And a recent study found out ended up in low-quality code. Even if it didnt, you would have to carefully read the resulting code and make sure that it wont break something else somewhere when you plug it into your repo. Its literally reading someone else's code for a pr review, but as opposed to a colleague's pr, you cant be sure of the skill level of the author. Its much easier to write that code yourself to avoid all those problems and the potential IP issues. And if you use it for very short, self-contained bits of code to avoid the quality problem, it brings nothing new to the table: Template code, autofillers, code completion, boilerplate code already existed for a long time now. And they are already vetted and tested.

But its easy to see why people who are not into the technical side of tech are very enthusiastic about AI-generated code: For those who dont know how to program and dont know how to use IDEs, stackoverflow and other resources to create code, it looks like a godsend But when you get out of the creating-an-mvp-to-test-an-idea level and actually get into the building of the actual app, things work the way how they did before gen-AI.