I’d be willing to bet the classes of bugs introduced would be different for humans vs LLMs. You’d probably see fewer low level bugs (such as off-by-one bugs), but more cases where the business logic is incorrect or other higher concerns are incorrect.