Key Takeaways

  • Meta designed its AI agent Muse by abandoning enterprise developer patterns like files, diffs, branches, and code threads in favor of familiar messaging conventions.
  • Alex Cornell grounded every interface decision in a single test: how two people would actually interact if they were speaking in person.
  • Muse uses message bubbles and embodied character avatars to establish the product as an exchange between two entities rather than a tool running commands.
  • Companion agents invert standard user flows by messaging users proactively, shifting software from reactive search boxes to relationship-driven surfaces.
  • Non-deterministic software requires progressive disclosure status updates to inform users without cluttering the screen with raw system output.

Strip the Code Editor Out of Consumer Software

Most consumer AI products still look like text-heavy developer terminals. They display raw markdown streams, file trees, and execution traces inherited from technical coding environments. When Cornell led design on Meta's Muse, his team deliberately threw those patterns out.

Mainstream consumers do not think in diffs, branches, or code threads. Handing everyday users a version-control mental model creates instant friction. Muse replaced developer artifacts with mobile messaging standards: visible message bubbles, conversational status updates, and separate tabs for ideas, chats, and feeds. Cornell noted that “we have bubbles around our messages you know and like that's typically something that you associate more with messaging than like an AI app.”

Putting rounded bubbles around text changes how users process the screen. Bubbles frame output as words sent by someone else, not system logs generated by a server. Host Claire Vo pointed out the deliberate contrast in Muse: “you have this like character first of all as a primary entity. You have chats, you have ideas, you know, you you have these um you have your feed like you you've just chosen different aspects to elevate.” By elevating social and conversational elements over file management, the interface becomes immediately approachable.

Design for an Entity, Not an Input Box

The standard AI design pattern is passive: an empty input box waiting for a prompt. Cornell treated Muse as an autonomous entity rather than a command prompt.

“The basic framing that always we went back to and I think was probably the most helpful was just imagining that this interface was more of an exchange between you and an entity,” Cornell explained. Whenever the design team hit an ambiguous product fork, they asked a simple question: “It was more like how would you handle this if you were actually speaking with another person, you know, and that was immensely clarifying, I think, in almost every decision that we had to make.”

That mental model immediately changes interaction patterns. People do not wait silently in a room until spoken to, and companion agents should not either. Cornell highlighted that “simple things like the fact that an agent can message you proactively changes the nature of like how you think about the product from a design perspective.”

Proactivity breaks the old search-engine paradigm where the human does all the work to initiate value. To manage the uncertainty of non-deterministic models, Cornell implemented progressive disclosure status updates. Instead of burying consumers under technical execution traces or hiding background work entirely, conversational status updates tell users what the agent is doing in plain language. You get clarity without clutter.

What to Do With This

Audit your consumer AI product's primary screen tomorrow morning. Strip out every developer artifact, including raw execution logs, exposed file paths, and terminal-style markdown. Replace them with conversational status lines, message bubbles, and at least one proactive touchpoint where the agent initiates the next interaction.