TLDR the article says that nodes are just a "frontend" and the real work is done by backend which is data-oriented anyway. tensorflow has python frontend and t

TLDR the article says that nodes are just a "frontend" and the real work is done by backend which is data-oriented anyway. tensorflow has python frontend and that is "slow" too, yet it doesn't matter because most of the work is done in C++ and python just compiles that graph to something lower-level