Retry and Error Recovery UX
Explore how to improve user experience when AI-generated responses fail or are incomplete. Learn to classify error types, implement retry and regenerate actions with clear state management, and provide fallback options that keep users in control of their input without confusing duplicates or dead ends.
Suggestions can be safely dropped when they arrive late. A failed assistant response cannot simply be dropped, the user still needs a way forward. Users report a familiar symptom pattern in production. The assistant stops mid-sentence, an error banner appears that does not say what to do, and pressing Retry appends a second assistant message that starts over without acknowledging the partial text.
That UX failure is rarely just one bug. It usually means the UI has no explicit taxonomy for what went wrong, and it has no rules for what state must be preserved across retries, so the interface cannot offer a safe recovery path.
Classify failures by what the user can do
With the symptom anchored, the next step is to classify failures at the UI layer based on observable behavior, not internal codes. The goal is to map each failure to an action the user can take and a promise about what the product preserves.
The same red banner can hide very different realities, so the UI should separate outcomes that imply different next steps.
Network error ...