Back when software was a craft

Agreed, but with the caveat that what you're describing is the bleeding edge of any industry. Even in manufacturing, the bleeding edge of an industry is craftsmanship - building new injectors, engineering systems that haven't existed before. As the bleeding edge becomes cutting edge and then passes into mature tooling, glue work dominates (and for good reason, most of us don't have a good reason to rewrite compilers or task schedulers) unless your use case is far off the beaten path.

This is not to say that a craftsman level understanding of the underlying technology is useless in glue work. The best parts of HN and the best of modern compsci is done by people who wondered if their tooling could be better - and the answer is yes. As the landscape changes, no one's going around updating the tools you use to be the best they can be. Someone here recently rewrote ripgrep, and I've been using it as a faster version of grep (for specific uses), for example. Even if your job can be done with glue logic, pushing for a deeper understanding of the components and doing better when you can is what keeps the industry as a whole moving forward, not just the bleeding edge.

I think this is pretty spot on. I'd like to think many of us are similar to home builders. We have all the standard supplies/tools at our disposal. The blueprints are quite similar to other homes because that's what our users expect, something familiar.

"They were self sufficient back then. Today we call that poverty"