ReAct (Reason + Act)

What ReAct Is

ReAct (Reason + Act) is an agent pattern where a model interleaves an explicit reasoning step with a tool call, then reads the result before deciding what to do next. The name comes from the 2023 paper "ReAct: Synergizing Reasoning and Acting in Language Models", from researchers at Princeton and Google Research.

Not to be confused with React, the JavaScript UI library, same name, unrelated concept.

The Loop

ReAct repeats three parts as many times as needed:

• Thought: the model reasons about the current state and what it needs to do next
• Action: the model calls a tool to gather information or take a step
• Observation: the result of that tool call gets added to what the model can see

How It Works Mechanically

Each tool call is the model pausing generation with a tool_use stop reason: it names a tool and its arguments, the calling code executes it, and the result is sent back as a tool_result so the loop can continue. The model never runs a tool itself: it requests, your system executes.

Why It Matters

ReAct is the default single-agent pattern behind most production agents, including LangGraph's create_react_agent and Claude Code's own loop. Newer patterns like plan-and-execute and reflection build on top of ReAct rather than replacing it, usually by adding an upfront plan or a self-critique step around the same core reason-act-observe cycle.