But do you also sometimes leave out AI for steps the AI often does for you, like the conceptualisation or the implementation? Would it be possible for you to do these steps as efficiently as before the use of AI? Would you be able to spot the mistakes the AI makes in these steps, even months or years along those lines?

The main issue I have with AI being used in tasks is that it deprives you from using logic by applying it to real life scenarios, the thing we excel at. It would be better to use AI in the opposite direction you are currently use it as: develop methods to view the works critically. After all, if there is one thing a lot of people are bad at, it's thorough critical thinking. We just suck at knowing of all edge cases and how we test for them.

Let the AI come up with unit tests, let it be the one that questions your work, in order to get a better perspective on it.