The agent loop is how an AI agent works: plan, act, look at the result, plan again, until the task is done.
Illustration: AI-generated · AI Compare Ninja · CC BY 4.0
Example
An agent searches for information, notices that the source is outdated, searches again and only then summarises.
For pros
Technically, the model calls a tool in each round, the result goes back into the context, and the model decides the next step. Good loops have a stop condition, a maximum number of steps and approval points. The ReAct paper (2022) describes the pattern “reasoning + acting”.
