Beyond the Basics: Build Prompt Systems That Work

🧭 THIS WEEK AT AI SECOND ACT

Howdy, this is the fourth edition, and I’d love your help shaping it. You’ll find a quick feedback poll at the end or a quick reply works too. Was a sad week in the life of this little newsletter when we had 1 unsubscribe. Hurting.

👉 Just hit "Reply" and let me know what you want more (or less) of. My goal is to make this as valuable and practical as possible as we navigate the new AI era. 🚀

• Expanding on edition 2, we dive deeper into some key prompting techniques. There are some crazy complex prompting manuals and infographics out there! We’ll try to keep it simple®.

  • TL/DR? → Scroll down to ‘The Master’

• 20 prompt pack, as practical as possible [10 here, 20 total on the website].

🧰 AI NEWS + LEARNING

Here’s a couple of things I found this week:

• Anthropic released a new version of Claude!

• Claude prompt engineering guide

• Check out Codex from OpenAI - Wow!

🗺️ FEATURED INSIGHT

Two weeks ago, we covered the 80/20 fundamentals: roles, structure, context, constraints, and iteration. If you've been practicing those basics, you're probably getting decent results.

When Basic Prompting Hits Limits

• Your AI gives surface-level analysis instead of deep insights

• Complex multi-step problems get an oversimplified or incomplete answer

• Responses lack the nuance you need

The Solution: 3R 

Progressive prompting techniques that match how these models were trained to reason, 3R Method:

1. Roles – Give the AI a clear identity.

• "You’re a leadership coach. Give feedback on this email I wrote to my team."

• "Pretend you're an SAT tutor. Quiz me on this text with 3 multiple-choice questions."

2. Routines – Build multi-step chains to guide it, which means break it down!

• "Act as a tech analyst. Read this article and summarize it in 5 bullet points."

• "Now summarize it again for a busy exec, in < 100 words."

3. Reviews – Check, compare, and iterate for accuracy.

• "Give me 2 alternative answers with pros/cons."

• "Review this answer for clarity and bias."

• Give here a very detailed description of/ example of the output expected, that is what is a great response.

Key Techniques

Expanding on 3R, use the following techniques:

• Specificity and Context:

  • Concept: Always be clear, precise, and descriptive in your instructions. Provide all relevant background information and details.

• Zero Prompt:

  • Concept: Very simple prompts, no example given.

• Chain-of-Thought (CoT) Prompting:

  • Concept: Guide the AI to break down complex problems into logical, sequential steps, similar to human reasoning.

• Few-Shot Prompting:

  • Concept: Provide the AI with one or more examples of the desired input-output pattern within the prompt.

• Prompt Chaining:

  • Concept: Break down highly complex tasks into a series of smaller, simpler prompts, where the output of one prompt becomes the input for the next.

• Tree-of-Thought (ToT) Prompting:

  • Concept: An extension of CoT, where the AI explores multiple reasoning paths simultaneously, evaluates each option and selects the most promising one.

The Master

Become the master by taking it further:

• The Professional Review Protocol: 'Review this as a senior partner at McKinsey. What would make this analysis more rigorous and actionable?'

• The Analysis → Synthesis → Critique Chain: Use Claude for deep analysis, GPT-4 for creative solutions, and then Claude again for risk assessment. Each model brings different strengths and solutions.

• The Research → Strategy → Communication Pipeline
  Perplexity for market research, Claude for strategic analysis, GPT-4 for stakeholder communication. Match the tool to the thinking style needed.

10 Prompt Pack

OK, so here’s a starter pack to reuse, iterate and improve.

Zero-Shot Prompting

These prompts require the AI to generate a response based on its general knowledge and instructions, without explicit examples or step-by-step guidance within the prompt.

1. Project Director: Write an email to the project team reminding them about the weekly sync meeting tomorrow at 10 AM CST in Conference Room A, asking them to come prepared with updates on their current tasks.

2. Project Director: Generate a concise summary of the project's current status for our monthly stakeholder report, highlighting that we are on track for Phase 1 completion but noting a minor dependency on the QA team.

3. Engineering Manager: List three clear, actionable goals for our engineering team for the upcoming sprint, focusing on code quality and reducing technical debt.

4. Engineering Manager: Draft a brief Slack announcement for my team about a new code review process, emphasizing its benefits for quality and knowledge sharing, and stating that all pull requests now require two approvals.

5. Product Director: Identify three common pain points for users when onboarding to new enterprise software platforms.

Chain-of-Thought (CoT) Prompting

These prompts encourage the AI to perform multi-step reasoning, break down a problem, or follow a logical sequence to arrive at the answer, often involving analysis, outlining, or structured development.

