Curiosity · Concept

ReAct (Reason + Act)

ReAct, proposed by Yao et al. (Princeton/Google) in 2022, combines chain-of-thought reasoning with the ability to call tools. The model emits a 'Thought' about what to do, an 'Action' that invokes a tool (search, calculator, API), receives an 'Observation' from the tool's output, then thinks again. This loop underlies almost every modern LLM agent framework.

Quick reference

Proficiency
Intermediate
Also known as
ReAct, Reason-Act, Thought-Action-Observation loop
Prerequisites
Chain-of-thought, Function calling / tool use

Frequently asked questions

What is the ReAct pattern?

An agent control pattern where an LLM alternates between explicit reasoning steps ('Thought: I need the current price of X') and tool invocations ('Action: search(X price)'), using the tool's observation to inform the next thought. The loop terminates when the model issues a 'Final Answer' action.

How is ReAct different from pure chain-of-thought?

CoT reasons inside the model's head with no external help — limited by parametric knowledge and prone to hallucination. ReAct lets the same reasoning process call out to search, calculators, databases, or APIs, grounding each step in real data. It is CoT plus actions plus observations.

What are common failure modes of ReAct agents?

Tool-use hallucination (inventing tool names or arguments), infinite loops, over-thinking simple tasks, poor error recovery when a tool returns garbage, and context-window bloat as Thought/Action/Observation traces accumulate over many steps.

Is ReAct still used in 2026?

Yes, although simpler forms dominate in production. Modern tool-use APIs (Anthropic tools, OpenAI function calling, Gemini function calling) implement the pattern natively — you send a message, the model may return a tool_use block, you execute and return the result, it continues. Frameworks like LangGraph, CrewAI, and Claude's agent SDK build on this loop.

Sources

  1. Yao et al. — ReAct: Synergizing Reasoning and Acting in LLMs — accessed 2026-04-20
  2. Anthropic — Tool use with Claude — accessed 2026-04-20
  3. LangChain — Agents concepts — accessed 2026-04-20

Related

  • Chain-of-Thought Prompting — Chain-of-thought (CoT) prompting is a technique where the model is asked to show its step-by-step reasoning before giving a final answer, which dramatically improves accuracy on math, logic, and multi-step tasks.
  • Retrieval-Augmented Generation (RAG) — Retrieval-Augmented Generation (RAG) is a pattern where an LLM is grounded on retrieved passages at query time — fewer hallucinations, up-to-date answers, no retraining required.
  • Structured Output — Structured output is the capability of having an LLM return JSON, a typed schema, or a tool call that conforms exactly to a declared structure — the bridge between free-form language models and deterministic code.
  • All Concepts — Browse the complete concepts library
  • VSET — Programmes offered — B.Tech programs (CSE, AI&ML, AI&DS, Electronics, IIoT)