The problem is that coders are used to dealing with code. GPT-4 is a robust processor for semantics as conveyed by strings of characters. The whole point is that you don’t need code. You just ask it what you want.
But programmers love to think they can still make improvements to such a system using code. In reality, any improvements that can be made are the responsibility of f-strings and/or a templating mechanism and even that may be overkill in many cases.
Often there is still a lot of programming to do, as a lot of the use cases involve some sort of integration into a bigger systems, or you might have to process larger quantities of data which gets you into queues and for example managing rate limits.
That said I 100% agree that often the basic tools are sufficient to solve these problems, and where it gets complex there are many battle tested solutions.
But programmers love to think they can still make improvements to such a system using code. In reality, any improvements that can be made are the responsibility of f-strings and/or a templating mechanism and even that may be overkill in many cases.