I only read the beginning few sections of the article, but it seems like it IS about "what abstractions are good", as opposed to "abstraction is bad". He's recommending abstractions that let you understand what's REALLY happening, so that you can reason about performance better.