Definition
Traditional monitoring asks whether a service is up. AI observability also records what the model was given, what it returned, which tools an agent called, how long each step took, how many tokens it used and how users or evaluators rated the result.
Standards are emerging: OpenTelemetry defines semantic conventions for generative AI so traces from different tools and providers can be collected in one place. Observability is the raw material for evals, cost control, incident response and audits.