Building AI-Powered UX
9 min readUpdated June 22, 2026
AI-powered interfaces fail in different ways from traditional software, so the UX needs to show different information.
In a standard web app, when a user clicks a button, the system usually does something predictable. It saves a record, fetches a list, or runs a calculation. The operation often finishes in under a second, and the result is either correct or there is a bug. Users understand this model because they have seen it for years.
AI changes several of these assumptions:
- Latency is variable. A simple query might take 2 seconds. A complex one might take 15. The user often cannot tell in advance.
- Results can vary. The same question can produce different wording, and sometimes a different answer, when context, tools, or model settings change. Users need ways to inspect, regenerate, and steer the result.
- Correctness is not guaranteed. The answer might be right, partially right, or plausible but wrong. Users need ways to verify important claims.
- The system has stages. A RAG pipeline searches documents, filters results, may rerank them, and then generates an answer. An agent may call tools before it responds. Each stage can fail or degrade independently.
These differences call for UX that shows progress, uncertainty, sources, and control. A spinner alone hides too much.
The first pattern is streaming.