agentopsd

Definition

A rendering of "AgentOps"; the trailing d looks like a tokenisation artefact rather than part of the term. AgentOps covers the operational side of running LLM agents in production: tracing each run, tracking token cost, catching failures, and evaluating changes before rollout. It borrows its framing from DevOps and MLOps, adapted to systems whose behaviour is non-deterministic. It is industry vocabulary, not a settled academic term.

Explainers using this term

• Paper Deep-Dive: AgentOPSD — Finding the Turn That Won the Game with Recursive Bayesian Belief UpdatesAgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning