I am consistently in awe of how well mathematicians have stacked layers of abstraction one on top of the other, and how many different ideas end up being very related to one another. Maybe I am romanticised it and the fact that I regret not going for pure math, but there is beauty in all those abstractions that fit together so nicely.
Defining better abstractions is part of the process that got us so far. Even in ML, we are starting to define some very good abstractions for Neural Networks through the perspective of symmetries and geometry.
I completely agree we should be defining better abstractions! I more meant it gets a lot harder to do so near the cutting edge.
It's unfortunate that such issues exist! But considering the issues within academia, you are most likely in a better situation now that otherwise with respect to free time and salary.