How to Use AI Without Becoming Dependent on It

The people who will get the most from AI over the next decade are not the ones who delegate the most to it. They are the ones who use it as a force multiplier while staying in control of their thinking, judgment, and domain expertise. The two things — high AI usage and strong independent capability — are not in conflict if you use the tools deliberately.

The problem is that passive AI use is much easier than active AI use. Asking a question and accepting the answer requires almost no effort. Forming your own view first, then using AI to challenge it or accelerate it, requires discipline. And over time, the path you choose shapes the kind of thinker you become.

The Dependency Mechanism

Cognitive offloading — outsourcing mental effort to external tools — is not new. Calculators did not make mathematicians worse. GPS did not make navigators unable to read maps. But the research on cognitive offloading does show that repeatedly outsourcing a specific mental task reduces engagement with that task, which over time reduces the development of the underlying skill.

With AI, the offloading happens at the level of reasoning, judgment, and synthesis — not just calculation. When you ask AI "what should I do about X" and accept the answer, you are offloading the judgment. Do that enough times in enough contexts, and you lose the ability to make that judgment without the tool. Not because the tool is bad — but because you have stopped exercising the judgment muscle.

The Two Modes of AI Use

There is a productive and an unproductive mode of AI use, and they are defined by who is doing the thinking.

The distinction is not about frequency. You can use AI heavily and still maintain independent capability — if every use involves active judgment. You can use AI sparingly and still become dependent on it — if every use involves passive acceptance.

What Makes You Valuable Cannot Be Delegated

The value of your work comes from things AI cannot provide: contextual knowledge of your specific situation, relationships and trust built over time, judgment about what matters and what does not, and the ability to make decisions under uncertainty with stakes attached. These are not capabilities AI lacks because it is not smart enough — they are capabilities it lacks because it does not have skin in the game, does not have your context, and does not carry consequences.

Using AI to write the first draft of a document is fine. Using AI to form your view on a strategic question in your domain — and then presenting that view as your own without having done the thinking — gradually erodes the capability that makes you trusted to answer those questions in the first place.

The Evaluation Gap

One of the most important skills in an AI-augmented world is the ability to evaluate AI output. This requires existing domain knowledge. You cannot tell whether a legal document AI drafted contains errors without legal knowledge. You cannot tell whether an Apex trigger is bulkified correctly without Salesforce knowledge. You cannot tell whether a financial analysis is sound without financial knowledge.

This is the deepest form of AI dependency: using AI in domains you do not understand, and therefore being unable to identify when it is wrong. The output looks plausible. You do not know enough to know it is incorrect. You deliver it and later discover the error.

The protection against this is building underlying domain expertise independently of AI — through deliberate practice, through reading source material, through making and learning from mistakes. AI can accelerate the journey, but it cannot substitute for it. If you skip the foundational work entirely and rely on AI output you cannot evaluate, you are not a skilled professional using a powerful tool. You are a conduit delivering outputs you cannot stand behind.

Practical Rules for Healthy AI Use

Form your view before asking AI

Before you open any AI tool for a thinking task — a decision, an analysis, a piece of writing — spend five minutes forming your own initial position. Write it down. Then ask AI. Compare. If AI's response makes you immediately discard your view without understanding why it is better, that is a signal. If AI's response gives you new information you genuinely did not have, that is the tool working correctly.

Own the final product

Your name is on it. Not the tool's. Before you submit, publish, or deliver anything AI contributed to, be certain you can defend every claim, explain every decision, and identify any errors. If you cannot, it is not ready — regardless of how confident the output sounds.

Practise without AI in domains you want to develop

If you want to develop as a writer, write the first draft yourself before AI touches it — every time, not occasionally. If you want to develop as a coder, write the logic yourself before asking AI to generate it. Deliberate practice requires effort and struggle. If AI removes all the struggle, it also removes the development. Be intentional about which skills you are actively building and protect the practice space for them.

Use AI for tasks below your expertise floor, not above it

Use AI to do things you know how to do but that are tedious, repetitive, or below the level where your judgment adds value. Use it to speed up the last 20% of a task, not to skip the first 80% where the real thinking happens. The more valuable the thinking, the more important it is that you do it.

Monitor your ability to work without the tool

Once a month, spend a day doing your most AI-assisted category of work without AI. Notice how that feels. If it feels impossible — if you genuinely cannot do the task without the tool — that is a signal about where your capability has eroded. Use that signal to direct where you need deliberate practice.

AI as a Thinking Partner, Not a Thinking Replacement

The most productive use of AI is as an interlocutor for your own thinking — a tool that pushes back, expands your view, identifies blind spots, and speeds up the parts of work that are execution rather than judgment. This is different from asking AI what to think and then adopting its answer.

The distinction maps onto how experts use assistants. A senior lawyer uses a junior lawyer to do research and draft documents — but reads everything, corrects mistakes, and makes the judgment calls. The senior lawyer gets faster and more capable over time because they are always evaluating, always exercising judgment. A lawyer who delegates the judgment along with the drafting — and who stops building the capability to evaluate — becomes dependent on their assistant in a way that erodes their own expertise.

AI is a very capable junior colleague. The question is not whether to use it heavily. The question is whether you are staying the senior partner — or gradually abdicating that role.