What is A2UI?
Agent-to-UI lets an agent send a structured description of interface components and data instead of unrestricted HTML. The client renders approved components.
Output can include forms, cards, tables, selectors and confirmation actions.
Why is text not enough?
Comparison, data completion and approval are clearer in structured controls than in a long text response.
An interface can reduce repeated interpretation and data-entry errors.
How does A2UI work?
The agent sends a structured message with components, properties, bindings and actions. The host renders only components in an approved catalog.
Google Cloud supports A2UI v0.9 and v0.8 and applies Gemini Enterprise styling to standard components.
- Structured UI message
- Approved component catalog
- Client-side rendering
- Action event returned to the agent
The agent and the interface
The agent proposes logic and data while the client owns rendering, accessibility and security boundaries.
Every action must map to a defined intent and server-side validation.
A2UI versus A2A
A2A supports communication and delegation between agents; A2UI carries an interactive interface from an agent to the user experience.
The two protocols can work together.
A2A
Agent ↔ Agent
A2UI
Agent ↔ User Interface
Security for generated interfaces
Component allowlists, schema validation, escaping, action permissions and human approval must be enforced outside the model.
Sensitive data must not appear simply because a component requests it.
- Component allowlist
- Schema validation
- Permission checks
- Human confirmation
- Audit trail
Enterprise use
A2UI fits request forms, document selection, option comparison, tool results and approval steps.
Dynamic UI should be used when it makes a user decision or action clearer.
A2UI at Aivan
At Aivan, generated interfaces must follow a controlled catalog, user identity and action policy. Execution remains within product and organizational boundaries.
Frequently asked questions
Is A2UI the same as generating HTML?
No. A safer pattern uses structured messages and pre-approved components.
How does A2UI differ from A2A?
A2A connects agents; A2UI connects an agent to the user interface.
Can an agent execute any UI action?
No. Permission, validation and required approval remain external controls.
Sources
- Google CloudGemini Enterprise Release Notes — A2UI and A2A GA17 August 2026 · Release Notes
- Google CloudRegister and manage agents using A2UI and A2A19 August 2026 · Official Documentation