THE CORE AGENT LOOP

CHECK YOUR UNDERSTANDING • What three moves make up the agent loop? ◦ Answer: think, act, and observe. • Why is writing reasoning down useful in ReAct? ◦ Answer: it stays in the history to keep the model on track, and humans can read it to understand and debug the run. • Give one task better suited to plan-and-execute and one better suited to a reactive style. ◦ Answer: a known multi-step report with fixed stages suits plan-and-execute, while diagnosing an unknown problem suits a reactive style. • Why should a harness check a model's claim of completion? ◦ Answer: because a model may say it is done when the work is incomplete, so the harness verifies the result. • What are the risks of a step limit that is too low and too high? ◦ Answer: too low cuts off healthy runs before they finish, and too high lets a stuck agent waste time and money.