Nowadays software is orders of magnitude more complex with layers and layers of abstraction and containerization to make app development and deployment as quick, easy, reliable and secure as possible, to expedite time to market and profit, that it's become unrealistic and unprofitable to expect your devs to be proficient across the whole stack from cloud, web, to kernel and hardware.
The vast majority of average dev jobs nowadays are a lot more specialized and containerized, and companies mostly care about your ability to grind through the stack of CRUD Jira tickets to solve the business problems the company needs to make money, not dig through the Linux codebase to find some bug/race condition in the kernel/libraries.
Outside of embedded/kernel dev and big-tech, where low level bug fixes and optimizations can translate to real $ at scale, jobs that need low level knowledge or that reward it, are very niche.
Many things became a lot easier with the resource abundance like having flat memory model with the 32bit systems in a stark contrast via the segment switch on 8086.
In the late 70s and 80s Assembler, Pascal, C, Fortran, Modula-2, Forth, Lisp were all used to create CAD, GIS, robot control, symbolic math, abstract algebra, etc. type programs.