Agent trajectory
By Gareth HoyleLast updated How we research
An agent trajectory is the step-by-step record of an AI agent or person completing a task, including each observation, action, tool call and result.
An agent trajectory is the full sequence of steps taken to complete a task: what was observed at each step, the action chosen (a click, a keystroke, a tool or API call), its result and the final outcome. Trajectories can come from AI agents or from people doing the same work.
Why do AI labs buy agent trajectories?
AI labs buy agent trajectories because they show models how multi-step work is actually done, which finished documents alone do not capture. They are used to fine-tune agents on successful paths and to train or score agents with reinforcement learning.
What makes an agent trajectory usable?
- Every step logged in order with timestamps
- The task goal stated at the start
- Tool inputs and outputs captured in full
- A clear success or failure label
- Personal data and credentials removed
Trajectories are most useful when paired with a replayable RL environment. See agent trajectory data and agent trajectories and RL environments explained.