People and models excel at understanding and responding to succinct, clear instructions without the need for extensive guidance. Keep prompts simple and direct. Unfold as you go.
Since people and models perform reasoning quite well, prompting either to "think step by step" is unnecessary. Avoid chain-of-thought prompts and instead, you (the human) must provide step-by-step adaptations as new information emerges.
By starting small, refining in steps (adding layers of specificity), preserving coherence (evaluating each time), and embracing emergence, we can expect better results. Delegation/prompting is not a one-shot-and-run kind of thing.