Trajectory
The full recorded path of an agent through a task — every thought, action, and observation, in order.
Think of it like
A hiker’s GPS track: not just where they ended up, but the whole winding route they took to get there.
Example
Debugging a failed run, you replay its trajectory and see it looped between two files three times before giving up.
How it actually works
A trajectory is the complete step-by-step record of one run, and it is the primary artifact for understanding agents — where it went wrong, where it wasted turns, whether the reasoning made sense. Trajectories are also training and evaluation data: good ones teach the model, and comparing them across runs reveals patterns. Reading trajectories is the closest thing agent builders have to debugging by hand.
For product teams
The receipt you read when an agent fails — how you actually learn why.
For engineers
The ordered sequence of states, actions, and observations produced over one agent run.
Related
- Rollout — A single sampled trajectory, often for training.
- Agent Loop — The loop that generates each step of it.
- Agent State — The evolving state it records.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome