Traditional software taught users that anything over 400ms is slow. AI products routinely take five to thirty seconds. That gap is a design problem, and spinners do not close it.
What works instead is staged, specific progress. A status line that reads “searching your documents, reading 12 matches, drafting summary” tells the user the right thing is happening, not just that something is happening. It also builds a mental model of what the system actually does, which pays off every time the output needs judging.
Three rules I apply on AI projects:
Acknowledge the request instantly, before any real work starts.
Name the stages in the user’s language, not the pipeline’s.
Stream the answer as soon as any of it exists.
Wait time you explain is anticipation. Wait time you hide is doubt.
Available
Think there’s potential for us to work together?
Tell me what you’re building and where design is slowing you down. If there’s a fit, I’ll show you exactly how I can help.
Think there’s potential for us to work together? Tell me what you’re building and where design is slowing you down. If there’s a fit, I’ll show you exactly how I can help.
Brendan Ho