What is OpenAI Intelligent UI?

OpenAI announced Intelligent UI in ChatGPT on October 7, 2026. An answer can include an interface the reader can operate. This is a product capability; it should not be treated as the same kind of artifact as a protocol such as AUP or A2UI.

From reading to acting

Imagine planning dinner: after reading the answer, you adjust the guest count and inspect the preparation list. That illustrates the use of an interactive answer. It does not tell us how state carries across systems.

What has been documented?

OpenAI describes native streamable components, a compiler that processes the interface incrementally, and model training for layout and interaction. Its help page says some component state survives a chat refresh but does not carry across chat threads. Do not infer identical persistence behavior for every component.

How should it be compared with AFS-UI?

Intelligent UI’s public emphasis is composing useful answers. AFS-UI also asks how interfaces connect to AFS resources and operations. Both face presentation and state questions. This is not a performance or security ranking.

Examine one task: after rebuilding a view, which object is still the same object? Through which interface can another operator access it? Who checks permission? A UI node and a business object can have different lifetimes; AFS paths also need explicit contracts for these boundaries.

The sources checked on 2026-10-08 do not provide a complete general protocol specification, component schema or cross-host state contract. That is a boundary of the public documentation, not a claim that an internal capability is absent. Appearance alone does not establish HTML, React or MCP Apps as the implementation.

Continue: Generative UI comparison · UI history · AUP

Official references