Hamel Husain argues that copying prompt recipes doesn't work. Instead, analyze your failures: examine many examples, find where things break, then write clear instructions. Managing an LLM is like managing an employee. It takes practice and iteration, not templates.
I think prompt recipes are overhyped. In most cases, I don't think you should reach for a prompt recipe. What you should do is do data analysis, look at lots of examples, and understand where your system is failing, and use that to provide clear instructions to your LLM. Think of it as an employee. You're managing employee, you want to give clear instructions, and you want to refine that over time. I think the best way to think about it is practice. When you reach for a recipe, what ends up happening is you don't think about the prompt. You're just borrowing it and cargo-culting it over. Don't do that. It's really important that you go through the process of thinking about the prompt.
