I find myself also forgetting that my prompts are more like outcomes than actual instructions. Perhaps that's why overtime people end up vibe coding like that because it just gets easier to expect the llm to walk into the right answer eventually.