AGENT OPERATIONS (AGENTOPS)

CHAPTER SUMMARY AgentOps is the discipline of running autonomous agents reliably in production, building on DevOps and MLOps and adding agent-specific observability, governance, cost control, continuous evaluation, and human oversight. Runtime governance enforces rules from outside the agent. Monitoring with baselines catches drift that no single run reveals. Written escalation paths make sure humans are alerted reliably. KEY TERMS REVIEW AgentOps: operating agents reliably in production. DevOps: joining development and operations for software. MLOps: operating machine learning models in production. Governance: rules and oversight over behavior. Policy engine: software that checks actions against rules. Drift: slow unintended change over time. Baseline: normal range for a measurement. Alert: automatic notice when a measure goes out of range. On-call: the person currently responsible for responding. Escalation path: written plan for who is notified and when.