What is a ReAct Pattern (Reason + Act)?

Introduced in AI agent research, ReAct interleaves chain-of-thought reasoning with tool use in a loop: the model thinks about what it needs, takes an action (a search, an API call, a calculation), observes the result, and reasons again before deciding the next step. This is different from a model that reasons purely internally and then produces one final answer, or one that calls tools blindly without narrating why. The pattern matters because it makes an agent's behavior both more accurate and more inspectable. Grounding each action in an explicit thought reduces hallucination — the model is reasoning from real observations rather than guessing — and the visible thought trail gives developers and users a way to audit why the agent did what it did, which is valuable for debugging multi-step failures. ReAct underlies most modern tool-using agent frameworks, including the loop that drives agentic coding assistants and autonomous task agents: plan, act, observe, replan, repeat until the task is done or a stopping condition is hit.