns. Reuters reported that Meta delayed Muse's launch after internal testing uncovered security and reliability problems, including an incident involving access to private iCloud photos.

The real story: Meta isn't just launching another AI chatbot. It's testing whether people are willing to hand an AI agent enough control to act as them online.