"If", indeed. That's the problem of metrics: it depends on what you measure, and sometimes it's hard to get a meaningful answer in the short term.

If you ship more for less, but less maintainable or less correct, then it's not necessarily an upgrade. Always the same question: does it benefit the developer? The product? The company?

It was already possible, without AI, to look like one is doing a great job ("they are producing so much! Let's promote them!") but actually just building a bad codebase. The art being to get the promotion and move to the next step before the project implodes.

Not saying that AI necessarily ends up doing that, but it most certainly help.