I'm glad that this process does happen (if slowly) in mainstream language design. IMH (and casual) opinion, up through the 80s mainstream programming languages were super ad hoc, but (again, IMHO) from the late 90s formalisms from computer science started to filter into the mainstream. I think this is in part because of the crisis of crappy code and a flood of new developers (before computers were so important, code quality was often not that important) so there was more impetus for "steering away from the bad path".
So for example CLU seemed like a purely academic (and restrictive) project to me back in the 80s, but ultimately spurred a lot of currently mainstream concepts like iterators and strong type systems. It just took 30 years (as did Liskov's well-deserved Turing). Lisp was scorned for its GC, anonymous lambda functions, dynamic type system, huge standard library and many other things, many of which are now table stakes today, 50 years later.
By the way, again IMHO, we lose something when language evolution increases guard rails. The Lisp environments of the 70s and 80s allowed lots of people to manufacture a rope and then hang themselves. But for others they were enormously powerful (e.g. I wrote the base of Cyc, a "blank sheet" reimplementation, in just a few weeks). The Smalltalk environment was like that too: infinitely malleable, designed for people who wanted to explore, and powerful enough to easily kill your machine. Nowadays languages with great expressiveness (e.g. C++, especially if you act as if its long ago "parent", C, doesn't exist) draw complaints that it's "too complicated" and "too hard".