Yeah, that’s my biggest worry. I always have to hold colleagues to the basics of programming standards as soon as they start using AI for a task, since it is easier to generate a second implementation of something we already have in the codebase, rather than extending the existing implementation.

But that was pretty much always true. We still did not slap another implementation onto the side, because it’s horrible for maintenance, as you now need to always adjust two (or more) implementations when requirements change.
And it’s horrible for debugging problems, because parts of the codebase will then behave subtly different from other parts. This also means usability is worse, as users expect consistency.

And the worst part is that they don’t even have an answer to those concerns. They know that it’s going to bite us into the ass in the near future. They’re on a sugar high, because adding features is quick, while looking away from the codebase getting incredibly fat just as quickly.

And when it comes to actually maintaining that generated code, they’ll be the hardest to motivate, because that isn’t as fun as just slapping a feature onto the side, nor do they feel responsible for the code, because they don’t know any better how it actually works. Nevermind that they’re also less sharp in general, because they’ve outsourced thinking.