Meta Muse: Useful, Hindered by Apps and Caution

A week-long trial of Meta's Muse agent, with access to emails, bank cards, and Apple Health data, yielded some successes and several dead ends. Meta has implemented safeguards that stop the agent in case of doubt, while missing integrations further limit everyday use cases.

Meta Claims Deliberate Friction to Avoid Errors

Meta states that Muse does not yet cover all use cases. Its spokesperson, Jimmy Raimo, asserts that part of the friction is intentional. If any fees seem unusual or if a connection appears incorrect, Muse is designed to stop and verify rather than proceed. Meta prefers that the agent ask questions instead of making mistakes. In the reported trial, the failures did not have serious consequences, as tasks were resumed on a computer. The user expressed a preference for these blocks over undesirable actions like spending $5,000, messaging all contacts, or signing up for a marathon.

Concrete Support Points: Email Sorting and Subscriptions

For well-defined tasks, Muse proved useful. It organized the inbox and provided a daily summary of emails deemed important. The agent also analyzed payment methods to suggest subscriptions to cancel, notably mentioning Flickr. Beyond text or simple searches, it can be dispatched for practical requests like sorting emails or searching for equipment for a trip.

Integration Roadblocks: Connections, Messaging, and Forms

In the educational and community context, the results were mixed. Tasked with tracking school and activity notifications, Muse successfully detected items received via email but struggled with more complex information. A Google registration form, accessible via ParentSquare and linked to the user's Google account, could not be reached because Muse's browser interface did not offer the Google login option. The agent also failed to retrieve the Girl Scouts' schedules, which were shared in a WhatsApp group that it cannot read.

Health: Accurate Analysis, Appointment Scheduling, and Logistics Failures

After connecting Apple Health and fitness data, Muse assessed a resting heart rate deemed acceptable and suggested improving sleep. However, scheduling the annual check-up, which included information like the date of birth and the last four digits of the social security number, failed during the insurance update, with the agent redirecting to a call to the office. Another dead end: scheduling a cookie delivery via Walmart+ two days before each cycle period was not possible, as Apple cycle data did not sync with Muse, and the agent did not connect to the other cycle tracking app.

Between Perceived Potential and Daily Adoption Conditions

Muse, Meta's personal agent launched earlier this month, was used for a week with access to emails, bank cards, and Apple Health data. The user believes it will improve and sees potential, while noting the failure of a birthday organization task that required phone calls. A colleague, however, reported successfully disputing a medical bill via Muse. According to the user, a significant part of the difficulty lies in the ability to formulate and delegate tasks suited for agents, which requires knowing the right use cases and being creative.