The Future of Design

Every interface you've ever used was designed before you opened it. Someone decided what goes where, shipped it, and hoped it worked for everyone. That's been the deal for decades — designers make layouts, developers build them, users get what they get.

I think that's about to change in a pretty fundamental way.

What's actually happening right now

There's this term "generative UI" floating around and most of what people mean by it is just chatbots with nicer formatting. A text response with a chart below it. Maybe some markdown tables. That's not it.

Real generative UI is when an AI has access to actual UI components — not HTML strings, not code snippets, real designed components — and it decides which ones to render based on what you're asking for. The interface isn't pre-built. It's composed on the fly.

Click different questions and watch the same system produce completely different interfaces:

Same components, same data, completely different output depending on what you need. The interface adapts to intent instead of being a static thing you navigate around.

Why this feels different

We've had "dynamic" interfaces forever. Dashboards with filters. Conditional rendering. Personalization engines. But all of that is still predetermined — someone sat down and mapped every possible state to a specific layout. The design system is static even if the content changes.

What's different now is that the AI is doing the composition itself. It looks at your question, understands what you're actually trying to learn, and picks the right combination of components to answer it. Nobody pre-designed that specific view. It was assembled in real time from building blocks.

And those building blocks are the key. The AI doesn't generate HTML or write CSS. It picks from a set of well-designed, purpose-built components. Toggle these on and off — every combination just works:

That's the whole trick. You design great individual components — stat cards, charts, progress rings, whatever your domain needs — and the AI just becomes really good at choosing which ones to use and what data to pass in. Any combination of well-designed primitives produces a coherent view.

What this means for design

This is the part I keep thinking about. If AI handles composition, the designer's job shifts from "arrange elements on a screen" to "design the elements themselves and the rules for combining them."

Think about it — right now, a designer might create 15 different dashboard views for different user types. With generative UI, you design maybe 10 really good components and a set of layout rules, and the AI generates thousands of possible views from those primitives. You're designing a system, not screens.

That's a massive leverage shift. The quality of your component library becomes everything. One well-designed chart component that handles edge cases gracefully is worth more than 50 hand-arranged dashboard layouts.

Where I think this goes

Right now, generative UI works best in structured domains — dashboards, analytics, admin tools — where you have clear data types and well-defined components. But the pattern extends a lot further than that.

Onboarding flows that restructure themselves based on what a user already knows. Help centers that assemble around your specific problem instead of making you dig through a knowledge base. Analytics that surface different stories depending on who's asking.

Same principle, different domains. Click around:

The limiting factor isn't the AI — it's the component libraries. Most design systems were built to be assembled by humans. They assume a designer is making layout decisions. The next generation of design systems will be built to be assembled by AI, which means they need to be more modular, more self-contained, and way more opinionated about their own layout rules.

Designers don't go away in this world. If anything, the job gets harder and more important. You're not just making things look good on one screen — you're designing primitives that need to look good in any combination, at any scale, for any question. That's a fundamentally harder design problem.

The thing nobody's talking about

Everyone's focused on the AI side of this — better models, better prompts, better tool calling. But the real bottleneck is the components. If your building blocks are mediocre, it doesn't matter how smart the AI is. You'll just get mediocre interfaces faster.

The companies that win at generative UI won't be the ones with the best models. They'll be the ones with the best component libraries. The best design systems. The best primitives.

That's a design problem, not an AI problem. And I think that's kind of beautiful.

I built a small experiment exploring some of this if you want to play with it. But honestly, we're just scratching the surface of what's possible here.