Generative UI
Generative UI, or GenUI, is an interface pattern in which an AI system selects, composes, or fills interactive interface elements in response to a user's goal and current context. Instead of returning only prose, the system can produce cards, forms, charts, or task-specific controls. GenUI is broader than generating front-end source code: a production design may map model output to trusted components rather than execute arbitrary generated code.
Origin and context
Vercel documented generative UI in March 2024 through AI SDK 3.0's component-streaming interface. Google used the same term for query-specific layouts and interactive tools in Search in November 2025, then introduced the separate A2UI interface-description project in December. Google's June 2026 Search article documented interactive visual generation already available to some subscribers. Together these sources show SDK-level and end-user implementations of the same general pattern, without establishing a single inventor or common wire format.
Why it matters
The output format can be part of answering a question. A weather card exposes structured values, while an interactive diagram lets someone explore a relationship that is difficult to explain in a chat paragraph. Generative UI moves some presentation decisions from a fixed screen into the response-generation process. The product-design question is which choices belong to the model and which remain fixed in the application. A component catalog can constrain that choice, but not every generative interface uses one.
Example
In an illustrative travel-planning interface, a user asks to compare several routes. The assistant chooses a comparison card and fills it with tool-returned durations, then offers filters appropriate to that question. The application controls the actual widgets and their behavior. In another implementation, the model generates an interactive diagram rather than choosing a predefined card. Both approaches adapt the interface to the task. A fixed link shown in every answer does not demonstrate the same response-specific composition.
How it differs
A2UI (Agent-to-User Interface)
Generative UI is the broad interaction pattern. A2UI is a specific declarative format for transmitting agent-generated interface descriptions to a client that renders trusted components. GenUI can be implemented with framework-specific component streaming, structured tool results, templates, or A2UI. Therefore, an A2UI response can power GenUI, but the two terms are not synonyms and GenUI does not require that protocol.
Maturity and evidence
Maturity is rated 3. The general interaction pattern has repeated, concrete use across Vercel's developer tooling and Google's agent-interface and Search products, extending beyond the launch of a single SDK. Its lifecycle is established at the pattern level; this does not make A2UI a final standard or imply cross-framework compatibility. The evidence supports an explanatory category, while accessibility, consistency and comparative usefulness remain implementation-specific questions.
Limits and open questions
Interface generation can mean component selection, declarative composition or code generation, and those mechanisms should not be treated as equivalent. Google's A2UI design specifically limits requests to client-approved components; it does not establish that arbitrary model-generated code has the same boundary. The product announcements also do not demonstrate universal improvements in usability or accessibility. Generated controls still need application-defined behavior, and an attractive visualization is not evidence that the underlying answer is correct.
Related terms
References
- Introducing AI SDK 3.0 with Generative UI supportVercel · 2024-03-01 · class A
- Introducing A2UI: An open project for agent-driven interfacesGoogle Developers Blog · 2025-12-15 · class A
- Google Search with Gemini 3: Our most intelligent search yetGoogle · 2025-11-18 · class A
- 4 ways to keep up with soccer using Google toolsGoogle · 2026-06-08 · class A
Last updated: 2026-09-05
This term is also covered in the Skills Atlas as llm function calling skill.
This term is also covered in the Skills Atlas as ai ux design skill.