Using LLMs effectively isn't about prompting