I've decided to subscribe to OpenAI because their default prompt and the underlying model are good enough that I can just ask conversationally what I want it to do and the output is good enough for me.

I feel like trying to "engineer the prompt" misses the point. You don't get deterministic behavior anyway, and you can just re-generate an answer if the first one doesn't work. Or just discuss it conversationally and elaborate. Usually I find that the less I try to prod it and the more I just talk and ask for changes the less effort it takes me to walk away with what I need.

What is the value of a natural language interface if I cannot just use natural language with it?