It makes sense in theory, but this is the type of thing I want to see data on before I consider it a possible problem, even.
No sense in worrying about things that you can't control, can't prepare for and have not occurred yet, after all. :)
Right now, today, when choosing a library, top of my mind is "has this been around long enough to be in GPT4's training set?"
Because to give up the 5-10x productivity boost GPT allows for, a library better be really good.
I've seen a couple companies offer a RAG enabled ChatGPT for their SDK, but without a large number of stack overflow and medium posts detailing how to use the SDK in different scenarios, well, there isn't much help an LLM can offer that I cannot just get from whatever is on a company's site.
Back in the day Microsoft used to pay boat loads of money to have absurdly complex sample applications shipped alongside new libraries and SDKs. The sample apps were often fully featured end to end solutions that had been gone over by testers as well.
The idea being, since there wasn't any really good online help back then (aside from news groups and such, which were useful if you knew about them), you had to rely on sample code or on books, books which Microsoft also commissioned to be written.
Maybe someday we'll get back to that point, companies paying money for large volumes of sample code to be written, but this time with the goal of having it ingested by LLMs.