Regression Detection (Agent)

Regression detection for AI agents is the monitoring of quality and performance metrics over time to identify when a previously acceptable agent has begun producing worse outputs: due to model drift, data distribution shifts, prompt degradation, or changes in connected tools.

Automated regression detection prevents gradual quality decline from going unnoticed until it causes significant user impact or compliance failures.

Closely related concepts include Drift Detection (Agent), Human Preference Annotation, Model Drift, Prompt Regression Testing. Each is defined in the Prefactor glossary.