Google releases A2UI 1.0 for AI agents to show screens, not text

Google released A2UI 1.0, the first stable version of its open protocol for generative UI. This is a screen that an AI agent makes. The agent tells the app which parts to show, and the app renders it with its parts. Atlassian and ByteDance use it, and kits for web, Android and iOS are available.

How it works

The agent sends a short description of a screen, for example a form or a chart. The app renders it with only the parts that the developer selects. Sensitive values stay on the server and do not go to the model.

The numbers

Google gave no test results or speed data. It says the screens render at native speed. Atlassian uses it in Jira, and ByteDance renders it in its Lynx tool.

What is not known

Other Google features with A2UI are not in use. The change of Gemini Enterprise to version 1.0 is in progress. No test result is available.

Insight: Google looks to set the standard for agent screens

The format works in Gemini and in Claude, and an agent can use only the parts that the owner selects. On 15 October 2026, Descope added an owner and an approval step for each agent. This seems to try to set the standard for how agents show screens, and the first format that many apps use can set it. Builders who wait can find it is set.