ReAct (Reasoning + Acting), introduced by Yao et al. in 2022, is one of the most influential design patterns in the AI Agent world. The idea is straightforward: at each step the model first produces a 'thought', then takes an 'action' (a tool call), reads the resulting 'observation', and repeats the loop. This interleaving combines the strengths of pure Chain-of-Thought reasoning with pure Tool Use, and reliably outperforms either alone on multi-step retrieval and decision tasks. LangChain, LlamaIndex and most agent frameworks default to ReAct or one of its variants — it is the shared grammar of modern agentic architectures.
External Links