6. Project Director: Create a draft agenda for a 60-minute project kickoff meeting for our new 'Atlas' initiative. Include sections for project overview, team introductions, scope review, key milestones, and next steps.

7. Project Director: Outline a risk mitigation plan for the 'Orion' project, specifically addressing potential delays due to third-party vendor integration issues. Suggest contingency actions and a communication strategy for stakeholders.

8. Project Director: Develop a presentation outline for a quarterly business review (QBR) focusing on the 'Gemini' program. Include sections for overall program health, key achievements, upcoming challenges, budget summary, and strategic next steps. The target audience is executive leadership.

9. Engineering Manager: Summarize the key differences between microservices architecture and monolithic architecture, including the pros and cons of a large enterprise system. Focus on scalability and maintenance.

10. Engineering Manager: Generate a template for a post-mortem analysis document for a critical production incident. Include sections for incident timeline, root cause, impact assessment, remediation steps, and preventive measures.

Clearly, with some of these examples, you would need to give the model more information and background to get reasonable responses.

There are 10 more on the AI Second Act site!

👉 Hit “Reply” and share your experience — I read every one!

+10 Prompt Pack

Meta Prompting

Having the AI generate or refine prompts for other AIs (or itself).

11. Manager/Director: "Generate a prompt for an AI to summarize a 50-page technical specification document into a one-page executive brief for a non-technical VP."

12. Manager/Director: "Refine the following prompt to ensure the AI generates a project status update that is concise, focuses on risks, and highlights upcoming milestones: 'Give me a project update.'"

13. Manager/Director: "Create two distinct prompts for an AI: one to generate a list of potential solutions for our software's scalability issue, and another to evaluate each solution based on implementation cost and predicted performance gain."

Tree-of-Thought (ToT) Prompting

Guiding the AI to explore multiple reasoning paths, evaluate each option, and select the most promising one, often by explicitly asking for exploration and justification.

14. Product Director: "Given the current market shift towards cloud-native solutions, propose three different strategic approaches our company could take to adapt. For each approach, outline potential benefits and drawbacks, and then recommend the best path forward with justification."

15. Engineering Manager: "Our engineering team is facing a performance bottleneck in our core API. Brainstorm at least three distinct technical solutions to address this. For each solution, evaluate its technical complexity, estimated development time, and potential impact on system stability. Conclude by recommending the optimal solution and explaining your choice."

16. General Enterprise Role: "We need to improve cross-functional collaboration between Product, Engineering, and Marketing. Suggest 2-3 different organizational or procedural changes that could achieve this. For each change, analyze the potential benefits, challenges, and required resources. Then, justify which change you believe offers the best return on investment for our company."

17. The Stakeholder Alignment Strategy (Product Director) "Address our feature conflicts by exploring different alignment approaches: Vector 1: Data-driven approach (user analytics, A/B tests, usage patterns) Vector 2: Strategic approach (business goals, competitive positioning, long-term vision)
    Vector 3: Collaborative approach (stakeholder workshops, compromise solutions, shared ownership) Fully develop each vector's pros and cons, then identify the optimal combination for our situation."

Prompt Chaining

Breaking down a complex task into a series of smaller prompts, where the output of one prompt becomes the input for the next.

18. Chain Example 1:

• Project Director: "Summarize the key decisions and action items from the attached meeting transcript."

• Project Director: "From the summary generated in the previous step, extract all action items, assignees, and deadlines, then format them into a bulleted list."

• Project Director: "Draft a concise follow-up email to the team based on the bulleted list of action items, reminding assignees of their responsibilities and deadlines."

19. Chain Example 2:

• Product Director: "Analyze the recent customer support ticket data (imagine a data source attached) to identify the top two recurring themes regarding product usability."

• Product Director: "Based on the themes identified in the previous step, propose three distinct feature enhancements for our product that would address these usability issues."

• Product Director: "For the most impactful feature enhancement proposed, draft a user story including acceptance criteria."

For complex initiatives that need both strategic thinking and tactical execution:

ReAct for Implementation Planning

20. The Project Recovery Planner (Project Director) "Guide me through getting this delayed project back on track using reasoning-action cycles: Reasoning: What are the real causes of our 3-week delay beyond the obvious ones? Action: Create an honest assessment of the remaining work and a realistic timeline Reasoning: Which stakeholders are most impacted and what are their primary concerns? Action: Design a communication strategy that rebuilds confidence Reasoning: What scope adjustments could recover time without killing project value? Action: Develop 3 scenario plans with different scope/timeline trade-offs Continue until we have a comprehensive recovery strategy."

Again with this last set of examples, you’ll need to add additional background information in some cases to get reasonable responses.

Photo by Alexander Shatov on Unsplash