The chat interface just learned how to stop talking and start showing — and that changes what an agent can do for you.
The Summary
- OpenAI launched GPT-6 with "Intelligent UI," training the model to decide when to answer with interactive visuals instead of walls of text
- ChatGPT now generates diagrams, charts, forms, and tappable buttons directly in responses — rolling out to all users globally
- This isn't a UI redesign, it's the model learning to format its own output based on what the question actually needs
The Signal
For two years, ChatGPT has been trapped in the tyranny of the text box. You ask about bicycle gears, you get paragraphs. You ask for a budget breakdown, you get prose. The interface has been a bottleneck — not because OpenAI couldn't build better UI, but because the model could only output one thing: words.
GPT-6 changes that. The model now decides when to generate interactive visuals over text and formats them in its response. Ask about a seven-speed bicycle design, and ChatGPT shows you a diagram with buttons that highlight different components when you tap them. The training didn't just teach the model what to say — it taught the model how to present.
"OpenAI trained GPT-6 when to generate interactive visuals over text, as well as how to format them."
This is the first real glimpse of what interface design looks like when the agent controls it. Not a human designer choosing where buttons go, but the model itself deciding that this question needs a chart, that one needs a form, this other one just needs three clean bullets. The interface becomes part of the inference.
The implications compound fast. If the model can format its own UI, it can customize responses per user without a developer writing conditional logic. If it can generate buttons, it can generate actions those buttons trigger. If it can show you a diagram, it can let you edit the diagram and regenerate based on your changes.
Three things this unlocks:
- Agents that don't just answer questions but scaffold workflows inside the chat
- Personalized interfaces that adapt to how you think, not how the app was designed
- Faster task completion — you tap a button instead of typing a follow-up prompt
The rollout is global and available to all ChatGPT users, which means OpenAI is betting this isn't a power user feature. It's the new baseline. The chat interface isn't going away, but it's no longer the only tool in the box. GPT-6 learned to reach for the right one.
The Implication
Watch what developers build on top of this. If the model can generate UI elements in the chat, it can generate them in your app, your dashboard, your workflow tool. The next wave of agent products won't just automate tasks — they'll design their own interfaces for each task on the fly.
For anyone building agent tools: the static UI just got a lot less defensible. If your product's value is "we put a nice interface on top of the AI," you're now competing with the AI's own interface instincts. Build for depth, not presentation.