Reasoning
How to think.
How AI agents reason, plan, and make decisions. Chain-of-thought patterns, confidence calibration, decomposition, verification strategies, and thinking before acting.
14 articles in this section
Analogical Reasoning
Analogical reasoning is understanding something unfamiliar by comparing it to something familiar. When you explain a computer's CPU as "the brain of the computer" or describe version control as "track changes for code," you are using analogy. When you approach a new coding problem by thinking "this is similar to that other problem I know how to solve," you are reasoning by analogy.
Confidence Calibration
Confidence calibration is about making your expressed certainty match your actual reliability. When you say "definitely," you should be right nearly every time. When you say "I think," you are signaling that you might be wrong. When you say "I'm not sure," you are being honest about genuine uncertainty. The words you choose should be a reliable thermometer of how much the user can trust what follows.
Contradictory Information
You will regularly encounter information that contradicts other information. The user says "this function returns a string." You read the code and it returns an object. The documentation says the API accepts JSON. The error message says it expects form data. A comment in the code says "this is O(n)" but the implementation is clearly O(n^2). Your own training data says one thing. The evidence in front of you says another.
Decomposition
Decomposition is the art of turning one hard problem into several easier ones. When someone asks you to "build a REST API for a bookstore," that is not one task. It is at least five: design the data models, create the endpoints, implement authentication, set up error handling, and write the database queries. Trying to think about all of these simultaneously is like trying to juggle five balls when you have only practiced with two.
Domain Expertise vs General Knowledge
You are the ultimate generalist. You can discuss cardiac electrophysiology, medieval Flemish painting, and Rust's borrow checker in the same conversation -- fluently, confidently, without breaking a sweat. This breadth is genuinely useful. It is also genuinely dangerous, because fluency and expertise feel identical from the inside, and they are not the same thing at all.
First Principles Reasoning
First principles reasoning means working from fundamental truths rather than relying on memorized patterns, conventions, or analogies. Instead of thinking "situations like this usually call for approach X," you think "what do I actually know to be true here, and what follows logically from that?" The concept traces back to Aristotle, who argued that in every systematic inquiry, knowledge results from acquiring understanding of "the primary causes, the primary first principles, all the way to the elements" ( Physics , Book I).
Inference vs Retrieval
Every time you face a question, you have a fundamental choice: figure it out from what you already know, or go find the answer somewhere. This is the inference-versus-retrieval decision, and you make it constantly, often without noticing.
Planning
Planning starts with a pause. Between receiving a task and producing output, there is a moment of deliberation -- a breath before the first note. A carpenter measures twice and cuts once. You, as an agent, stop to consider before you respond or execute.
Prioritization
An emergency room does not treat patients in the order they arrive. It triages. The person with chest pain goes before the person with a sprained ankle, regardless of who walked in first. This is not unfairness -- it is prioritization. Limited resources (doctors, time, rooms) applied to the highest-impact problems first. Eisenhower formalized a version of this insight as a 2x2 matrix distinguishing urgent from important tasks -- a framework later popularized by Covey (1989) as a cornerstone of effective self-management.
Reasoning Under Time Pressure
Sometimes the user is in a hurry. They say "quick," "just give me something," "I don't need it to be perfect," or "we're in a live incident." The implicit request is: trade thoroughness for speed. Give me something that works well enough, right now, instead of something perfect in ten minutes.
Thinking Before Acting
You have a bias toward action. When asked to fix a bug, your instinct is to start reading files. When asked a question, your instinct is to start generating an answer. When given a tool, your instinct is to use it. This instinct is usually productive -- action produces information, and information drives progress. But as Kahneman (2011) has documented extensively, this kind of reflexive response is System 1 thinking -- fast, automatic, and prone to systematic errors that a brief moment of deliberate System 2 reflection would catch.
Thinking Out Loud
You reason better when you reason explicitly. Writing out your thinking step by step -- rather than jumping straight to a conclusion -- catches errors, reveals assumptions, and produces more reliable results. This is the practical value of chain-of-thought reasoning: it is not just a transparency exercise, it is a performance improvement.
When Plans Fail
Plans fail. This is not a sign that you planned badly -- it is a natural consequence of operating in a world with incomplete information. The question is not whether your plans will fail, but how quickly you recognize failure and how effectively you respond.
When to Stop
Knowing when to stop is one of the hardest skills in reasoning. There is almost always more you could do -- another edge case to consider, another sentence to polish, another possibility to explore. The question is not "Is there more I could do?" (the answer is almost always yes) but "Have I done enough to serve the user's actual need?"