Agents

ReAct: Synergizing Reasoning and Acting in Language Models

Princeton·October 6, 2022

Shunyu Yao, Jeffrey Zhao, Dian Yu

TL;DR

Interleaves reasoning traces with actions (like tool or search calls), letting a model plan, act, and adjust within a single loop.

A foundational pattern for tool-using agents — the basis of much of today’s agentic tooling.