Generative AI is turning a large part of what I know how to do into a commodity. That isn't a prediction, it's visible in my day to day. The models write most of my code and they're better at it than I am on the majority of use cases.
That's exactly the moment I decided to start learning again, after about a year of not really bothering. Learning something a machine already knows better than me sounds like a waste of time, and a year ago that's precisely why I stopped.
Why I stopped
For years I tried to keep training continuously, without any particular plan. Reading, side projects, digging into whatever I was curious about at the time. My work is mostly web, a lot of front-end but also full stack, and my side projects go wider than that. At some point I mostly stopped, because I had the impression I was wasting my time.
The reason was fairly visible to me. My programming knowledge was becoming a commodity. Everything I had taken time to learn, at school, from reading, or on the job, the model already has in its knowledge base, and it trained on far more cases than I will ever run into personally.
I still think that's true. It's more true today than it was then.
The commodity argument turned around on me
What I got wrong wasn't the observation, it was what I concluded from it.
The more the models improved, the more my work moved away from writing code and towards deciding whether the code was right. That second job needs more knowledge than the first one, not less. The knowledge became a commodity. Knowing whether you were handed the right one didn't.
I didn't work that out by thinking about it. I worked it out by spending a few months doing something I wasn't already good at.
A few months of mobile
Over the last few months I've done a lot of mobile, because I became the main contributor to the daFi AI mobile app. The stack is React Native, which has a lot of patterns very close to React on the web, so a good part of my experience carries over. But I have less experience in mobile, and I can see it. A feature I have to implement on the mobile app takes me more time than the same kind of feature on the web.
Sometimes I have the impression that it's the platform doing that, and that even with more experience it wouldn't have changed much. There are problems like the liquid glass of iOS 26 on iPhone that I don't have to deal with on the web for now.
But it isn't only the platform. My lack of expertise there costs me in two places.
- My prompts aren't precise enough to reach the right result directly.
- If the AI goes in a wrong direction, it's much harder for me to see it.
The second one is the expensive one. On the web I catch a bad approach almost immediately. On mobile I sometimes catch it three steps later.
Learning again, differently
So I've started learning again, but not the way I used to.
I'm not trying to be able to write the code, because I'm not going to write it. What I want is to read it fast, to know what a good approach looks like on a platform I don't know well, and to notice what's missing from what I asked for.
In practice that means spending more time on mobile, precisely because I'm less comfortable there. The uncomfortable stack is the one worth learning.