Google just released A2UI 1.0, the first stable version of its open protocol that lets AI agents generate interactive user interfaces instead of responding with walls of text.
And the same AI-generated interface can work across web apps, mobile devices, desktop applications, and AI assistants like Claude and Gemini through MCP Apps.
This could fundamentally change how we interact with AI agents.
Instead of generating HTML or JavaScript directly, an agent describes the interface it wants to display.
The application then renders that description using its own trusted, native UI components.
WHAT'S NEW IN GOOGLE A2UI 1.0?
• Generative UI: AI agents can dynamically compose forms, charts, cards, buttons, dashboards, and other interactive interfaces.
• Native rendering: applications use their own design systems and registered components instead of executing arbitrary AI-generated code.
• Cross-platform support: React, Angular, Lit, Flutter, and SwiftUI renderers are available, with Jetpack Compose currently supporting A2UI 0.9.1.
• MCP Apps integration: the same A2UI experience can appear inside compatible AI assistants, including Claude and Gemini.
• Bidirectional function calls: interfaces can communicate with applications and agents, allowing actions like checking flight availability when a user selects an option.
• Accessibility support: version 1.0 introduces accessibility metadata and extensible component information.
• Secure data binding: sensitive application data can be connected server-side without exposing it directly to the language model.
• Faster UI generation: A2UI Express uses compact representations, while decision models such as Jev can select predefined components instead of generating lengthy text.
• Developer SDKs: Python, TypeScript, and Dart support for building agent-driven interfaces.
AND COMPANIES ARE ALREADY USING IT.
Atlassian is integrating A2UI into Jira so AI agents can ask for missing information through interactive cards inside work items.
Alation uses A2UI for data feeds, charts, and SQL-backed insights that can also appear inside Claude.
ByteDance has developed an A2UI renderer for its Lynx framework.
Bloomberg is also among the companies building with the protocol.
Inside Google, A2UI already powers interactive experiences in Gemini Notebook and Gemini Enterprise, although these integrations are still using earlier protocol versions.
WHY DOES THIS MATTER?
Today, most AI agents communicate through text.
You ask for something, receive a long response, and then explain what you want to change.
A2UI introduces a different interaction model.
Ask an agent to organize a trip, and it could generate an interactive itinerary.
Ask it to analyze business performance, and it could display charts and filters.
Ask it to book a flight, and it could present available options with controls for making a selection.
The important innovation is the separation between interface generation and interface rendering.
The agent describes what should appear. The host application decides how to render it and which actions are permitted.
A2UI doesn't automatically make every agent secure or compatible with every application. Developers still need to implement renderers, components, permissions, and backend integrations.
But it provides an open foundation for building more consistent, interactive AI experiences.
And with MCP Apps support, these interfaces can move beyond a single AI platform.
Google A2UI documentation and quickstart:
https://a2ui.org/
GitHub source code:
https://github.com/a2ui-project/a2ui
Official Google announcement:
https://developers.googleblog.com/a2ui-10-build-delightful-fast-reliable-agentic-experiences/
Could A2UI become a standard interface layer for AI agents?
#GoogleA2UI #AIAgents #GenerativeUI
YouTube / ABVX
