"So new languages with different abstractions come along and things are simpler to do in it. That simplicity is then used to build even more complex things, which then expose how the abstractions doesn't work in the formerly new language."

This is true for mathematics, but not in the same way for computer languages, because we haven't unlimited computing power and memory space.

Yes, with a more abstract language you can solve a problem easier on the implementation, source code side.

But with every abstraction you're loosing flexibility, because to get an abstraction you're putting the lower parts in a concrete combination. Thus every abstraction has it's costs.

Yes, you could use two computing languages, one for the performace/memory critical part, and one for the other parts of the programm. But in practise it's not that easy to combine the two.

There's a reason why people are still using C/C++.