technique

ReAct pattern

ReAct interleaves Reasoning + Acting in an agent loop — the model writes a thought, then decides to call a tool, then observes the result, then thinks again.

ReAct (Yao et al., 2022) was the original recipe for LLM agents. Each step has three parts: Thought (free-form reasoning), Action (tool call), Observation (tool result fed back in). The loop continues until the model emits a Final Answer. Modern agent frameworks (LangGraph, OpenAI Agents SDK, Claude Agent SDK) all implement variations of ReAct, often with a planner that pre-decomposes the task before the executor loop begins. Pure ReAct can drift on long horizons — add budget caps and a planner-executor split for reliability.

When to use react pattern

Common mistakes

FAQ

What is react pattern?

ReAct interleaves Reasoning + Acting in an agent loop — the model writes a thought, then decides to call a tool, then observes the result, then thinks again.

When should I use react pattern?

Agents with tools where each step depends on the previous result. Web research, code editing, browser automation.

What are the most common mistakes with react pattern?

Letting the loop run unbounded — set a max step count. Pure ReAct on tasks with branching planning needs — use a planner first.

Sources

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/react-pattern.md.