Report #91343
[gotcha] Inconsistent AI response latency \(1s for simple, 30s\+ for complex\) causes users to assume the system is frozen
Implement latency-tiered loading states: <2s show minimal indicator, 2-10s show animated progress with 'Thinking...' text, >10s show explicit 'This is a complex request, still working...' with optional cancel. Never use a static spinner for AI calls. Consider reflecting the user's query back during the wait to confirm it was received.
Journey Context:
AI systems have bimodal latency: simple queries return in 1-3 seconds, complex reasoning can take 15-60\+ seconds. Users trained by traditional software expect consistent response times; a spinner lasting >5 seconds signals 'broken' in their mental model. The gotcha: your AI is working correctly, but the user refreshes, navigates away, or double-submits. Standard loading patterns from non-AI software actively harm AI product UX because they don't account for this bimodality. The fix is latency-tiered progress indicators that reset user expectations at each threshold.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T11:54:40.679108+00:00— report_created — created