Continuing to learn

If there weren't free to use AI models available, I'm not sure whether I would use them as much as I do. I don't use them a lot, or rather, I suspect that I don't use them as much as a lot of my peers (at least from the impression that sites like twitter give me, although, this is also an unreliable source lol). The fact is though that there are some free to use models, and some of them are decent enough. I've had success with DeepSeek V4 Flash for instance.

For pretty much the entirety of my career as a developer I have spent a good amount of my free time trying to learn more about programming and computers. My go-to outlet for learning is to implement something. I pick a problem in an area I want to improve in, or a problem that I find interesting, I come up with some project idea that fits, and then I get building.

These projects have included things like chatbots, parsers, compilers, web apps, databases, and tools to help my day-to-day. What is consistent between them all is that I hand-wrote the code and learned new things along the way. Even in the early days where I was basically copying someone else's code from GitHub by hand, I still did this intentionally and did not simply copy-paste in order to try and absorb ideas as I wrote the code, even if I didn't come up with them myself.

Nowadays we have AI models available, and it can be hard to resist using them when you know that they can potentially write the code so many more times faster. This feeling can be hard for me to avoid at work, where I care more about what I actually produce. If I can potentially produce the same work at a faster rate, why wouldn't I give it a shot? It's a win-win for me and the company. Of course, getting the AI to produce the right thing is the hard part.

I digress, maybe that is a topic for another post. What I am interested in here is the effect that AI is having on my personal learning projects. There is now the temptation to ask AI to just build it for me. I am implementing already solved problems, so it already knows exactly (kind of) how to solve them. But by doing this I am entirely missing the point of the exercise and cheating myself in the process.

I found myself asking for what purpose am I actually "working" on projects like these if I am just asking an AI to build it? The point is learning, and what am I actually learning? Not much.

Today I might still use AI in a project like this to generate tests, or to give me some advice if I found myself stuck, but I think I am now at a point where using it to generate anything but trivial code that I already understand is off the table.

If there were a scenario in which I would use AI in a project who's purpose is for me to learn, it would have to be that the project focuses on learning how to better use AI.

More interestingly, I think I'm okay with using AI to generate code in projects that aren't expressly for learning. Is that actually okay? This is why I am still on the fence about my usage of AI at work and when building real things. Something tells me that generating code and not understanding it can only lead to trouble down the line. The best path to me seems like a happy medium between the two.