Technique: ReAct (Reasoning + Acting)
ReAct is a foundational pattern for AI agents. The model alternates between thinking (Thought), taking an action (Action), and observing the result (Observation) in a loop until it reaches a final answer. This is how modern AI agents work — they reason about what to do, execute a step, process the result, and decide what to do next.
In a real agent system, the “Actions” would call actual tools (web search, calculator, database query, API call). In this playground, the model simulates the entire loop. The pattern demonstrates the core agent reasoning cycle that powers tools like ChatGPT’s browsing mode, coding assistants, and autonomous research agents.
When to use: Complex questions requiring research, multi-step problem solving, tasks that need external information, and any scenario where the AI needs to “think, act, and adapt” rather than answer in one shot. This is the entry point to understanding how AI agents are built.