Skip to content
BENIBIX Learn

Agent Trajectory

AI🤖 AI AgentsAgent Control

Also known as Agent Trace, Agent Transcript, Agent Run Log

The record of each step an AI agent took in one run.

Think of it like

It is like the trail of footprints in snow. You can see every step someone took to get where they are.

Example

Emma's agent sent the wrong email. She reads its trajectory and sees that at step 6 a search returned an old address, which the agent then used.

Why it matters

You cannot fix what you cannot see. The trajectory shows the exact step where a run went off course.

Definition

An agent trajectory is the full record of one agent run, step by step. It lists what the agent thought, which tools it called, what came back and what it did next. People read it to find where a run went wrong. Teams also score whole trajectories to test an agent, since a good final answer can hide bad steps. A trajectory is often saved as a trace in an observability tool.
The log under the row of steps is the trajectory. It keeps every call and result in order.

Common mistake

Checking only the final answer. An agent can reach a right answer by a risky path that fails next time.

Key takeaways

  1. 1
    Every step. Thoughts, tool calls and results, in order.
  2. 2
    Find the fault. Shows where a run went wrong.
  3. 3
    Test the path. Score the steps, not just the answer.