"surprising depth in ... ancient code" seems extraordinarily condescending to the past, upon which computer science has evolved little.
Also older software tended to be documented better because there was no 'fire off an email' or 'ask Bob on Slack' what this code means. The barriers to quick communication where higher, hence reading the code comments would have been the path of least resistance.
In particular, some old video games feature the absolutely most wonderful hacks to achieve what they wanted to with the available hardware, but also often lead to engines having to make particular compromises in possibility of environment design that many players failed to notice.
Many older game engines were actually incapable of stacking walkable planes on top of each other, though the levels were three dimension and one could ascend and descend, there was never any walkable surface directly under another, which was necessary for some optimizations.
Shipping meant producing that gold master build and sending it off to a floppy replication service and then got distributed to the store shelves. That 1.0 version was frozen until you'd ship a new version a year or more later. There was no second chance for that release, code had to be as close to perfect as possible, first time out.
Is it arrogance to say that medicine is far better than it was in the past?
Irrelevant. Medicine and Software Development are not correlated, just because we want them to be. That's the point. Software Development results seem to have devolved (quality metrics such as binary size for equivalent reliability, functionality, and notably time to release), despite improvements in almost every aspect of development tooling.
Also interesting to note, these quality metrics follow along expected curves (accounting for development tooling) when DUPLICATING (sometimes with small improvements) working software. This tells us a great deal about what is a large problem with modern software.
Sometimes even playing music doesn't work on my mobile phone, and I have to restart it. I never had that problem with my casette player 30 years ago. But that doesn't mean that I go back using a casette player instead of my phone for listening music.
Binary size is not a quality metric.
Server back ends don't care about it, front ends don't care about it despite the lip service paid, games sure as hell don't, etc
A small minority of web devs seem to care, together with creators of already gigantic mobile apps and as I was saying, embedded devs working in very constrained environments.
The market, as a whole, maybe has this as their 100th priority. Which means it's not.
No, they're not worthless, they're worth less :-)
Don't tell me, tell the business people financing everything we build. By and large, they don't care, it's reflected in incentives and it's reflected in what we build.
So yes, I stand by that strawman.
Another common scapegoat for developers who write bad software. I just wish we had a little more integrity as an industry, especially since so many of us insist on calling themselves "engineers".
Software designed to spec and user requirements does (where those requirements could be fast development time, high performance for whatever the performance criteria are, longevity, easy and cheap extensibility, etc.).
Binary size, as I was saying before, is like 100 on the list of priorities for most categories of software. Mobile apps sometimes have it as a priority and embedded apps frequently have it, too. But for mobile apps those limits are loosening, so soon they'll stop caring, too, and embedded apps are a minuscule percentage of all apps developed out there (most of them are web apps, especially Line of Business - LOB - apps).
Expecting code from the past to be "worse" is like expecting math from the past to be worse.
It's just improving classical computer science has less practical benefits than machine learning nowdays, so the focus of research changed.