Imagine a world where documenting code is as easy as chatting with an AI. That's the promise of large language models (LLMs). But can developers actually harness their power effectively? A new study investigated how well developers, both professionals and students, could prompt LLMs to generate useful code documentation. The researchers built two VS Code extensions powered by GPT-4. One allowed developers to freely write their own prompts (ad-hoc prompting), while the other used a pre-defined, optimized prompt (few-shot prompting). The results? Professionals, with their deeper Python knowledge, were more successful at ad-hoc prompting, often by simply including keywords like "Docstring." However, students struggled without the guidance of a pre-defined prompt, often producing lengthy, less readable explanations. Overall, both groups preferred the simplicity and efficiency of the few-shot prompting tool. It generated concise, readable documentation in a consistent format. However, developers in both groups saw the generated documentation as a starting point, emphasizing the iterative nature of documentation. This research has important implications for AI-powered tools. It suggests that developers need more support in prompt engineering, either through better interfaces or educational resources. Additionally, generating multiple documentation versions at once could spark developer creativity. The study also highlights the importance of human-centered metrics when evaluating code documentation quality, as automated measures often miss the nuances of what makes documentation useful for developers. This points to a future where AI assists, but doesn’t replace, the developer's role in creating clear, concise, and helpful documentation.