I remember before I started using AI (I was pretty late to the game), I assumed all the bugs and instability were because people were writing software with AI. Now that I've seen how good the code that AI writes can be (CAN be), I'm moving more to the thinking that it's from AI being used to find vulnerabilities.
It’s more that management pushes more output and with AI it’s very easy to achieve but that will create slop.
I've never had AI one-shot even a 10 line script that I didn't manually revise. To be fair, it usually produces something that works, but after seeing it I'll realize that my prompt was too vague or I left out some edge case. So it's the old problem with software "it does what was asked for but not what was needed."
So I speculate that a lot of problems with AI code are just that it is accepted uncritically because it builds and doesn't have any syntax errors.