Google has released A2UI v0.9, a framework-agnostic standard that lets AI agents declare user interface intent and render it natively across web, mobile, and desktop without shipping arbitrary code. Announced on the Google Developers Blog, the release reframes generative UI around a simple idea: agents should speak the language of an application's existing design system rather than invent new components of their own.
The headline change is a shift in philosophy. According to CopilotKit, v0.9 is not a minor update, with the core philosophy, the JSON structure, and the schema all changing, and the protocol becoming bidirectional. The optional component set previously called "Standard" has been renamed "Basic" to signal that frontend developers should connect agents to the components they already own. On the client side a shared web-core library now underpins browser renderers, and the release lands an official React renderer alongside version-bumped Flutter, Lit, and Angular renderers, plus a dedicated home for community renderers.
The agent side gets its own toolkit. The new A2UI Agent SDK adds caching layers for lower latency, and adding it to a Python agent is a single install:
pip install a2ui-agent-sdk
The SDK handles version negotiation, dynamic catalogs that switch schemas at runtime, and resilient streaming that incrementally parses and heals partial LLM output so components render as they arrive. New language features include client-defined functions for validation, client-to-server data syncing for collaborative editing, improved error handling, and a simplified, modular schema. Transports were also streamlined, with A2UI now able to run over MCP, WebSockets, REST, AG-UI, and the newly launched A2A 1.0.
Migration guidance is published. The v0.8 to v0.9 evolution guide documents the move to a "prompt-first" schema embedded directly in the model prompt, the refactor into modular files, and a property rename quick reference. Teams should note that the current production release is actually v0.9.1, a patch in the stable v0.9 family, and that a v1.0 release candidate specification is now available on the website, complete with its own v0.9.1 to v1.0 evolution guide.
CopilotKit's Atai Barkai called v0.9 "a BIG improvement, with AG-UI support out the box" and described it as mass-adoption-ready. Others are skeptical. A Hacker News commenter asked why anyone would trust an LLM to output a UI:
Why the hell would anyone want this? Why on earth would you trust an LLM to output a UI? You're just asking for security bugs, UI impersonation attacks, terrible usability, and more.
While a Reddit post argued the catalog model is too restrictive and warned that "every UI will become the same." Architect Brian Love proposed a pragmatic middle ground of fixed catalogs with dynamic overlays and deterministic fallback when validation fails.
A2UI is not alone in this space. It sits among AG-UI, MCP Apps, Vercel's json-renderer, and Oracle's Agent Spec, while web-focused projects like syntux argue that chatbot-style generative UI produces "disposable" interfaces that struggle with the consistency and cacheability real web apps demand. For now, Google is positioning A2UI as the portable contract sitting beneath all of them.
A2UI is an open, framework-agnostic specification developed by Google for declaring generative user interfaces across web, mobile, and desktop, with the full v0.9 specification, evolution guides, and official React, Flutter, Lit, and Angular renderers published at a2ui.org, and the A2UI Agent SDK available for Python via PyPI.