ReAct (Reasoning + Acting) is the most widely used agent architecture — the LLM interleaves reasoning steps with action steps in a structured loop.

ReAct Pattern

Thought: I need to find the current CEO of Apple.
Action: search("Apple CEO 2024")
Observation: Tim Cook is the CEO of Apple.
Thought: I have the answer.
Final Answer: Tim Cook is the current CEO of Apple.

Why ReAct Works

• Reasoning step catches errors before they happen
• Explicit thought process makes debugging possible
• Observation step grounds next reasoning in actual results
• Natural fit for tool-using LLMs

Implementation

Available out of the box in LangGraph: create_react_agent(llm, tools)

Reference:

ReAct paper

https://arxiv.org/abs/2210.03629