Even if you are using AI to program, never copy paste, read and write the code by yourself. Next time you find a similar problem you are more like to remember how to solve it

Avoid the frontier models. They make it much easier to say "I want X app", and theyll spit it all out. If you use an older model, you have to break down the task a lot more, which at least forces you to think about the architecture. And each task is easier to review.

I use gemma4-12b. It is fine for generating code, but you definitely can't one shot things with it.

I love the challenge of building whatever my imagination dreams up. Often my problem is stopping.

I think if you are learning to code you cannot use AI to generate code. You can ask questions and get small snippets to explain concepts but you must be solving the actual problem yourself even when it feels simple and feels like you know the solution.

I used it once and had to draw a hard line. Because it was so much better than me that if I used it then I would have no reason to write any code.

The way I use it is to talk through the problem and my solution and why with an instruction to not generate code, give suggestions and to be critical and teach.

Then once I'm happy with a high level plan I write the code and test it and when I get stuck I give it a bit of time trying to figure it out myself then go to AI for review.

The motivation to not use it is because i want to learn to code. The end result doesnt matter as much as the process to me. If I had 10years experience I'd be vibe coding like a man possessed.