FollowAgents AI AGENTS DIRECTORY
Home / AI Agent Glossary / Chain-of-Thought (CoT)
Skills & Workflow

Chain-of-Thought (CoT)

Also called: CoT

Chain-of-thought is a prompting technique where a model is asked to write out its intermediate reasoning steps before giving a final answer, rather than producing the answer directly.

QuestionStep 1Step 2Answer
The model writes reasoning steps before its final answer.

Ask a model a multi-step question and tell it to “just answer,” and it may skip straight to a guess. Chain-of-thought (CoT) prompting instead asks it to show its intermediate reasoning — the steps it took to get to the answer — before stating the final result. This was documented as a prompting technique in language-model research and has since become a standard, widely-used approach for tasks that involve arithmetic, logic, or multi-step reasoning.

CoT is purely about the shape of the model's output — reasoning text followed by an answer. It doesn't call any tools or take any actions on its own; it's a prompting pattern, not an execution framework. Writing out intermediate steps tends to make it easier to catch where reasoning went wrong, since the steps are visible rather than hidden inside a single jump to a conclusion.

CoT reasoning shows up in agent systems as a component: an agent might reason step by step (CoT) about what to do, and then act on that reasoning — which is closer to what react-prompting describes.

How it works

In practice, a CoT prompt either explicitly asks the model to “think step by step,” or is built from a small number of example question-reasoning-answer triples so the model picks up the pattern. The model then produces its reasoning as visible text before its final answer, rather than jumping directly to a one-line answer.

Example

Question: “If a train travels 60 mph for 2.5 hours, how far does it go?” A chain-of-thought response would show: “Speed is 60 mph, time is 2.5 hours, distance = speed × time = 60 × 2.5 = 150 miles” before stating the answer, rather than just outputting “150 miles.”

How it differs

Chain-of-thought vs. ReAct: CoT is reasoning-only text, with no tool calls or actions involved; react-prompting interleaves that kind of reasoning with actual actions and observations from tools, which is what turns reasoning into something an agent can act on.

Common misconceptions

Often assumed: A longer chain of thought is always more accurate.
Actually: Written-out reasoning helps in many cases, but a model can still reason its way to an incorrect conclusion — more steps don't guarantee correctness.
Often assumed: Chain-of-thought is itself an agent planning system.
Actually: It's a prompting pattern for how a model's output is structured, not a planning or execution framework on its own — planning and task-decomposition are the broader activities it can support.

FAQ

What is chain-of-thought prompting?
Asking a model to write out its intermediate reasoning steps before giving a final answer, instead of answering directly.
Does chain-of-thought make an AI more accurate?
Often, especially for multi-step or arithmetic problems, but it's not a guarantee — a model can still reason incorrectly.
What's the difference between CoT and ReAct?
CoT is reasoning text only; ReAct interleaves that reasoning with actual tool calls and observations.

Last checked: 2026-08-28

Related terms

ReAct

ReAct is a prompting technique that has a language model explicitly interleave written reasoning ('Thought') with actions and their results ('Observation'), rather than jumping straight from a question to an action.

Planning

Planning is the step where an agent turns a goal into an ordered set of smaller steps before — or while — it starts acting, so it isn't deciding what to do purely one action at a time with no larger structure.

Task Decomposition

Task decomposition is the process of breaking one large or ambiguous task into smaller, more concrete subtasks that can each be executed and checked individually.