I have an idea for a Python project and want to make it part of my long-term career. I'm still fairly inexperienced with Python, Git, deployment, and the surrounding development workflow. At first, I used Claude to build the project and made rapid progress, but after a few weeks I realized I no longer understood much of the code or felt in control of it. That was discouraging because I want to keep coding and genuinely understand what I create, not just deploy something once.
I've started over in a clean repository and would like advice from experienced developers. Should I avoid AI-generated code entirely and work through problems myself, using documentation or questions only when I'm stuck? Or is it reasonable to use AI while learning, perhaps as a mentor or reviewer? I'm not looking for the fastest way to launch—I want to develop real programming ability, even if that means progressing more slowly.
5 Answers
If your goal is to become an expert, you need to get comfortable with difficulty. Struggling through progressively harder problems, checking your assumptions, and correcting your own mistakes is how you build judgment and lasting understanding. If an assistant removes all of that struggle, it may help you produce working code while preventing you from developing the skills behind it.
The answer depends on your objective. If you only want a working product quickly, let AI generate more of it. Since you want programming to become your career, write the important parts yourself. Use AI like a tutor: ask focused questions, request hints, or have it critique your solution instead of asking for large chunks of finished code.
Start by attempting every problem independently. When you’re stuck, consult the documentation or ask AI about the underlying concept and possible approaches, not just for the final code. Once you understand the solution, implement it yourself. That approach is slower, but it gives you much more control and makes the knowledge stick.
A good compromise is to use the chat interface as a demanding mentor rather than an autonomous coding tool. Explain your design, reasoning, and tradeoffs, then ask it to challenge your assumptions or lead you toward an answer with questions. Avoid handing over the whole repository or accepting large generated implementations you haven’t worked through yourself.
You don’t necessarily need to ban AI, but don’t keep code you can’t explain. Try solving each problem yourself first, then use AI to review your approach, point out gaps, or explain a specific concept. Keep changes small, read every line, and rewrite anything you don’t understand. A useful rule is that your understanding should never fall more than one commit behind your code.
That makes sense. I may eventually be comfortable letting Claude write parts of the code, but I want to reach the point where I could implement those parts myself if I had to.