ReAct Connects Language-Model Reasoning to Actions

The ReAct method prompted a language model to alternate between reasoning about a task, taking an action and incorporating the resulting observation. In question answering, actions could query an external knowledge source; in interactive environments, they could change the model's state. This did not create a dependable autonomous worker, but it supplied a clear pattern for later agents: think, use a tool, inspect the result and decide what to do next.