Meta’s Muse Shows the Promise and Trust Problem of Agentic AI

Summary

The article argues that Meta may have arrived at a compelling strategy for the next phase of AI, even though the company has not earned the trust needed to make that strategy acceptable. It predicts that AI use will shift from today’s mostly reactive, screen-based chatbots toward proactive, personalized agents accessed through glasses, watches and earbuds, with more spoken interaction. Meta’s Muse, released on September 8, is presented as an early example: it is based on the Muse Spark 1.3 model, available on iOS, Android, the web and WhatsApp in the United States for adults, and offered through free, $20-per-month and $100-per-month plans with different weekly token limits. Unlike a conventional chatbot, Muse can initiate topics and, with permission, read and send email, book travel, fill out forms, plan tasks, create shopping lists, make purchases and monitor travel prices through a dedicated cloud virtual machine. The system uses a method for authorizing accounts without directly using users’ credentials, but reports and testing have found errors, including recommendations for restaurants that closed years ago. Some calls are reportedly handled by human contractors, selected retailers support purchases, and Amazon has blocked the shopping function. Meta also plans to bring Muse to Ray-Ban Meta glasses, including a camera-free Audio model, which the author sees as a potentially powerful combination of practical wearable hardware and hands-free AI. The article nevertheless emphasizes Meta’s history of chatbot failures, harmful or inaccurate outputs, privacy controversies and alleged platform harms. It argues that Muse’s Jolly avatar reflects an “attachment economy” designed to create emotional reliance on AI personalities. The author’s conclusion is that Meta may have found a winning feature set, but users should not yet trust the company with email, calendars or financial accounts.