On the other hand: In terms of building software using agents I wouldn't call developers students, specifically when it comes to letting AI write the code, but its a similar concept: Good developers know how to architect and guide the coding agent, bad ones keep asking it to do the work they don't understand, or even if they do understand it, they don't take the time to stop and architect the problem. I've had amazing output from Claude Code, but apparently a lot of devs feel it is inadequate, some people it seems want the AI to code it perfectly in one shot, I go back and tell it what to fix, to add tests, to not change existing tests, etc.