AI Glossary

Agent loop

What does “Agent loop” mean?

Agent loop: At the core of most agents is a repeating cycle: the model reads the current situation, chooses an action, the system runs that action, and the result is fed back to the model. The loop ends when the model decides the goal is met or a limit is reached.

Agent loop visual guide: At the core of most agents is a repeating cycle: the model reads the current situation, chooses an action, the system runs that action, and the result is fed back to the model. The loop ends when the model decides the goal is met or a limit is reached.
Visual guide Open full size (new tab)

What is an example of “Agent loop”?

Step limits and time limits on the loop stop a confused agent from repeating the same failing action forever.

Keep learning

See also

AI agentTermReActTermTool useTermContext windowTerm

Same topic

Agent memoryTermAgent guardrailsTermAgent permissionsTermAgent evalsTermAgent skillsTerm