This page may contain stale information. Last updated: 2026-07-03

Definition

Generative UI is an architectural pattern where AI agents dynamically produce user interface structures at runtime — declaring UI intent via structured messages rather than pre-built screens or arbitrary code execution. The client renders using its existing design system and component catalog.

A2UI Protocol (July 2026)

google released A2UI v0.9.1 (production) on July 3, 2026 — an Apache 2.0 open standard co-developed with CopilotKit:

  • Prompt-first philosophy (v0.9): schemas embedded in LLM system prompts vs. v0.8 structured-output-first
  • Message types: createSurface, updateComponents, updateDataModel, deleteSurface
  • Transports: MCP, A2A 1.0, AG-UI middleware (day-zero integration)
  • SDKs: Python Agent SDK (pip install a2ui-agent-sdk), React renderer, Angular/Flutter/Lit renderers
  • v1.0 RC: client-to-server RPC (actionResponse), action IDs

A2UI separates UI structure from application data, enabling progressive rendering as JSON streams arrive — no arbitrary code execution on client.

Key Design Principles

  1. Framework-agnostic: agents declare intent; clients render natively
  2. Design system preservation: uses application’s existing component catalog
  3. Transport-agnostic: any JSON message delivery mechanism works
  4. Security: declarative UI only — no remote code execution

Sources