Is AI going to slowdown the creation of new frameworks and libraries?

I work in game dev and a lot of game code isn't nearly as well represented in the training set for LLMs as other areas like web dev. Because of this I haven't found LLMs nearly as useful for generatin

I work in game dev and a lot of game code isn't nearly as well represented in the training set for LLMs as other areas like web dev. Because of this I haven't found LLMs nearly as useful for generating game code and I have tried experimenting with it for about the last year with only small gains. What confuses me is that if these models perform decently on tasks that are well represented in their training set, e.g. frontend with react, what happens when eventually new frameworks/libraries or even languages are created and there no longer exists a decent collection of data for these models to train on? If the majority of coding at that point is just AI generated code then it will just be training on its own slop which leads to model collapse, or there won't be enough data and they won't perform well. If the models can generalize so well that this isn't an issue, then I guess every area of programming that people aren't able to get AI to write decent code is apparently just a skill issue? I added this part because I already know that a dozen people are going to reply with "ai writes all my code, if you aren't getting 100x gains then you're using it wrong!". I've found AI useful, just more like a 5-10% boost, not 200% or anything. As an indie game dev I would like nothing more than to have an inexpensive prodigy programmer to assist with my development. This leads me to then wonder if developers are becoming reliant on AI to do their work, that innovation might end up just slowing down to accomodate AI. As in, whats the point in a new techonology if no one uses it because it doesn't work well with their AI workflow? Or are we going to see waves of AI jumping between useful and useless for when new techonologies arrive or there is too much AI generated content diluting human generated content.