Generative UI: interfaces that assemble themselves

A static interface assumes every user wants the same screen. Generative UI starts from the opposite premise: the interface should assemble itself around the person in front of it.

Instead of designing one fixed layout for every scenario, generative UI composes the screen at runtime from a library of trusted components, guided by intent, context, and data. The designer’s job shifts from drawing final states to defining the rules, constraints, and building blocks the system is allowed to use.

Designing for intent, not screens

Jakob Nielsen has argued that accelerating AI will push UX from static, single-purpose interfaces toward generative ones — and that, paradoxically, this makes UX the primary business moat rather than a commodity. When anyone can generate a passable interface, the advantage goes to whoever best understands what the user is actually trying to do.

That reframes the deliverable. We’re no longer shipping “the dashboard”; we’re shipping a system that knows how to build the right dashboard for someone onboarding, someone in a hurry, and an account in trouble — each from the same set of blocks.

Guardrails are the new grid

The risk of a self-assembling interface is obvious: incoherence. If the system can do anything, it will eventually do something ugly, confusing, or off-brand. The antidote is constraint. A strong design system — with clear rules about spacing, hierarchy, motion, and tone — becomes the guardrail that keeps generated screens recognizably yours.

In practice this means investing less in pixel-perfect comps and more in those contracts: what each block needs, what it must never do, and how blocks compose. The design system stops being a static library and becomes the runtime grammar of the product.

Generative UI moves the designer from author of screens to author of the system that authors screens.

Where to start

You don’t need a research lab to begin. Pick one surface where users arrive with very different goals — onboarding, search results, an empty state — and design it as a set of rules and components rather than a single fixed layout. Define the intents you’re serving, the signals that reveal them, and the blocks the system may assemble. Then watch what it builds, and tighten the constraints where it gets things wrong.

Generative UI won’t replace careful design. It raises the stakes on it. The teams that win will be the ones whose systems are opinionated enough to generate something good — and humble enough to be corrected.