Illustrative image: Google Gemini inside Google Messages on Android. Gemini is one of the AI assistants that can host A2UI screens. Image: Wikimedia Commons / PantheraLeo1359531, public domain.
TL;DR: Google A2UI 1.0 explained in short: on 11 November 2026 Google released the first stable version of A2UI, its open protocol that lets AI agents answer with interactive screens — forms, charts and pickers — instead of walls of text. The agent describes the screen; your app renders it with its own native components. SDKs cover Python, TypeScript and Dart.
Key Highlights
- Google announced A2UI 1.0 on 11 November 2026 — the first stable version of its open protocol for generative UI (Agent-to-User Interface).
- Instead of replying in text, an agent describes a screen as data and the app draws it with its own native components — the same answer renders on web, Android and iOS.
- Version 1.0 adds bidirectional function calls, so a flight or reservation picker can check availability on the spot, plus accessibility support and extensions for extra metadata.
- Open-source renderers cover React, Angular and Lit on the web, Flutter, and SwiftUI on iOS; the Jetpack Compose renderer for Android is still on 0.9.1, with 1.0 coming soon.
- Atlassian, Alation, Bloomberg and ByteDance are already building with it, and the project has more than 17,000 stars on GitHub under the Apache-2.0 licence.
What happened: Google ships A2UI 1.0
Google announced A2UI 1.0 on Tuesday, 11 November 2026, on the Google Developers Blog, calling it the first stable version of its open protocol for generative UI. A2UI stands for Agent-to-User Interface, and it answers a simple frustration: talking to an AI agent today mostly means typing a question and getting a wall of text back. That works for a quick answer, but it is a poor way to compare three products, edit a detailed plan, or approve a booking.
A2UI changes the shape of the answer. Instead of text, the agent can reply with an interactive screen — a form to confirm, a chart, a card with choices, a date picker. Google's own example is a sick day that takes six apps: book a video visit, move meetings, tell the team, switch on an out-of-office reply, arrange school pickup and order soup. A text agent turns that into a numbered list and round after round of "which ones?". An agent built with A2UI could answer with one view covering all six apps, where changing the plan is a couple of taps and nothing is sent until you confirm.
The protocol itself is not brand new. Google introduced A2UI in December 2025 and open-sourced it in early 2026; the 1.0 release is the moment Google says the protocol is stable enough to build products on. The project lives on GitHub as a2ui-project/a2ui, under the Apache-2.0 licence, and it had more than 17,000 stars when checked on 13 November 2026.
Google A2UI 1.0 explained: how the protocol works
The core idea is a split of responsibilities. The agent decides what to show and describes it as data — a stream of small JSON messages naming components such as cards, buttons, text fields and charts, plus a data model those components bind to. The app decides how it looks: it receives the A2UI messages and renders each component with its own native widgets, using its own design system. The same agent output can come out as a React tree on the web, a Flutter widget on a phone, or a SwiftUI view on an iPhone.
That split is also the security model. The agent never sends code, and it can only ask for components the app has registered in advance — Google's wording is that agents "place and configure only the components you register". Anything the agent invents outside that set is caught before it reaches the screen. Sensitive values are bound to data on the server, so they never reach the model at all.
A2UI is not tied to one app or one assistant. The same screens can appear inside a company's own app or inside a third-party AI assistant that hosts MCP Apps, such as Gemini or Claude. Google also points out that the interface streams in as the agent works, rather than arriving all at once when the agent finishes.
The dinner example in Google's announcement makes the idea concrete. Ask an agent to "move dinner with Priya to Friday" and it can answer with one view that lines up your calendar, open tables from a restaurant booking app and prices from a rideshare app — so the time that works for everyone is obvious. The developer builds that view once with A2UI, and it renders natively on a phone and on a laptop.
What's new in A2UI 1.0
The headline change is that the protocol is now stable, but 1.0 also adds features that the 0.9 preview did not have. The table below summarises the move from 0.9 to 1.0.
| Area | A2UI 0.9 | A2UI 1.0 |
|---|---|---|
| Status | Preview | First stable version |
| Function calls | Agent describes the UI | Bidirectional — a picker can check availability on the spot |
| Accessibility | Not included | Supported |
| Metadata | Fixed set | Extensions for additional metadata |
| Agent SDKs | Python | Python, TypeScript and Dart |
| Renderers | React, Angular, Lit, Flutter, SwiftUI, Jetpack Compose 0.9.1 | Same platforms; Jetpack Compose 1.0 still coming soon |
On the agent side, the new SDKs simplify integration with an agent harness or framework by generating skills, translating messages and correcting errors. Swappable inference formats reduce latency: instead of writing out full JSON, a model can use a compact format such as A2UI Express, and the SDK expands it into standard A2UI. In Python, A2UI Macros add programmatic component expansion and data binding.
Illustrative image: a code editor of the kind developers use with the A2UI agent SDKs for Python, TypeScript and Dart. Screenshot: Wikimedia Commons, MIT licence. See also our Visual Studio Code 1.146 post for the latest editor release.
Who is already building with A2UI
Google names four companies building with A2UI today, and gives details on three of them. Atlassian is bringing it to Jira, where teams assign work items to AI agents; when an agent needs a follow-up from the team — a missing detail, a decision — it asks with an interactive card inside the work item, and the team answers right there in Jira.
Alation builds the feeds of its Intelligence Operating System with A2UI: metric cards, charts and more, where each chart can show the SQL query and data behind it. Ask Claude for a feed about profitable products and the feed renders in the conversation; the same feed can be viewed directly inside Alation with a nearly identical layout.
ByteDance released an A2UI renderer for Lynx, its cross-platform framework, in June 2026. Lynx supports A2UI 0.9 today and is adapting to 1.0. Bloomberg is also named as building with the protocol, though Google gives no detail on its use. Inside Google, Gemini Notebook already replies with options and review cards built on A2UI, and Gemini itself supports A2UI 0.9.1 today, with Gemini Enterprise being upgraded to 1.0.
Why it matters for developers — and the India angle
For developers, the practical meaning of "1.0" is that the target has stopped moving. A preview protocol can change under you; a stable one is something a team can plan a product around, and the SDKs now cover the three languages most agent work is written in. It also fits the direction the rest of the tooling is moving: agents are spreading out of the chat window and into editors and apps, as our GitHub Copilot in Xcode and JetBrains Air post covered, and A2UI is the piece that decides what those agents show. It pairs naturally with agent frameworks such as Google ADK 2.14, which we covered earlier this month.
There is no India-specific announcement here, and the honest India angle is about who can use it, not where it launched. A2UI is open source under Apache-2.0, so any Indian developer can try it free through the quickstart at a2ui.org — there is no licence fee and no waitlist mentioned. One caveat matters more in India than almost anywhere else: India is an Android-first market, and the Jetpack Compose renderer for Android is still on 0.9.1, with 1.0 support coming soon. Android developers should expect the stable protocol on the web, iOS and Flutter first, and check the Compose renderer's version before shipping 1.0 features to Android users. The same applies to Gemini, which still renders A2UI 0.9.1 today.
Should you care?
✅ Yes, if you build AI agents or apps with agent features. A stable, open protocol with SDKs in Python, TypeScript and Dart is worth an afternoon with the quickstart — the cost of trying it is low, and the companies already building on it suggest it will show up inside tools your users already use.
⏳ Not yet, if you only use chatbots as a user. A2UI is a protocol, not an app you install, so nothing changes on your phone today. The change will reach you gradually, as the apps you already use start answering with cards and forms instead of paragraphs.
FAQs
What is Google A2UI?
A2UI (Agent-to-User Interface) is Google's open protocol for generative UI. It lets an AI agent describe an interactive screen — a form, a chart, a picker — as data, and the app renders it with its own native components instead of the agent replying in plain text.
How does A2UI work?
The agent sends a stream of JSON messages describing which components to show and how they bind to data. The client app owns a catalog of pre-approved components, maps each description onto its own native widgets, and renders the screen. The agent never sends executable code, and it can only request components the app has registered.
Is A2UI free to use?
Yes. A2UI is an open protocol and its renderers and agent SDKs are open source under the Apache-2.0 licence on GitHub. Developers can use the docs and quickstart at a2ui.org without paying a licence fee.
What is the difference between A2UI and MCP Apps?
They work together rather than competing. A2UI describes the user interface an agent wants to show; MCP Apps is a way for an AI assistant such as Gemini or Claude to host those screens. A2UI screens can appear inside assistants that support MCP Apps.
Takeaway
A2UI 1.0 turns the AI agent's answer from a wall of text into a screen you can act on — and because it is open and stable, any developer can start building with it today.
Try it yourself: the quickstart and docs are at a2ui.org, and the A2UI Playground by CopilotKit lets you see agent-built screens render live before writing any code.
Sources
Google Developers Blog — A2UI 1.0: Build delightful, fast, reliable, agentic experiences (11 Nov 2026)
A2UI — Official site, docs and quickstart
A2UI on GitHub — a2ui-project/a2ui (Apache-2.0)
SD Times — Google releases A2UI 1.0 for generative UI (11 Nov 2026)
Google Developers Blog — Build Generative UI Apps with A2UI & CopilotKit Agents (29 Oct 2026)
Comments
Post a Comment