> Yeah, but you can't trust your own code!
No, really, you can't trust the AI's code. It spits out code that references functions that don't exist. It spits out code that does something you didn't ask it to do, or doesn't do something that you did. It spits out code that does something vaguely related that the Stack Overflow guy it's cribbing from did, but you aren't doing. It spits out code that confuses your syntax for JS's halfway through. You can skim other people's code; you know what common failure modes are at their experience level, and you know where the complicated bits are that will house the bugs if there are any. You cannot skim the AI's code. Every word of it must be examined. You must be in full reviewer mode all the time, which is an unproductive state to be in when actually writing code, which makes specifically Copilot less useful. You cannot use it to replace a no-code tool, because you must understand the language it's emitting.
I find that proponents shift between whether its ideal use-case would be Copilot or no-code; any flaws with the one approach get interpreted from the perspective of the other, where they can be dismissed.