Key Takeaways

  • Pure generative UI creates disorientation; Alex Cornell argues consumer agents require rigid structural primitives rather than unpredictable layouts that redraw every session.
  • Dynamic personalization works best inside fixed frames, such as translating a custom avatar into predetermined action scenes like typing at a desk.
  • Meta's Muse uses a native 'Ideas' module to solve prompt paralysis, populating dynamic suggestions based on conversation history to expose hidden capabilities.
  • Consumer agents succeed by borrowing UX conventions from familiar messaging apps rather than technical developer consoles.

Why Full Generative UI Fails Consumer Apps

Design circles love discussing generative UI, where an artificial intelligence generates an bespoke screen layout tailored on the fly to every user request. In practice, completely fluid layouts confuse people. When buttons move, containers reshape themselves, and navigation patterns change on every prompt, users lose their spatial memory.

Cornell, design director on Meta's Muse, took the opposite approach. Instead of turning the software into an unpredictable canvas, his team built rigid architectural containers. As host Claire Vo observed: “Generative UI in a really elegant way which is like the core primitives stick. You've got your avatar, you've got your status message, you've got your ideas, you've got your feed, but then within that, every time I log in, I have a new set of ideas that are based on the things that we've talked about.”

Reliable software requires stable anchors. When an interface remains predictable, users feel safe exploring non-deterministic outputs. The frame stays static so the content can surprise them.

The Tomato Test: Delight Inside Fixed States

Personalization does not require inventing new layouts. Cornell found that user delight comes from mapping custom inputs into tightly scripted, pre-designed component states.

In Muse, users can customize their companion into anything they want. Cornell explained the mechanical result: “I think people find a lot of delight I have seen when they do make their own new avatar scene that it will then be translated into each one of those states, typing on its little computer or making something.” He added, “if you turn yours into a tomato, then you'll see a tomato typing on the computer. And that really does feel very personal and is I think a very delightful thing.”

The interaction works because the scene itself has strict bounds. The desk, the computer, and the physical motion are stable native components designed by human artists. The dynamic element is simply the character asset slotted into the animation. The software feels deeply personal without breaking visual consistency.

Native Surfaces for Progressive Discovery

AI agents suffer from a blank slate problem. Most users do not know what a model can do, so they default to basic search queries or quit after two prompts. Instead of expecting users to guess instructions, Muse uses pre-designed native surfaces to suggest potential actions.

Cornell points to their designated 'Ideas' surface as a primary driver of discovery. Cornell noted: “exposing some of those additional capabilities just with the additional surfaces that we have like ideas in particular. It's serving a very specific purpose which is basically just telling you some things that you may not know that it can do.”

Even the personalized feed relies on fixed rails. Cornell pointed out: “it's still a native surface though, the feed unit that you're looking at is still one that we've designed and it's being powered by the intelligence and the personalization that your muse has, but it's not yet like a fully a generative surface.”

Keep the layout predictable. Feed dynamic data into native containers. That simple rule keeps consumer agents functional.

What to Do With This

Audit your product's primary interaction loop tomorrow morning. Pick three core primitives, such as your input bar, header status, and a suggestion card, and lock their dimensions and positions permanently. Restrict dynamic generation to the data inside those cards, using the user's last three queries to populate suggestions that point directly to unused product features.