Reliability and fallbacks
Also called: graceful degradation, failover, timeouts and retries
Designing for the times the model is slow, unavailable, or returns something unusable — with timeouts, retries, and a defined safe behaviour when the AI step can't complete.
External model APIs have outages and bad days. A feature with no fallback fails hard and visibly; one that degrades gracefully — a cached answer, a simpler path, an honest 'try again' — keeps the product usable when the AI part doesn't cooperate.
Accuracy. Reliability is about the system still behaving sensibly when the model fails or stalls, not about whether the model's answer was correct when it did respond.
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1U.S. National Institute of Standards and Technology · 2023-01-26
First-pass citations, limited to primary sources; a reviewer will broaden and verify these before this entry leaves draft.