ReAct Pattern

ReAct is an agent pattern that interleaves reasoning steps with actions — the model alternates between thinking about what to do and calling a tool — so it can gather information and adjust as it works.

By combining reasoning with tool use in a loop, a ReAct agent can plan a step, act on it, observe the result, and revise, rather than committing to a fixed plan up front.

It is a foundational loop for tool-using agents, and its reliability across steps is a major driver of real-world agent quality.