I think LLMs constitute in a real way a potential "end" of software. if they have their way we are going to start producing generations of people who cannot actually write software anymore, and it's completely unclear in that scenario how software maintenance let alone production is supposed to happen. who is maintaining the Linux kernel in this world. who is maintaining Python, CUDA, all of the conventional software that makes up the world and, incidentally, lays the foundation for LLMs themselves.
the guy in this PR is not doing it, and this is the kind of guy LLMs produce. in the near term he is an enormous waste of time and resources which is a wrench in the already precarious machine that is software production. but in the long term if the project of LLMs wins this guy is the baseline software developer, having known nothing but AI generated code his whole life and career, having never struggled to make a computer program do what he wanted it to do, bending his brain into the appropriate shapes to think in terms of computation. just "prompt engineering" and an insistence that the face he sees in the cloud sees him too.
I've had two conversations with people where I asked who is maintaining Linux kernel in this scenario and the answer on both occasions was that the expectation was that the models would get good enough that they could do it with good prompting. and, like, I don't know how to interpret that other than the end of software as a human project.