← Back to Learnings
Rookie10 min read

Prompt Engineering 101

Why Prompt Engineering Matters

The prompt is your primary interface with an LLM. A well-crafted prompt can be the difference between a useless response and a production-quality output. As a PM building AI products, prompt engineering is one of the highest-leverage skills you can develop.

Core Techniques

1. System Prompts

System prompts set the context, personality, and constraints for the model. Think of it as the "job description" for your AI.

You are a senior product analyst. You analyze feature requests
and provide structured recommendations. Always consider:
- User impact (who benefits, how many)
- Engineering effort (t-shirt sizing)
- Strategic alignment (company goals)
Respond in markdown with clear headers.

Best practices:

  • Be specific about the role and expertise level
  • Define output format upfront
  • Include constraints and guardrails
  • Keep it concise — long system prompts can dilute focus

2. Few-Shot Examples

Provide examples of the input-output pattern you want. This is one of the most reliable ways to steer model behavior.

Classify these feature requests:

Input: "Add dark mode to the dashboard"
Category: UI/UX
Priority: Medium

Input: "Fix login timeout on mobile"
Category: Bug
Priority: High

Input: "Add bulk export for reports"
Category: ???
Priority: ???

3. Chain-of-Thought (CoT)

Ask the model to think step by step. This dramatically improves reasoning quality for complex tasks.

Think through this step by step:
1. First, identify the core problem
2. Then, list possible solutions
3. Evaluate each solution's tradeoffs
4. Recommend the best approach with reasoning

4. Structured Output

Tell the model exactly what format you want. JSON output is especially useful for building products.

Respond with a JSON object containing:
{
  "summary": "one-line summary",
  "sentiment": "positive | negative | neutral",
  "action_items": ["list", "of", "items"],
  "confidence": 0.0 to 1.0
}

Common Pitfalls

  1. Being too vague — "Summarize this" vs "Provide a 3-bullet executive summary focusing on revenue impact"
  2. Prompt stuffing — cramming too many instructions degrades performance
  3. Ignoring output format — always specify how you want the response structured
  4. Not testing edge cases — your prompt should handle weird inputs gracefully

Key Takeaways

  • Iterate relentlessly — prompt engineering is empirical, not theoretical
  • Version your prompts — track changes like code
  • Test with diverse inputs — what works for one example may fail for others
  • Combine techniques — system prompt + few-shot + CoT is a powerful combination