# Streaming AI & Generative Interfaces

**Track:** Models, Tools & Agents — AI-Native Full-Stack Development — complete (30)
**Framework / surface:** web dev (streaming UI)
**Level:** Intermediate
**Prerequisites:** Models as Typed Capabilities, HTTP, JSON, APIs & Streaming, Components, State & Events

**In one line:** Design interfaces around partial results, tool events, and evolving state rather than one final block of text.

## Theory, aesthetics & inspiration

Human-factors research has held steady since Robert Miller's 1968 paper on response times, and Jakob Nielsen's 1993 formulation made it canon: at a tenth of a second an interface feels instantaneous, at one second the flow of thought survives, and by ten seconds attention is gone. Model latency lives in exactly the danger zone — seconds of generation — but unlike a database query, a model produces useful partial output the whole way, and streaming turns the wait into an interaction surface. The important idea is progressive state, not a chat window: text appearing incrementally, structured data arriving in parts, tool and agent events updating a board, a timeline, a composition before the task completes. The interface should distinguish provisional output from committed state and keep cancellation and correction reachable mid-stream. The browser primitive underneath is the WHATWG Streams API, with typed model and UI streams layered above it; learn both levels — what actually crosses the network, and how those events become rendered state.

**Builder question:** What useful state could your interface reveal before the model or agent reaches its final answer?