Google A2UI 1.0 Goes Stable With Native UI for Agents

A chat window full of text may be the dumb-terminal phase of AI agents. Google's A2UI 1.0 is one attempt to end it, and the idea is plain: the agent describes the screen, and each app draws it with its own native components.

The spec went stable on Nov 11, and the same output can run in your own app or inside Claude and Gemini through MCP Apps. You describe the interface once, and the host renders it.

What Happened

A2UI 1.0 is now stable, marked Nov 11. Instead of an agent answering in plain text, it describes the screen it wants, and the app that receives that description renders it using its own native components.

Because the description is the output, the same result can be shown in your own app or inside Claude and Gemini through MCP Apps. One description, two very different places to display it.

Why It Matters

Text-only agents are the terminal era, and this is one way out of it. An agent that can describe a screen can put real controls in front of the person reading, instead of handing them a wall of markdown to interpret.

Rendering on the app's side is the part worth watching. The interface still looks like the app it lives in, not like a stranger's widget pasted on top, and the agent only has to say what it wants shown.