One of the reasons I am not yet worried about LLMs is that this is not actually possible, at sufficient levels of domain complexity, and contextual complexity.

One of the reasons I am not yet worried about LLMs is that this is not actually possible, at sufficient levels of domain complexity, and contextual complexity. The domain complexity that LLMs start failing is pretty low at the moment. As well, as domain complexity increases so does the required amount of precision, and again I find LLMs start getting less precise when put in the deep end. You don't tend to find this kind of software in your "programmers programming". You tend to find this kind of software in industry, where large businesses consist of wild business rules built up over decades, and that has been half-codified into software. I just don't see LLMs swooping in to take that over.