Knowledgeable human friends also make mistakes! They are not infallible experts who never get things wrong.

Both humans and LLMs may know more than you and may have additional perspectives and experiences to draw on. Just like you'd not trust a knowledgeable friend to be omnipotent and 100% correct 100% of the time when you ask for advice, neither would you an AI.

But the point I was trying to make was to not use AI to implement things for you (even tests) as you're not going to learn if the AI is doing the repetition for you. You can supervise yes, but only when you've been there and learnt the ropes already (i.e. when you are already an experienced software engineer and know the details)