Evaluating Agent Trajectories

Know the Vocabulary

This lesson uses the terms below many times. Read them once now, and come back here when a later section needs one.

• Agent: A system in which the model chooses some of the next steps at runtime, such as which tool to call. The application still decides which actions are allowed.
• Tool: A function the agent may ask the system to run, such as looking up an order or issuing a refund.
• Step: One turn of the agent's loop. The model reasons, asks for one action, and reads the result.
• Trajectory: The full path one task takes