I agree with the premise, but it's quite painful in practice constantly probing and prodding for justification and explanation -- especially because _even with_ the justification, explanation, etc, one's mental map / the "topology" of the thing being built is only very loosely being populated as a result of the conversation. I say this having continuously tried to find a way to keep my learning rate comparable to if I was writing the code myself, and having somewhat failed.
I'm starting to wonder if the thing to address is the anxiety itself rather than the "fuzziness about the code" that creates the anxiety - and more explicitly model myself as an engineering and/or product manager counterpart to these things. I wonder how non-IC EMs/PMs do it - it seems maybe fundamentally anxiety-inducing? – but they _do_ do this already (tolerate the fact that the underlying technical system is not fully within their grasp).
A few different differences.
Unlike with an ai, a pm knows what it means to understand something. Just in general, they have had the part of the human condition which is gaining an understanding of something. Not memorizing a fact or learning how top perform some recipe steps, but understanding several dimensions of something.
They don't code, but there are other things they have learned and understand, and witnessed other people showing how they do or don't understand that thing. It could be about fishing or calligraphy or being careful what you say to kids or anything not just work stuff. From that they know how to recognize it in others on other topics. Modulo the effect of the exceptionally good and bad on both sides, a good bullshitter can fool a simple pm, but the exceptions to a generality are irrelevant.
And they know that some other people do understand the code. They haven't performed all the steps to reproduce someone else's reasoning and correct final result, but they don't have to do that every time to know that it's possible to do and would arrive at the same or equivalent result. It's not faith, or not blind faith like religion. It's just letting someone else do a job that you know if you wanted to you could do exactly the same and it would work exactly the same. You don't have to carry a load of grain from one pile to another to know for a fact that it's possible to do and roughly what it takes to do it.
They also know how to detect cues about consensus, or lack of it. When most people who understand a topic, you can tell, without relying on any simple rules about what words are spoken, you can tell when most of the people who have put in the time to understand something agree or disagree on some premis. And the people you are guaging that consensus from are again other humans who you share the human condition with. You have a power to understand and interpret them that you don't for anything else.
They may also understand some level of code. It will have been explained to them at some point to some degree. They will have some sort of simple example in their history where they were walked through something in some class or a meeting.
These things are all missing with an ai.