What Is ReAct (Reasoning and Acting)?

ReAct, short for reasoning and acting, is a method that lets an AI agent alternate between thinking through a problem and taking actions with tools. The model reasons about what to do next, acts by calling a tool, observes the result, then reasons again. This loop lets it work through multi-step tasks instead of trying to answer everything in one shot.

How ReAct works

ReAct runs as a cycle of three moves: think, act, observe. The model writes out its reasoning about the current state and what to try next. It then acts, usually by calling a tool such as a search, a calculator, or an API. It reads the result of that action, folds it into its reasoning, and decides on the next step. The cycle repeats until the task is done.

The key idea is interleaving. Earlier prompting methods asked a model to reason in one block or to call a tool in isolation. ReAct combines them, so each action is informed by fresh reasoning and each round of reasoning is grounded in real results from the world, rather than in the model's assumptions.

Why ReAct matters for AI

ReAct is one of the foundations of practical AI agents. Letting a model check its work against real tool results, step by step, makes it far more reliable on tasks that need current information or several actions in sequence. It also makes the agent's behavior easier to follow, since its reasoning is written out at each step.

For teams building agents, ReAct is a pattern worth knowing because it explains how a model goes from producing text to getting things done. It is a common building block inside the agentic workflows that power real systems.