Its similar for us to a certain extent. I honestly don‘t know yet where this will lead to. Personally I also don‘t really follow the arguments that the agent does the coding and the human does the understanding. In my opinion one is thinking differently about the code when not coding it by himself, on a higher level or lets even say a more superficial one. To keep the understanding on the same level you would have to limit the agent to just „typing“ but this is definitely not whats happening.

Yes, may be a skill issue[tm], may be an inherent one or it may not even be relevant anymore, we will see.

For me it currently still works well because I am working on legacy systems I more or less have a good understanding about - so I can judge the agent code. Not sure how this will be with new, green field code.

Not even starting with the discussion how it should be possible to review the sheer amounts of generated code.