Observability
Working knowledgeGovernance, Safety and Ethics
Also called: AI Observability
Observability is the ability to see, trace and explain what an AI system actually did. Every model call, tool use and reasoning step is visible in enough detail to debug a failure and prove compliance. Traditional monitoring was built for predictable software: it can tell you a system is slow, but not why an agent reached a particular decision.
In practice
Without observability you cannot answer the two questions that follow any AI incident: what did it do, and why. Put it in before agents touch a live process, because a decision cannot be reconstructed after the fact from logs that were never designed to capture reasoning.