This is a very silly argument: compile and runtime speed is not the only consideration to take into account.

Otherwise we would all be programming assembly, or heck, just write machine code directly and skip the time-consuming assembler.

Speed is important, but so is good debugger support, productive UI tools, developer productivity, training requirements, available libraries, target market, etc. At the end of the day, most people make pragmatic decisions, taking lots of factors into account.

A good historical example of this is Word vs. WordPerfect. There are many reasons why Word won, but one of them is that WordPerfect was written in assembly. This was helpful at the beginning as it was good for performance, but their developer productivity suffered as the codebase increased. This resulted in fewer features, which is also something that customers care about.

Few customers care how quickly you can do something they don't want to do.

Also, he's conflating toy problems, where getting the big-O right is the only important thing, with real-world applications where the constants are also important.