Skip to main content

COMPEL Glossary / agent-observability

Agent observability

The logging, tracing, and evaluation infrastructure that makes an agent's plans, tool calls, memory reads/writes, and decisions auditable after the fact.

What this means in practice

Governance artefact — not an engineering nicety — because without observability an organization cannot evidence compliance with NIST AI RMF MEASURE/MANAGE or reconstruct an agentic incident for regulators.

Synonyms

agent tracing , agent telemetry , agentic audit trail

See also

  • Agent memory — The stored context, preferences, observations, and state an agent accumulates across turns or sessions.
  • Evaluation harness — The infrastructure that runs capability, regression, safety, and human-review evaluations on an LLM feature on a defined cadence.

Related articles in the Body of Knowledge