Chain of Thought

A prompting technique where the model is encouraged to produce step-by-step reasoning before answering, improving accuracy on complex tasks.

Chain of Thought (CoT) prompting was introduced by Wei et al. in 2022 and is the foundation for most modern agent reasoning patterns. Instead of asking a model to produce an answer directly, you prompt it to “think step by step” — and the intermediate reasoning measurably improves accuracy on multi-step problems.

CoT is the conceptual ancestor of ReAct: ReAct extends CoT by interleaving actions (tool calls) between reasoning steps, turning the chain into a loop. In 2026, “thinking” or “reasoning” model modes (OpenAI o-series, Anthropic’s extended thinking, Google’s Gemini with thinking) implement CoT internally — the model generates a private reasoning trace before emitting the visible answer.

Practical implication: for agent applications, you usually don’t write “think step by step” in prompts anymore. The reasoning models do this implicitly and bill for the thinking tokens. Older non-reasoning models still benefit from explicit CoT prompting on complex tasks.