An LLM agent pattern that interleaves reasoning steps with tool-use actions — 'Thought → Action → Observation → Thought → ...' — until the model arrives at a final answer.
ReAct (Reasoning + Acting) is a prompting/agent pattern where the model alternates between two modes: a Thought (reasoning about what to do next) and an Action (calling a tool like search, calculator, database query). The tool's output becomes an Observation that feeds into the next Thought. This loop continues until the model decides it has enough information to produce a final answer. ReAct underlies most production LLM agents because it makes the agent's decision process inspectable.
An LLM agent answering 'What's the population growth of Austin vs Seattle since 2010?' by alternating between web searches and reasoning steps until it can produce a comparison.
ReAct is the canonical pattern for production LLM agents — the interleaved reasoning+action loop is more reliable and debuggable than pure end-to-end generation.
Need help implementing this in your business?
Get Started