agent execution

A single run of an LLM agent — from initial prompt through tool calls until the agent emits a final answer or stops.

Definition

An agent execution is one end-to-end run of an LLM agent: the model plans, calls tools, observes results, and loops until it answers or hits a stop condition (token budget, time limit, max iterations, repeated tool-call loop). Execution traces are essential for debugging; LangSmith, Helicone, and the Anthropic console all visualise them.

When to use

See also