MONDAY, NOVEMBER 16, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  Dev
Dev

Google's A2UI 1.0 Lets AI Agents Answer With Buttons Instead of Walls of Text

Google released A2UI 1.0 on November 11, the first stable version of its open protocol for generative UI, letting agents answer with interactive screens your app renders natively.

Google's A2UI 1.0 Lets AI Agents Answer With Buttons Instead of Walls of Text
Source: Martin Vorel (CC BY-SA 4.0) via Wikimedia Commons

A2UI 1.0 turns agent answers into interfaces

A2UI 1.0 is Google's stable open protocol for generative UI, released on November 11, 2026, and it solves a problem every agent developer has hit. Agents today answer almost everything in text. Ask an agent to compare three products, reschedule a dinner, or approve a plan, and you get a wall of words. A wall of text is a poor way to compare three products or edit a detailed plan. A2UI lets the agent answer with an interactive screen instead: a form to confirm, a chart to read, a picker to tap.

The idea is simple and deliberately boring, which is a compliment for a protocol. The agent sends JSON messages describing an interface. The application renders that description with its own native components. The agent never sends HTML or JavaScript, and it can only use the components the developer registered in advance. Google's phrase for this is safe like data, but expressive like code. The host keeps control of what gets drawn and what a tap is allowed to do.

What changed in A2UI 1.0

The headline addition in 1.0 is bidirectional function calling. The agent and the renderer can now request functions from each other and return structured results or errors. That sounds abstract until you picture the example Google gives: a flight or reservation picker that checks availability on the spot, instead of showing you options that no longer exist. A single interface can also combine components from multiple catalogs, so one screen can mix a calendar from one app, open tables from a restaurant booking app, and prices from a rideshare app.

The 1.0 specification also adds accessibility support and extensions for extra metadata, and it removes theme colors from the protocol so styling comes from the target framework's native theme. That last change matters more than it sounds: it means an A2UI screen looks like it belongs in your app, not like a web page stapled into it.

Where A2UI 1.0 works today

The official SDK documentation lists 1.0 support for React, Angular, and Lit on the web, Flutter on mobile, web, and desktop, and SwiftUI on iOS. Agent SDKs are available for Python, TypeScript, and Dart. Jetpack Compose on Android supports 0.9.1, with 1.0 still to come, so Android developers should check the version matrix before they plan around it. A2UI can also run inside an MCP App, where the MCP App contains a web renderer that turns A2UI messages into an interface inside the host's isolated frame.

The project is Apache 2.0 licensed, created by Google with contributions from CopilotKit and the open-source community, and the repository had passed 17,000 stars by November 12. Companies such as Atlassian, Alation, Bloomberg, and ByteDance are already building with it, according to Google's announcement. That list is the strongest signal that this is not a demo protocol: enterprises adopt UI standards slowly, and they do not adopt them at all unless the standard solves a real integration problem.

Why this matters for agent developers

A2UI arrives in a week when the agent tooling stack filled in fast. Claude Code moved MCP tools to the background, VS Code 1.146 gave AI coding agents an inbox, and Google shipped Gemini 4 Hydrogen for the agents themselves. The missing piece was the last mile: what the user actually sees and touches when an agent does something on their behalf. A2UI is Google's answer to that piece, and it is an open answer, which means it can work across hosts instead of locking developers into one assistant.

There are fair criticisms. A protocol is only as good as its renderers, and platform support still varies: Jetpack Compose lags, and not every host has upgraded. The security model also depends on developers doing their part. A2UI separates UI generation from tool authorization, so a button the agent proposes cannot perform a real-world action unless a separate execution layer authorizes it. That boundary is the whole point, but it only holds if developers respect it.

The practical move for developers is to try the quickstart at a2ui.org and build one small surface: a confirmation form, a comparison card, a picker. The protocol's own evolution guide lists the changes from 0.9 to 1.0, including new message names and catalog definitions, so existing experiments need a migration pass. If your agent currently answers in paragraphs, A2UI 1.0 is the fastest way to find out whether your users prefer screens. Most of them will.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.