The paradox of perfection
Then Codex came. Then Claude. Then everything else.
And a paradox appeared that I didn't see coming.
I describe what I want. AI generates code. It works. It's clean. It has error handling. It has tests. Documentation built in. Seriously.
So I won, right?
No. I stopped writing.
Not because the code is bad. Because the code is too good. When a tool does something better than you - and does it instantly - suddenly something disappears that wasn't just a "side effect" of the process. It was the process itself.
When AI writes code for me, it removes from the process exactly what was valuable about it: the struggle, the error, the learning.
Is this like outsourcing running to a car? You reached the finish line. But why did you run?
What I actually lost
It's not about lines of code. It's about the learning loop.
When I wrote by myself - even badly, even messily - I was training a particular kind of thinking. Decomposing problems. Tracking down the source of a bug. Reading error messages instead of ignoring them. Learning from documentation instead of copying the first result from Google.
AI shortens that loop to zero. Not to a minimum - to zero. The effect: I no longer get errors I don't understand. I get solutions I don't understand. And that's a different, more subtle kind of loss.
A junior developer once told me: "I've been using AI for a year and I've stopped understanding what I'm doing." He didn't say it proudly.
He was scared.
There is one counterargument worth raising honestly: AI can also create new learning loops. Reading generated code and asking "why this way?" is learning. Understanding the model's decisions, comparing your own intuitions against its solutions - that's learning too. Just different. Whether this new loop replaces the old one or merely supplements it - I don't have an answer yet.
The bigger question: AI and learning through experience
This doesn't only apply to hobbyists. And here it's worth making a distinction the article easily blurs.
For a hobbyist, losing the learning loop is a personal loss - you lose something that brought joy and engagement. For a professional - especially a junior, especially in the early years - it's a professional risk. Understanding what you're doing is the foundation of further growth. Handing that understanding over to a model at the start of your career is an investment with an uncertain return rate.
Technical students who will never have to struggle with a basic bug - because AI will fix it instantly. Junior developers who won't go through the phase of "I have no idea why this doesn't work" - because AI will tell them. Everyone who learns by making mistakes - which is most of us - in conditions where AI removes mistakes from the process.
What happens to learning then?
I honestly don't know. There are studies pointing both ways. There are arguments that AI is like a calculator - frees you from manual computation, lets you think at a higher level. And there are arguments that if nobody calculates by hand, nobody understands what they're calculating.
Both sides have a point. And that's exactly the problem.
What to do about it
Not "don't use AI." That's foolish and impossible. And that's not the point.
Three proposals - tested on myself:
1. An AI-free zone. One task a week that you do yourself. Rough, slow, your own way. Not because the result will be better - it won't. Because the process has value independent of the result. How do you know if it's working? After a month, ask yourself: do I still feel like I "understand what I'm doing" - or just "delegating to the model"?
2. Try first, ask AI later. Before you describe the problem to AI - spend 20 minutes on your own attempt. Even if nothing comes of it. That's 20 minutes of thinking practice, which AI will then confirm or correct. The sign it's working: you start noticing where AI makes mistakes or unusual choices. That means you understand.
3. Read the code AI generates. Don't accept it like a restaurant order. Ask questions: why this way? what's happening here? could I write this differently? Learning happens right there - in the friction between what AI did and what you would have done. You don't need to understand everything immediately - but it's worth keeping a list of things you want to understand before accepting the code.
Will this restore my hobby? I don't know. Maybe this is the new hobby - understanding AI instead of wrestling with code.
A question to close with
Do you have something AI can't take from you - because the value is precisely in doing it yourself, even messily, even without purpose, even at 2am?
If you do - protect it.
Because tools that do things for us are multiplying faster than our ability to decide what we actually want to keep learning.