A natural way to use an LLM for coding is to create tests and ask it to write code that passes those tests, so I disagree with the author.