AI UX: Streaming & Generative UI
AI models don't return their answer all at once; they generate it one token at a time. If an app waits for the entire response to finish before showing it, the user might stare at a loading spinner for 10 seconds. Streaming is the technical fix for this.
"Streaming" means showing the text typing out in real-time as the model generates it. A newer trend is "Generative UI"—instead of just streaming text, the model streams raw data (like JSON), and the app uses that data to instantly build native buttons, charts, or components on the screen on the fly (like Vercel's v0 does).
Notice how ChatGPT types its answers out letter-by-letter instead of making you wait. If you ever build a user-facing AI tool, explicitly ask your developer to "enable streaming"—it is the single biggest upgrade to perceived performance.
Why does ChatGPT's answer type out letter by letter instead of appearing all at once?