adelphoï labs | AI Training & Automation

The Skill That Changes Everything

If you take one thing from this entire course series, let it be this module. Prompt engineering is not a gimmick or a buzzword. It is the practical skill that determines whether AI is a frustrating toy or a transformative tool in your hands.

A good prompt is the difference between "write me a report" (useless output) and getting a polished, specific, actionable document that saves you two hours of work. The same AI, the same model, the same subscription -- completely different results based solely on how you ask.

The CRAFT Framework

We use CRAFT because it is easy to remember and covers everything you need:

C -- Context: What is the situation? What background does the AI need?

R -- Role: Who should the AI act as? An expert, a teacher, a critic, a colleague?

A -- Audience: Who is the output for? A CEO, a customer, a 5th grader, your team?

F -- Format: What should the output look like? Bullet points, email, report, table, code?

T -- Tone: How should it sound? Professional, casual, urgent, empathetic, authoritative?

From Bad to Exceptional: A Real Example

Attempt 1 (No framework):

"Write me an email about our new service."

Result: Generic, vague, unusable. The AI has no idea what service, who it is for, or what tone to use.

Attempt 2 (Adding Context):

"Write an email announcing our new AI-powered bookkeeping review service for small business clients."

Result: Better -- it knows what the service is and who the audience is. But still generic.

Attempt 3 (Full CRAFT):

"Context: Our accounting firm (Johnson & Associates, 50 employees, serving small businesses) is launching an AI-powered monthly bookkeeping review that catches errors 3x faster than manual review. It costs clients $200/month and saves them an average of 8 hours. Role: You are our client communications manager. Audience: Existing small business clients who already use our tax preparation services but not our bookkeeping services. Format: Email, 200-250 words. Include subject line, brief intro referencing our existing relationship, 3 bullet points on benefits, and a CTA for a free trial month. Tone: Professional but warm. These are people we have worked with for years."

Result: A polished, specific, ready-to-send email that sounds like it came from someone who knows the business.

Advanced Prompt Techniques

1. Chain of Thought

Ask the AI to think through its reasoning step by step before giving a final answer. This dramatically improves accuracy for complex tasks.

"Before answering, think through this step by step. What are the key factors to consider? What are the potential approaches? What are the tradeoffs? Then give me your recommendation."

2. Few-Shot Learning

Give the AI examples of what you want before asking it to produce.

"Here are three examples of client testimonial summaries in our style: Example 1: [your example] Example 2: [your example] Example 3: [your example] Now write a similar summary for this testimonial: [new testimonial]"

3. Iterative Refinement

Do not try to get the perfect output in one prompt. Use follow-ups:

"Make the tone more conversational"
"Shorten the second paragraph by half"
"Add a specific example about inventory management"
"Rewrite the opening to lead with the cost savings statistic"

Think of it as a conversation with a very fast, very patient assistant.

4. Constraint Setting

Tell the AI what NOT to do. This is often as important as telling it what to do.

"Do not use jargon. Do not include disclaimers. Do not start with 'In today's fast-paced world'. Do not exceed 300 words. Do not use bullet points for this section."

5. Output Templating

Give the AI a structure to fill in:

"Format your response using this exact structure: SUMMARY: [2 sentences] KEY FINDINGS: [3-5 bullet points] RECOMMENDATION: [1 paragraph] NEXT STEPS: [numbered list] RISKS: [brief list]"

Common Prompt Mistakes

Being too vague: "Help me with marketing" vs. "Draft 3 LinkedIn post variations promoting our April webinar on AI for accountants"
Not specifying format: You get a 500-word essay when you needed 5 bullet points
Forgetting the audience: Getting a technical explanation when you needed something for non-technical stakeholders
Not iterating: Giving up after one mediocre output instead of refining
Overloading: Putting 10 tasks in one prompt instead of breaking them into focused requests

Your Deliverable: The Prompt Portfolio

Choose 3 tasks from your opportunity list.
For each task, write a full CRAFT prompt in the Sandbox.
Iterate at least 3 times on each one, refining based on the output you get.
Save your best version of each. These are the beginnings of your personal prompt library -- reusable prompts you can deploy whenever you need them.

By the end of this module, you should have 3 production-quality prompts that save you real time on real tasks.