loading
loading
A prompting pattern where the model alternates between reasoning out loud and taking actions, using each observation to guide the next thought.
ReAct (Reason + Act) interleaves chain-of-thought reasoning with tool calls: the model writes a thought, picks an action, reads the observation, then reasons again. The reasoning step keeps it from blindly firing tools, and the observation step grounds it in real results instead of hallucinated facts. It's the conceptual ancestor of most modern agent loops — when an agent 'narrates' its plan before searching or running code, that's ReAct.
Plainly
Think of ReAct as a simple recipe for doing the work better. A prompting pattern where the model alternates between reasoning out loud and taking actions, using each observation to guide the next thought.
In practice
Use it when you need a repeatable method instead of guessing from vibes. In practice, define the owner, input, output, and failure mode before you rely on it.