Google releases A2UI 1.0, a stable protocol for agent-generated interfaces

In A2UI, the agent describes the view it wants to show, and each app draws that view with its own native components. The app receives the A2UI messages and renders them, so the same agent output can appear natively in a company's own app or in an AI assistant that supports MCP Apps, such as Claude or Gemini.

Version 1.0 adds bidirectional function calls, so an interactive picker, such as one for flights or reservations, can check availability during the conversation. The protocol also adds accessibility support and extensions for extra metadata. Renderer libraries cover React, Angular, Lit, Flutter, SwiftUI and Jetpack Compose, with the Compose renderer at v0.9.1 until 1.0 ships.

Atlassian, Alation, Bloomberg and ByteDance are already building with A2UI. Atlassian is bringing it to Jira so agents can ask team members for missing details through an interactive card, and ByteDance has released a renderer for its Lynx framework.

Agent developers can return forms, charts and pickers that render with the host app's own components instead of long text replies. Sensitive values can stay on the server so they never reach the model. Teams building on earlier 0.9 versions should check the evolution guide before upgrading.