Common questions.
What is Agent Trajectory?
An agent trajectory is the full sequence of steps an AI agent takes to complete a task: each reasoning turn, every tool call with its arguments and result, each retrieval and what it returned, and the intermediate outputs that led to the final answer.
How is Agent Trajectory used in production?
It matters because an agent can reach a correct-looking answer through a broken process, calling the wrong tool, looping unnecessarily, or grounding a claim in a document it never retrieved, and output-only scoring passes all of that. Trajectory-level evaluation scores the path as well as the result: tool-call accuracy, step efficiency, error recovery, and goal adherence, usually with an agent-as-a-judge that inspects every step. Capturing the trajectory is the job of agent tracing; scoring it is where agent evals and benchmarks such as tau-bench measure reliability rather than a single lucky run.
Which terms are related to Agent Trajectory?
Closely related concepts include Evaluation Harness (Agent), Ground Truth Evaluation, Hallucination Rate, Continuous Integration Agent. Each is defined in the Prefactor glossary.
Where can I read more about Agent Trajectory in practice?
The guide "Agent evals: scoring the trajectory, not just the answer" covers the applied side in depth.