Anyone else feel like prompt engineering is mostly trial and error?
retained text
Sometimes a prompt gives exactly the answer you're looking for, and then a small change in wording completely throws everything off. I've been thinking about how much effort goes into getting consistent results. There's so much discussion around prompt engineering, AI development, and teams like geekyants, but the actual challenge seems to be making things work reliably beyond a few successful tests. I'm curious what people here focus on more. Do you spend time refining the prompt itself, or have things like examples, structured outputs, and testing made a bigger difference?"
document fetched 2026-11-12 00:41 · canonical https://reddit.com/r/PromptEngineering/comments/1xtgkhj/anyone_else_feel_like_prompt_engineering_is