Definition
AgentOps extends MLOps and LLMOps to systems that take actions. Beyond tracking model quality, it covers what agents actually did: which tools they called, with what inputs, what it cost, where they failed and whether a person had to step in.
Typical AgentOps capabilities include tracing of every step, evaluation suites that are rerun on each change, cost and latency budgets, permission controls for tools, and incident handling when an agent behaves unexpectedly. The aim is the same as in any operations discipline: know what is running, catch problems early and improve safely.