Google has shipped A2UI 1.0, the first stable version of an open protocol that lets an AI agent reply with a form, a chart, or a picker instead of a paragraph. The Google Developers Blog announced the release on November 11. The bet is that agent output stops being something users read and becomes something they tap.

The design splits the job in two. The agent sends a description of what to show, and the host app draws it with its own native components. A developer registers the components the agent may use, and the agent can pick and set up nothing outside that list, so the brand’s design system stays in charge. Sensitive values are bound to data on the server, which keeps them away from the model.

Tooling covers most stacks. On the web, renderers exist for React, Angular, and Lit. Flutter handles phones, browsers, and desktops, SwiftUI draws screens on iOS, and Jetpack Compose serves Android. Agent SDKs now cover Python, TypeScript, and Dart. Version 1.0 itself lets the interface call functions in both directions, so a booking widget for flights or tables can query availability while the user is looking at it. It also adds accessibility support and extensions for extra metadata.

Distribution is where the commercial argument sits. The web and Flutter renderers work inside MCP Apps, so A2UI screens can appear in assistants that host them, including Claude and Gemini. Alation’s Intelligence Operating System already does this. Ask Claude for a feed about profitable products and metric cards and charts render in the conversation, while a nearly identical layout runs inside Alation’s own Data Products App.

That changes what a company builds. Until now, a custom interface was a reason to keep users inside your own product. A2UI lets one view travel into someone else’s assistant and still use the company’s own components, which turns the interface from a moat into a distribution channel.

Other adopters point to the enterprise end. Atlassian is bringing A2UI to Jira, where an agent assigned a work item can ask the team for missing information or a ruling through an interactive card in the item itself. ByteDance shipped an A2UI renderer for its Lynx framework, which targets several platforms at once, back in June 2026. Inside Google, Gemini Notebook uses cards to ask what kind of slide deck a user wants before generating it.

The post leaves gaps. It offers no latency measurements, and the claims of native speed and low latency are Google’s own. The stable version also trails the protocol on several surfaces. The Compose renderer is at v0.9.1, with 1.0 listed as coming soon. Lynx supports 0.9 and is adapting to 1.0. Gemini supports 0.9.1 today, and the upgrade of Gemini Enterprise to 1.0 is still in progress.

For teams already on 0.9, Google points to an evolution guide that covers every change. For everyone else, the mismatch is the practical risk: a team planning an agent launch on Android or inside Gemini in the next quarter should budget for a version bump instead of assuming 1.0 behavior on every surface.

Reported by the Google Developers Blog on November 11, 2026.