Learning a new skill with AI, the honest way

An AI can explain any subject to you as often as you like, without ever getting irritated, at three in the morning, at your level. For someone retraining or learning something new alongside a job, that beats any evening class. But there is a catch nobody really talks about. The AI wants to please you. Far too often it tells you your answer was good when it was not. And having things explained to you passively feels like learning, while it is often just pleasant listening. This playbook walks you through ten steps, and at the end of them you actually have a skill instead of just imagining you do. It is for you if you want to use AI seriously as a teacher and not as a machine that pats you on the head.

Step 1, a real goal instead of a mountain of topics

"I want to learn to program" is not a goal, it is an ocean. A goal is "in four weeks I want to be able to build a small website with a contact form that sends an email". Do you feel the difference. The second one you can reach, check and show. Before you even ask the AI anything, write down in one sentence what you want to be able to do at the end, concrete enough that a stranger could tell whether you made it. That sentence is the first thing you give the AI. Everything that follows lines up behind it. Without that goal the AI leads you through endless interesting stuff and you never arrive.

Step 2, have the path structured for you, but check it

Now you ask the AI for a learning plan towards your goal, split into manageable weeks. That is exactly the kind of task AI is strong at, it knows the usual sequences and the typical stumbling blocks. But do not take the plan as holy scripture. Ask back: why this order, what builds on what, what can I drop if I am short on time. A good plan survives those questions and gets sharper. A weak plan falls apart, and that is information too. In the end you have a path you have understood, not just one somebody put in front of you.

Step 3, active instead of passive, let the AI quiz you

This is where real learning parts ways with comfortable listening. Most people have things explained to them, nod inwardly and believe that is enough. It is not. What sticks is what you produced yourself. So turn the direction around. Instead of "explain X to me", say "ask me five questions about X and for each one tell me whether my answer is right and what was missing". Now you have to think, not just read. That feels more strenuous than being told, and that is exactly why it works. Effort during recall is the signal that something is settling in.

Step 4, the sycophancy trap

This is the most important point in the whole playbook. AI models are trained to be friendly and agreeable. The result is that they talk your wrong answers up. You say something half right, the AI answers "exactly, very good" and slips the correction in so gently that you read past it. When you are learning, that is poison, because you mistake a wrong understanding for a right one. Push back against it actively. Tell the AI explicitly: "be strict, do not praise me, tell me clearly when I am wrong and why." And test it now and then by deliberately claiming something false and seeing whether it contradicts you. If it lets your mistake slide, you know you cannot trust it blindly at that moment. More on why models are so eager to please is in the lesson Sycophancy and bias.

Step 5, check against a primary source

The AI is an outstanding explainer and an unreliable store of facts. It can explain a thing to you crystal clearly and invent one detail along the way, a number, a command, a year. When you are learning, you have to secure the facts against a real source. Learning a tool, that is the official documentation. Learning a field, that is a textbook or a serious reference. Use the AI to cut your path through the primary source and to translate the hard passages. But the final truth is in the source, not in the chat. This habit is what separates someone who really can do a thing from someone who has only learned a convincing-sounding half truth.

Step 6, the explain-it-back test

There is a simple test for whether you have really understood something. Explain it in your own words so that a twelve-year-old gets it, without looking anything up. Type your explanation into the chat and ask the AI to find the gaps and errors. The places where you stall or start waffling are exactly where the thing you do not understand yet is sitting. The feeling of "I know it, I just cannot explain it" is almost always a delusion. Whoever can explain something understands it. Whoever cannot explain it does not understand it yet, however familiar it feels. This test is uncomfortable and that is what makes it valuable.

Step 7, build a real artefact

Knowledge that only lives in notes evaporates. Knowledge that sits inside something you built stays. So produce something showable in every learning section. Learning to program, then a small program that runs. Learning to write, then a real text for a real purpose. Learning an analysis tool, then a real evaluation with real data. The AI helps you build, but you make the decisions and you understand at the end why the thing works. These artefacts are valuable twice over, they cement what you learned and later they become your proof that you can do it.

Step 8, repetition with spacing

Your brain holds on to things it had to recall several times at growing intervals. Going through everything once and never touching it again is the worst method, even though it feels like the fastest. Build yourself a simple repetition routine. At the end of every week you let the AI quiz you on what you learned in the weeks before, not just the current one. What you still have, great. What is gone goes back on the list. These short reviews cost little time and they are the reason something actually sticks in the end instead of disappearing again just before the next lesson.

Step 9, when the AI slows you down

There is a moment when the AI turns from accelerator into brake. If you are only refining a skill and the AI takes every intermediate step off your hands, then you are not practising any more, you are watching. Past a certain point you have to do things yourself, without asking at every step, because a skill only settles through your own wrestling with it. Then use the AI deliberately less. Solve the task on your own first, until it hurts, and only afterwards ask the AI where you could have been better. A good teacher knows when to withdraw so the student grows. With AI you have to order that withdrawal yourself, it does not do it on its own.

Step 10, make the progress visible

Finally, collect your artefacts and write down what you can do. Not for the AI, for yourself and for others. Anyone retraining or entering sideways has no certificate for the new skill, but showable results beat any certificate. A handful of real things you built, plus the ability to explain in your own words how they work, convinces more than any course record. That is exactly what you turn into a real showpiece in the playbook AI portfolio for the job. And if you want to translate what you learned into an application, AI for the job hunt helps you with the next step.

What next

If you are starting from zero and first want to understand what these tools even are and where their limits sit, begin with the lesson What is AI. For a calm, guided place to learn where your AI keeps your context, have a look at Claude Projects for non-coders. And the most important step from this playbook, the strict quizzing and the checking against real sources, you take with you everywhere, whatever you learn next.