These kind of articles show a deep misunderstanding about how computers work. The only code ever run is machine code burned into the chip. Humans write programming languages that are compiled into that code, and so do LLMs.
Suggesting that using an LLM to write a programming language which is then compiled is the same as using "English to program" is skipping steps. No one does that. We use English to generate program code in a traditional language. That's how the LLM was trained and that's what it outputs.
Generating code this way, later to be compiled and run, does turn out to be faster than generating such code the other way and that's why it's popular.
