I think that this kind of novel problem modeling was always in (comparatively) short supply in programming. It's just so _easy_ to tell the machine to do what you want it to do, that it's a tough sell to figure out how to tell it to do something completely different that your modeling says will magically make your solution easier.

It feels like progress here is really jerky, with periods of aimless milling about interspersed with breakthroughs that spread rapidly. Some examples that come readily to mind are A* graph traversal, BNF parser generators, BSP trees for game world rendering.

It seems rare that a working developer gets to sit back and think, "huh, this problem actually feels a bit like that physics/math/statistics thing I was reading about the other day/year/decade" and spend any time usefully thinking about it. So any language or library we have to hand gets bent into solving the problem the way we're thinking about it right now.