I've had that same sense starting years ago, and this is the first time I've seen another programmer articulate it. For me, the disillusionment started with Microsoft Foundation Classes and J2EE. They made creating systems challenging in all the wrong ways, and the learning not really learning. Now that it no longer matters whether I learn the latest fad or not, I'm happy to let AI work the libraries. But I'm grateful I'm not entering the field today and that AI wasn't around when I started, because I never would have learned what it really takes to create something from nothing. Still, ML is and damn fun, or more accurately, seductive.
That's a bummer. While I'm not in your shoes, I can understand the excitement of solving a new library's "puzzle". And, how that may seem less valuable when a code generator can nearly instantly apply itself to using the same library.
Have you tried branching-out into different areas of programming?