Meta’s Muse AI took over real life tasks, here’s where it failed
One writer tested Meta's Muse AI assistant on email, credit cards, health data, and kids' logistics
Handing an AI agent your email, credit cards, and health data sounds like the setup to a cautionary tale. One writer did it anyway, testing Meta's new Muse AI agent against a stack of real, messy personal tasks to see whether it could actually lighten the load.
Muse launched earlier this month with a warmer reception than Meta's earlier AI misfires, including the largely forgotten video generator Vibes. Unlike a typical chatbot used for text generation or search, Muse is built to execute tasks directly, things like "organise my inbox" or "find a travel stroller", rather than just answering questions.
Early tests went smoothly: Muse sorted a cluttered inbox into a daily digest of important emails and scanned credit card statements to flag an unused subscription worth cancelling.
A more challenging test was in children’s administration, where there is the greatest opportunity for AI agents to make an impact in theory. The Muse could not organise a trampoline park birthday party, as it relied heavily on human judgement, and failed to complete a school signup form that was blocked by a Google login.
It was unable to read a Girl Scouts organising discussion because the organisation lived in WhatsApp, which Muse could not see, and failed completely at renewing a Girl Scouts membership when the payment page timed out.
Connected to Apple Health data, Muse reviewed a resting heart rate and suggested better sleep, a reasonable if unremarkable answer. Booking an annual physical went worse: after collecting a date of birth and the last four digits of a Social Security number, Muse still couldn't get past an insurance verification step.
A recurring cookie order tied to a menstrual cycle also failed, since period-tracking apps don't sync with Muse, though the agent did catch itself adding unwanted same-day delivery fees before the order was cancelled.
