The issue is there are multiple dimensions of quality, and people mean different things when they talk about it. Personally I think code quality for its own sake has always been a false flag that only gained currency because of skill moat of software development. On the other hand, user experience and fitness for purpose has always mattered a great deal, all the moreso when code is cheap, and this is definitely tied to code quality. The connection is even more stark when you talk about evolution of a system over time and data integrity. Here, the tacit assumptions, invariants and external interfaces chosen have a huge impact on the quality of the system and fitness for purpose over time. Every choice made affects the future, and it's very easy to box yourself in with bad choices, which ultimately depend on the offline human concerns that system is impacting.

I don't think any of this changes with AI, the only thing is that it's possible to handle much more incidental complexity. Ultimately I expect it to land on a similar equilibrium in terms of software quality that is gated by what the users of the system will tolerate. The more things change, the more they stay the same.