Agent Loop
One tool result can lead to another decision. The loop connects those turns.
An agent loop repeats a cycle of model decisions, actions, and feedback until a stopping condition is met. It lets the system adapt its next step to the result of an earlier action.
[Anthropic]In practice
An assistant searches for a file, reads the result, and then decides whether another search is necessary.
[Anthropic]A little deeper
Completion, an error or a configured limit can end the loop. [Anthropic]
Helpful to know: Language Model · Tool Calling