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

Alex Cornell is a design director at Meta, where he leads design on Muse, Meta’s consumer AI agent. He came to the role with a long background in software product design, including time as a founder, and has spent the past year building one of the most thoughtful agent interfaces in the consumer AI space. Muse ships with a custom avatar, a three-layer status display, a personalized ideas surface, and a generative feed. Alex has been sharing the reasoning behind each of those design decisions publicly on Threads. What you’ll learn: Why Alex built Muse to feel like messaging, not an AI app How Muse’s three-layer status system accommodates different levels of technical expertise What generative UI actually means in a shipped consumer product How Alex turned years of forgotten Instagram saves into a daily shopping catalog Why his mom apologized to her Muse, and what it means for AI design How he uses Muse to track his golf game, shot by shot, as he works toward a single-digit handicap The mindset shift designers need when the output is never pixel-perfect When to use the main chat versus a side chat, and where the architecture still falls short — Brought to you by: WorkOS —Make your app enterprise-ready, with SSO, SCIM, RBAC, and more — In this episode, we cover: (00:00) Welcome and guest intro (02:50) Designing Muse (05:33) Choosing which primitives to surface: characters, ideas, feed (08:50) Speed of model and response times (11:01) Generative UI and personalization (14:25) SkyMall demo: Instagram saves turned into a shopping catalog (17:16) Shopping, monitoring, and price alerts with Muse (19:07) Instagram and Pinterest connectors for visual data (21:40) Golf use case: round recaps, shot data, practice breakdowns (25:28) Main chat versus side chats (27:56) Old blog turned morning podcast: 20 years of posts via the Wayback Machine (29:39) Chess dashboard: game analysis with a snarky coach (30:59) What designers need to learn for AI, and what still works (34:33) Prompt