What always excited me about software was that I always had renew myself, learn a new thing, change the way I think about something, reflect on what is now possible that wasn't before, etc. I.e. it was never stagnant.
In addition I could try out everything myself, and (more of less) understand what it is doing and why. When I wanted to understand RSA, I read the paper and implemented a PoC myself, or a BTree, or an LSM tree, same for Paxos and Raft.
I worked a lot on databases and "BigData" and on the re-convergence of the two. Much of this is open source, so I could play with it, change it, etc.
In that AI is indeed different. I can install (say) Ollama on my machine and play with it, look at the source code (llama.cpp), etc. And, yet, when I get a response to a prompt, even locally on my machine, I feel blind.
And I used to work on neural networks in the late 90s, when their use was limited, so I understand what they do and how they work.
How exactly was that model trained? On what data? What did it actually learn?
(Aside: Here I am reminded of early usage of neural networks to detect enemy tanks. It worked perfectly in the lab, would correctly classify enemy vs friendly tanks, and in a field test it failed terribly - worse than random. What happened? Well it turned out that the set of photos with enemy tanks mostly showed a particular weather pattern, whereas the friendly photos predominantly showed another. So what the neural network had actually learned was to classify the weather. You might laugh about this now... But that's what I mean.)
So, yeah, I can related to OP, even though I am excited about what AI might bring.