Large language models (LLMs) are revolutionizing how we summarize text, mimicking a human's iterative refinement process. But what's the best way to guide them? Researchers explored two prompting strategies: "prompt chaining," which uses a sequence of separate prompts for drafting, critiquing, and refining a summary, and "stepwise prompting," which combines all these stages within a single prompt. Intuitively, the single prompt approach seems more efficient, but does it actually work better? The surprising finding: prompt chaining consistently produced superior summaries. Why? It appears stepwise prompting might create a sort of artificial refinement process, where the LLM introduces errors just to correct them later. This intriguing discovery has implications beyond summarization, offering valuable insights into how we can better harness the power of LLMs for various text generation tasks. While stepwise prompting might seem like a shortcut, this research suggests that a more structured, chained approach leads to more meaningful and accurate results. The future of prompt engineering may lie in finding the right balance between conciseness and allowing LLMs the space to truly refine their output.