I've been learning programming for a few months and started turning to AI whenever I got stuck. It was convenient at first: I could describe a problem and get working code in seconds instead of spending an hour debugging or reading documentation.
The downside became clear when I tried building a small project without AI. I recognized what the finished code should look like, but I didn't know how to reach that result on my own. I had been skipping the uncomfortable parts—understanding error messages, searching documentation, debugging, and working through problems step by step. My projects were improving, but my own skills may not have been improving at the same rate.
I don't want to stop using AI completely because it's genuinely useful. I'd like to use it more deliberately, though. What rules or workflows help people use AI as a learning tool without letting it do the actual learning for them?
4 Answers
Use AI after you have something of your own to evaluate. For example, write the tests or implementation first, then ask it to review your code for bugs, edge cases, or confusing parts. Treat its suggestions as opinions rather than truth—you still need to verify them and understand why they matter. Documentation links, diagrams, and test ideas can be more useful for learning than generated code.
A useful rule is to use AI as a tutor, not an implementation service. Ask it to explain a concept, point you toward relevant documentation, quiz you, generate practice exercises, or review code you already wrote. You can also ask it to explain why a particular approach works instead of asking how to implement the whole feature. Don’t move on until you could explain the answer yourself.
The struggle is the part that builds the skill. Try solving the problem yourself first: read the error, inspect the relevant code, check the documentation, and write down a hypothesis. Only ask AI after you’ve made a real attempt, and ask for a hint or explanation rather than the finished solution. If it gives you code, make sure you can explain every important line and reproduce the fix later without copying it.
That uncomfortable stretch of staring at one error for ages is frustrating, but it’s also where debugging instincts start forming. If AI removes that entire process, you lose the chance to build those instincts.
It’s also reasonable to set aside regular AI-free sessions. Build a small project, recreate an earlier exercise from memory, or fix bugs without assistance. You’ll quickly discover which parts you truly understand and which parts you only recognized when the tool produced them. AI will probably be available in future jobs, but strong fundamentals still matter because you need to judge its output, describe the problem accurately, and notice when its answer is wrong.
That makes sense. I think I was confusing being able to recognize a solution with being able to develop one. I’m going to try attempting problems independently first and using AI mainly for hints, explanations, and reviews.
Generating small exercises and tests from each chapter of a programming book works well. It reinforces what you just learned while still requiring you to write the solution yourself.