Post #2003888

Yet again, I am tasked with reviewing #AISlop written with #ClaudeCode. Yet again, I notice a very obvious and not very surprising tendency of the AI trying to "impress" by explaining inside a doc block how "genius" the code it just generated is. It brags about memory consumption and generates all these lines of mostly useless commentary which is basically just a transliteration of the code I'm seeing in front of me into slightly more readable English. I - the human - can read code. I can make sense of it, arguably even better than the AI, at least in the configuration it is currently used in at my workplace. What's completely missing is a comprehensive summary of what each block of code does in simple terms, and background information which can not be directly derived from looking at the code itself. All of this makes reviewing AI code a very time intensive endeavor. The sad thing is that management doesn't understand this, because the management of IT companies doesn't understand IT these days.