Not sure that I'd look at python package stats to build this particular argument on.

First, I find that I'm using a lot fewer libraries in general because I am less constrained by the mental models imposed by library authors upon what I'm actually trying to do. Libraries are often heavy and by nature abstract low-level calls from API. These days, I'm far more likely to have 2-3 functions that make those low-level calls directly without any conceptual baggage.

Second, I am generalizing but a reasonable assertion can be made that publishing a package is implicitly launching an open source project, however small in scope or audience. Running OSS projects is a) extremely demanding b) a lot of pain for questionable reward. When you put something into the universe you're taking a non-zero amount of responsibility for it, even just reputationally. Maintainers burn out all of the time, and not everyone is signed up for that. I don't think there's going to be anything remotely like a 1:1 Venn for LLM use and package publishing.

I would counter-argue that in most cases, there might already be too many libraries for everything under the sun. Consolidation around the libraries that are genuinely amazing is not a terrible thing.

Third, one of the most recurring sentiments in these sorts of threads is that people are finally able to work through the long lists of ideas they had but would have never otherwise gotten around to. Some of those ideas might have legs as a product or OSS project, but a lot of them are going to be thought experiments or solve problems for the person writing them, and IMO that's a W not an L.

Fourth, once most devs are past the "vibe" party trick phase of LLM adoption, they are less likely to squat out entire projects and far, far more likely to return to doing all of the things that they were doing before; just doing them faster and with less typing up-front.

In other words, don't think project-level. Successful LLM use cases are commit-level.