OpenAI's Intelligent UI Lets Agents Build Real Interfaces

Ask an AI agent for a tool and you usually get a wall of text describing it. Intelligent UI is OpenAI's new answer to that: the agent stops writing raw code and builds the interface you actually need instead.

The agent never writes raw code. It emits DIL, a new language, and the server compiles and streams it while the client renders it from a component set, so the result feels native rather than bolted on.

What Happened

OpenAI has launched Intelligent UI, a system where the agent's output is no longer plain text. The agent never writes raw code. Instead it emits DIL, a new language, which the server compiles and streams, and the client renders through a component set so the interface feels native to the app it lands in.

Alongside it there is a template for bringing interactive, generative UI into your own product, and it is built to work with any agent, not one specific model.

That template supports your design system, your components, and your real data and actions, which is the part that matters if you already have a product with its own look and its own backend.

It is available on Web, Mobile, and Slack.

Why It Matters

Most agent output today ends in a chat bubble, and you copy the useful part out by hand. Generative UI moves that work into the agent itself, so the answer arrives as something you can click, not something you have to read twice.

Because the template works with any agent and keeps your own design system and data, you are not rebuilding your front end to try it. If you build with agents, this is the direction to keep an eye on.