I think of LLMs as essentially translators - taking natural language and translating it into something else. It works great with writing HTML for example. The more declarative and high-level the language is, the better it does. Which makes intuitive sense, the closer the output is to the input the better it does imo.

So generally the people getting the most use out of LLMs are people who are using these higher levels of abstractions. And I imagine we will be building more abstractions like HTML to get more use out of it.