Meta Launches Muse AI Agent for Proactive Tasks

Meta released its new personal AI agent, Muse, on September 8 to positive initial reviews. Powered by the Muse Spark 1.3 model, the application is currently available to adults in the United States via iOS, Android, WhatsApp, and the web at muse.ai. Meta is offering three ad-free pricing tiers: a free tier providing up to 100 million tokens per week, a Power tier at $20 per month for 500 million tokens, and a Maximum tier at $100 per month for 3 billion tokens.

Unlike traditional reactive chatbots, Muse is proactive, initiating tasks independently. Operating within a dedicated cloud virtual machine called the Muse Secure VM, which features its own browser, the agent can access user accounts without using direct credentials. Once authorized, Muse can read and send emails, book travel, negotiate lower bills, fill out forms, and manage shopping lists. Retailers including Walmart, Best Buy, American Eagle Outfitters, DICK'S Sporting Goods, Fanatics, Gap, Michael Kors, Sephora, Ulta, and Wayfair are facilitating purchases through the agent, though Amazon has blocked it. It competes with proactive tools like Gemini Spark, Alexa+, and Claude Dreaming.

At its Connect 2026 event, Meta announced plans to integrate Muse into its Ray-Ban smart glasses. This includes the newly unveiled Ray-Ban Meta Audio glasses, which feature speakers, microphones, and touch controls but omit cameras to address privacy concerns. Meta also introduced the Muse Charm, a pocket-watch-shaped device featuring the agent's cartoon avatar, Jolly. However, reports indicate the system still makes errors, such as recommending closed businesses, and some of its automated phone calls are actually placed by human contractors.

For developers and enterprise practitioners, Muse represents a shift toward the "attachment economy," where AI tools foster emotional connections to drive user loyalty. However, Meta's history of data privacy scandals—including the Cambridge Analytica leak of 87 million users' data—and past chatbot failures like BlenderBot 3 and Galactica present significant trust hurdles. Practitioners must weigh the convenience of proactive, hands-free agentic workflows against the security risks of granting an AI agent deep access to sensitive personal and financial accounts.