GenUI

Everything you see and can click on a screen is called a user interface: buttons, menus, input fields, tables. Normally, a human has designed and programmed this interface beforehand. That’s why it looks the same for every user, no matter what they actually want. GenUI turns this around: a program assembles the interface only at the moment you need it. It decides for itself whether you see a table, a form, or a map. The basis for this is an AI system that has learned from vast amounts of text and code which presentation fits which request.

Why chat windows alone aren’t enough

Since ChatGPT, many people have been talking to software in full sentences. This works well for questions. But for actions, plain text is often cumbersome. If you want to book a flight, a calendar you can tap is faster than the sentence “please March 14th, but if that’s expensive, then the 15th instead.”

GenUI is meant to combine both worlds. You describe your request in words, but get real interface elements back. A slider for the price, a map with hotels, a button to book. This reduces misunderstandings, because you see what the system intends to do instead of just having it described to you.

For companies, this holds an economic promise. Classic software needs its own screen for every new feature, which someone has to design and maintain. If interfaces are generated automatically, this effort decreases. Whether this actually becomes cheaper in practice, however, remains an open question.

How the interface is created at runtime

The AI doesn’t paint pictures of buttons. Instead, it generates a description of the interface in a structured form, similar to an assembly instruction. It might state, for example: a heading at the top, below it a table with three columns, at the bottom a button labeled “Confirm.” The actual app reads this instruction and builds the visible page from it.

Importantly, the AI usually doesn’t invent freely. Developers give it a fixed toolkit, a so-called component library. This contains ready-made, tested building blocks: calendar, chart, search field, warning notice. The AI only chooses which building blocks to combine, in what order, and with what data to fill them.

This is an important difference from AI that simply writes program code. When writing code, a text is produced that a human reviews and later publishes. With GenUI, the interface is created live, while you’re using it, and disappears again afterward. This is exactly where the biggest risk lies: if the model picks the wrong building block, the user sees a nonsensical screen, and no one has reviewed it beforehand.

GenUI in current products

You can already see early examples in chatbots. If you ask about the weather, a small weather card with icons appears instead of a paragraph of text. If you ask about stock prices, a chart appears. Shopping assistants also show product tiles with images and buy buttons, assembled differently depending on the request.

In the tech industry, GenUI is considered a possible next step after the chat window. Providers like OpenAI, Google, and Vercel are working on tools for this, and developers are discussing standards that would let AI systems request interface building blocks. When you read about “generative interfaces” in business news, it’s usually about this race.

A common misconception is that GenUI makes designers obsolete. The opposite is more likely: someone has to design the building blocks the AI chooses from, and set the rules for what it’s not allowed to do. The work shifts from the individual screen to the toolkit behind it.