AgentOps

AgentOps is the discipline of running multi-step, tool-using agents in production. It covers everything LLMOps covers plus the agent-specific concerns: tool call success rates, agent loop iteration counts, escalation rates, multi-agent handoff metrics, plan-vs-actual divergence, and failure-mode classification. The work is operational rather than experimental: tracking which user segments hit which failure modes, which tools fail most often, which prompts produce the longest agent loops, and where humans actually have to intervene. AgentOps also names a specific commercial platform (agentops.ai) but the broader term refers to the practice.

Adopt AgentOps practices as soon as you have one autonomous agent in production. Without it, you cannot tell whether agent behavior is improving or degrading over time.

• AgentOps platform
• Langfuse agent observability

Related Terms