Break down tasks: workflow thinking
Day 4 of 30 · 30 Days of AI
Guide the model step by step
Learning goal
- Split complex tasks into phases.
- Use stepwise prompts to reduce errors.
Why it matters
- One-shot prompts for complex tasks often fail.
- Phases improve control and quality.
Explanation
- Decompose: input → process → output → review.
- Ask for outline first, then details.
- Insert QA checks (assumptions, missing data, risks).
Examples
- One-shot (weak): “Write a full campaign plan.”
- Phased (good): “Step 1: give a 5-bullet outline. Step 2: expand each bullet to 80 words. Step 3: add risks and metrics.”
Guided exercise (10–15 min)
- Choose a complex task (e.g., campaign, report, SOP).
- Write a 3-step prompt (outline → expand → QA).
- Run it and note improvements vs one-shot.
Independent exercise (5–10 min)
Apply the pattern to a different task (e.g., meeting summary → action plan).
Self-check
- Task split into steps.
- QA step included.
- Output quality better than one-shot.
Optional deepening
- Prompt chaining examples: https://platform.openai.com/docs/guides/prompt-engineering