I'm seeing people compare LLM generated code to "a hacky shell or Python script", as in a rough throwaway tool not meant to be built atop or understood.
Gently, I'll argue that this is misunderstanding the value of shell scripts in particular & rapid prototyping in general. The value and the point of shell scripts is to set down on disk the smallest kernel of a working idea specifically so it *can* be understood clearly, then iteratively expanded & replaced to cover surrounding special cases.







