[immibis]

Low level "machine sympathetic" programming is just one type. If you restrict yourself to it, you'll be no better than the programmer who's restricted to any other type. All these abstractions didn't spring out of nowhere - in many cases they make for very flexible software that is cheap to create. Go learn both ends. And the middle. And the weird outliers.

[7thaccount]

Agreed. I stated above that abstraction hell has its benefits. At the end of the day though you mainly are just slapping components together with that and not understanding how the problem truly works.

In my industry we do a lot of optimization and the standard technique is to not reinvent the wheel and to use a solver (e.g. Gurobi) where you give the silver your decision variables and constraints and have it quickly return the optimal decisions. This is great, but costs a lot of money (the best solvers are almost entirely commercial) and you treat a core part of your program entirely like a black box and you also have to fight things like the license manager and so on. Another option is not to create your own generic solver that solves all problems, but rather a solver that solves just your problem. The one company I know that does this has been enormously successful and their product is incredibly fast and lightweight with zero dependencies besides Windows (really cool). The only negative is that now the application is tightly connected to his own solver and they can't just switch it out easily with a different one like most people can. It's all a set of tradeoffs at the end of the day.