AI AgentAgent PatternsUpdated 2026.04.28

ReAct

Reasoning + Acting

Also known asReAct 패턴Reasoning + Acting

In one line

ReAct (Reasoning + Acting) is the classic agent pattern where an LLM loops through Thought, Action and Observation steps — reasoning out loud and calling tools as it goes.

Going deeper

ReAct came out of a 2022 paper by Yao et al. (Princeton + Google Brain, arXiv 2210.03629). Instead of producing one big answer, the LLM cycles through Thought, Action and Observation — reasoning, calling a tool, looking at the result, and feeding it back into the next thought.

It matters because almost every modern agent framework — LangChain, OpenAI Assistants, Claude's tool-use loop — is a variation or extension of ReAct.

The limits are real. Long loops accumulate hallucinations and bad tool calls, and tokens add up fast. In production teams usually wrap ReAct with planner and critic components rather than running it raw.

Sources

Related terms

How does your brand show up in AI answers?

Villion measures how your brand appears across ChatGPT, Perplexity and AI Overviews, then automates the work that lifts citation rate and share of voice.

Get a free audit