I tell it the function name, input, output, and an LLM writes the function for me. I prefer to keep an eye on the architectural decisions myself.