How Meta designed Muse: an AI agent people actually want to talk to | Alex Cornell (Design Director)

Designing effective AI agent experiences requires shifting from traditional software-centric interfaces to human-like, conversational exchanges. Meta’s Muse prioritizes accessibility through progressive disclosure, allowing users to view high-level status updates while hiding complex technical logs unless needed. By treating the agent as an entity capable of proactive communication, designers can reduce user anxiety regarding latency and system progress. Practical applications of this agentic model include personal data synthesis, such as organizing Instagram saves into shopping catalogs or consolidating fragmented hobby data like golf statistics and old blog posts into actionable insights. Successful agent design hinges on balancing non-deterministic system behavior with intuitive, personalized UI elements that provide clear feedback, ultimately transforming mundane data management into a delightful, goal-oriented experience.