I think my response would be that, all of those lessons are trained into an llm, and are also available as documentation for every platform/OS. So if the llm can just read the docs for three operating systems and generate three native binaries following three sets of documentation why not let it? Humans obviously cannot, but we shouldn't let our hangups impede our tools.