Meta's Muse AI can't conquer the hassle of my kids' frustrating school apps

Handing over personal information such as health data, credit cards, and email access to a technology platform may seem risky, especially when that platform is operated by Meta, a company whose track record has often raised privacy concerns. Despite such reasonable apprehensions, a new AI agent from Meta, called Muse, has been met with a notably positive reception since its launch earlier this month. Muse is positioned as a user-friendly digital assistant capable of handling routine tasks typically viewed as tedious or time-consuming.

Unlike predecessors or rival AI services that focus on text generation or search alternatives, Muse’s functionality is centered on practical, everyday assignments. Tasks like organizing a cluttered email inbox or identifying unnecessary subscriptions are completed through simple prompts. In one test, Muse efficiently sifted through large volumes of emails, delivering a daily digest of messages flagged as important - a welcome relief for inboxes saturated with unwanted content. The AI also reviewed financial statements to suggest which subscriptions could be canceled, in one case targeting a Flickr membership.

The platform’s capabilities were further tested with more complex family logistics, such as managing the mountain of digital paperwork and scheduling that comes with school-age children. Muse was asked to plan a child's birthday party at a local trampoline park, a task that involved checking venue availability and issuing invitations. This ultimately hit a snag as too many decisions required direct human input and phone calls, underscoring the current limitations of AI in handling multi-step, nuanced tasks that involve human negotiation.

Muse performed better in consolidating notifications from various schools and activity apps. Its ability to flag important emails was effective but was hampered by the labyrinth of platforms and single sign-on barriers, like attempting to access forms hidden within ParentSquare that require Google authentication. Meta’s team provided troubleshooting assistance, but such concierge support will not be feasible for the typical user at scale.

When tested on accessing more sensitive information, such as health data via Apple Health, Muse offered generic but accurate health advice, specifying that the user’s resting heart rate was already healthy but could be improved with better sleep. Attempts to automate medical appointment bookings and periodic delivery orders, such as ordering cookies timed to a user’s menstrual cycle, highlighted persistent roadblocks. Muse could not complete either task due to issues with insurance forms and third-party app data integration.

Despite some partial successes - such as a colleague’s ability to dispute a health bill through the platform - Meta acknowledges that intentional friction exists within Muse. Jimmy Raimo, a spokesperson for Meta, explained that Muse is programmed to stop tasks if unexpected fees or questionable logins appear, prioritizing user safety over unfettered automation. “We’d rather it ask than get it wrong,” Raimo told Business Insider.

Ultimately, while Muse represents a credible step toward the broader adoption of personal AI agents, its value is currently limited by the intricacies of daily life, the range of third-party technology barriers, and the need for user discretion in assigning tasks. For now, while nothing attempted was irreversible or hazardous, the experience highlights how both human and technical factors shape the promise and limitations of AI tools in real-world scenarios.