Google has released A2UI 1.0, the first stable version of its open protocol for generative user interfaces, giving AI agents a way to answer with interactive screens instead of long blocks of text.

The release, announced on the Google Developers Blog on 11 November, is aimed at the growing number of agents that live inside apps and assistants. Until now, an agent asked a question usually replied with a paragraph. A2UI lets it reply with a form to confirm, a chart, a booking picker or a comparison view, drawn with the host app's own native components.

What A2UI changes

Under the protocol, the agent describes the screen it wants and the app draws it. That separation matters: the agent never sends code, and it can only place and configure components the app has registered. The same agent output can render on the web, on Android and on iOS, and inside AI assistants that support MCP Apps, such as Gemini and Claude.

Google's pitch is that this ends the wall-of-text problem. Its example is a sick day: booking a video visit, moving meetings, telling a team, switching on an out-of-office reply and ordering soup currently takes six apps. A text-only agent turns that into round after round of questions; an A2UI agent can put the whole plan in one view, and nothing is sent until the user confirms.

What is new in version 1.0

The headline change is a stable protocol, so developers can build on it without expecting breaking changes. Version 1.0 adds bidirectional function calls, meaning a flight or reservation picker can check live availability on the spot rather than showing stale options. It also adds support for accessibility and extensions for extra metadata.

Speed and safety are part of the design. Agents can compose interfaces with A2UI Express, a compact format, or with decision models that pick components instead of writing text. Data binding happens on the server, so sensitive values never reach the model, and the interface streams in as the agent works.

Developers can use the release now with open-source renderer software for the web (React, Angular and Lit), Flutter for mobile, web and desktop, SwiftUI for iOS, and Jetpack Compose for Android, plus agent software for Python, TypeScript and Dart. Documentation and a quickstart are published at a2ui.org.

Who is already building with it

Google names Atlassian, Alation, Bloomberg and ByteDance among the companies already building with A2UI. Atlassian is bringing it to Jira, where an agent working on a task can ask the team a follow-up question through an interactive card inside the work item itself. Alation builds the feeds of its Intelligence Operating System with A2UI, including metric cards and charts that show the query and data behind each figure, rendered identically in its own app and inside Claude.

ByteDance released an A2UI renderer for Lynx, its cross-platform framework, in June 2026, and is adapting it to version 1.0. Its playground demo renders a two-day trip planner for Kyoto from A2UI messages. Inside Google, Gemini Notebook already uses the approach for slide-deck options and review cards, and Gemini can render responses from registered agents, currently on version 0.9.1.

What is still to come

Not every piece is at version 1.0 yet. The Jetpack Compose renderer for Android still supports 0.9.1, with 1.0 support coming soon, and Gemini Enterprise is being upgraded to 1.0. Google's roadmap covers design-system primitives that render natively on every platform and official support for decision models such as Jev.

For developers, the release is a signal that the agent era is moving beyond chat. The question is no longer only what an agent can say, but what it can show, and whether what it shows can be trusted, tapped and confirmed in the app where the work happens.