A2UI Architecture Lets AI Agents Describe UIs Without Writing Any Code

A2UI is a design architecture that allows language models to build user interfaces on the fly by sending structured data descriptions rather than executable code. Instead of generating JSX or HTML, the agent produces a small JSON object that names a pre-existing client-side component and specifies which values it should display. This approach keeps all executable code on the client, making the model's output validatable against a schema before anything is rendered. The architecture splits responsibilities across three distinct roles: an agent that decides what to show, a protocol that defines the wire format, and a renderer that handles visual presentation. Because components are reused and only values change, descriptions can be cached and the format can be versioned and treated as a formal contract across multiple client implementations.