> And then you try to write a JIT for a dynamic (and concurrent, let's say, as Erlang succeeded in this in the last release) language and, with just "arithmetics, pointers, functions and structs", it becomes obvious it's not nearly enough. Then you start reading books and papers and soon you realize that 5 years have just passed. And JITs are not really that complex!

I actually wrote a JIT for one of Googles internal machine learning frameworks, so yes. It was used in production for Google search, Google ads etc. Only a very small part of that traffic, of course, I didn't do something huge, but at least I know what it takes to make such systems production ready and performant enough, as these systems are extreme resource hogs. So I know very well what it takes to create a language, parse it, execute it, optimize it and ensure that there are no errors in production and help the researchers who are supposed to use it to debug when errors happen.

I learned programming by just implementing a lot of complex algorithms and systems. I wrote several high throughput http servers using sockets and threads for example. I also got good enough at competitive programming to occasionally place in single digits in some world programming competitions, but mostly double digits. All of that is very easy to understand compared to math, math just forces you to bend your mind in strange ways, programming is super concrete.

I have no doubts that programming will one day get as deep and mindbending as math, physics definitely is for example, but today it isn't even close.