The reality is closer to the "it has always been like this" end of the spectrum, than to the other end. But, it makes some sense that software which you will be working on for 20+ years is more worth optimizing than software which you will never see again after 1-2 years from now, because you will switch jobs or the company will get bought and forced to switch to the buying company's preferred language or any of a dozen other reasons.

But, in the course of my career I've looked at a lot of really old software as well, and it wasn't better. I've never worked on COBOL, but people I know who have, were not impressed and don't miss doing it. So I think it has always been the case in software that most work is done in a manner which is later cause for regret.