I made a comment somewhat related to this on a post here not too long ago. I think one of my largest hesitations to using AI for writing code is my lack of desire to do so much code review. It turns writing code into reviewing code, which isn’t my preference. Of course code review is a necessity, and I enjoy it, but only when balanced with writing my own.

And then, as you noted, all of a sudden your codebase consists of a majority of code that you (or anyone) didn’t write, only reviewed. I think it will make a codebase less maintainable long-term because you don’t have that connection to the code you can only get when you author it yourself.