Which is prompt engineering, since you just ask the LLM for a good context for the next prompt.