I do agree, if someone's just going to shove what you say to an AI at what point do you ask "do we need to replace you with a shell script?" before they stop being overtly reliant on AI, that's a bridge too far for me, and I am all for letting Claude hack away at code for a while, and then review the code and refine it, a lot of the time even if its not perfect, it gets me like 80 to 90% of the way there, especially if I tell it HOW I want it written, and to consider edge cases.