AI Prompt Cheatsheet

Principle 1: Stop using AI as a search engine

Most people type a vague question and hope for the best: "Write me a lesson about photosynthesis." That's treating AI like Google. Instead, describe your specific situation, your constraints, and the exact output you need.

The difference between "fix my code" and "I'm a backend developer working on a Python REST API with PostgreSQL. My auth middleware throws a 500 on expired tokens. Here's the stack trace. Suggest a fix that doesn't invalidate existing sessions" is the difference between garbage and gold.

Principle 2: Define the role and experience level

Start every prompt by telling the AI who it is: the role, the years of experience, and the domain. The higher the experience level you set, the deeper the AI will reason.

"You are a project management expert with 20 years of experience in IT, hands-on with Scrum and Agile."

Think of AI as a sparring partner: it can be a junior, a peer, a senior, or a subject matter expert. You decide which by how you frame the role.

Principle 3: Tell it your tool stack

After defining the role, list the tools you actually use. This prevents the AI from suggesting random alternatives you'll never adopt.

"I work with Jira, Trello, Confluence, and Slack."

Now every recommendation fits your existing workflow instead of requiring you to learn something new.

Principle 4: Plan first, then proceed

Don't hit enter after writing the role and context. Add this line:

"Ask me 2-3 clarifying questions, then generate the plan."

This forces the AI to understand your situation before producing output. In the workshop, this single step turned generic advice into a tailored career plan in under 3 minutes.

Know your models

Not all AI models are equal. Pick the right one for the task:

Opus

The Architect

Planning, creativity, reasoning, complex strategy

Sonnet

The Developer

Coding, writing, well-known patterns, daily tasks

Haiku

The Junior

Quick lookups, basic text, fast simple tasks

Use Opus for planning, then hand the plan to Sonnet to execute. Same project, less tokens burned.

Templates from the Workshop

Each template below was built live during Workshop 1 with an attendee's real problem.

Template 01 Project Manager

The AI Mentor for Your Career

Live demo: Abhijit wanted an AI mentor for project management activities

You are a project management expert with 20 years of experience in IT, hands-on with Scrum and Agile. My mentor for specific IT projects to accelerate my process. I work with [Jira, Trello, Confluence, Slack]. Ask me 2-3 clarifying questions about my current situation and goals, then provide a structured mentorship plan.

The "ask me 2-3 clarifying questions" line is the secret. It forces the AI to gather context before jumping to advice. Without it, you get generic PM tips. With it, you get a plan tailored to your specific projects.

Template 02 Teacher

Lesson Plan for What Students Actually Struggle With

Live demo: "Write me a lesson about photosynthesis" vs. the focused version

I teach [8th grade biology]. My students struggle with [light-dependent reactions in photosynthesis]. Create a [30-minute] lesson plan with a hands-on activity that helps them understand [the specific concept]. Include: learning objective, materials list, step-by-step activity, and 3 assessment questions I can use to check understanding at the end.

The original prompt was "write me a lesson about photosynthesis." Adding the specific struggle point ("light-dependent reactions") and format constraints (30 min, hands-on activity) changed the output from a Wikipedia summary to a usable lesson plan.

Template 03 Content Creator

Research Workflow That Cuts 4 Hours to 90 Minutes

From the workshop: "How creatives can go beyond their current capabilities"

I write [2 blog posts per week] for a [SaaS company] about [technical topics]. I currently spend [4 hours on research] per post. Create a research workflow that cuts that to [90 minutes] using AI. Include specific prompt templates for: 1. Topic research and trend identification 2. Outline generation from multiple sources 3. Fact-checking and source verification Format: Step-by-step with the exact prompts I should paste at each stage.

The "specific prompt templates for each stage" constraint means you get a reusable system, not a one-off answer. Run this once, use it every week.

Template 04 Product / Business

Professional Presentation from Technical Architecture

Live demo: Emil needed a McKinsey-style deck explaining a transcription tool

You are a UI/UX designer with extensive experience creating professional presentations and translating technical information into business terminology. I need a [McKinsey-style] presentation for [business stakeholders]. The topic: [describe your system, tool, or project]. Here's the technical context: [paste your architecture, stack, or flow]. Core message: [what you want them to remember]. Ask me 2 questions about the audience and key message, then generate the slide-by-slide outline with talking points for each slide.

Defining the style ("McKinsey-style") and audience ("business stakeholders") prevents the AI from creating a technical deep-dive when you need an executive summary. The 2-question constraint ensures it gets the nuance right.

Template 05 Data Analyst / Career Changer

Data Portfolio Plan from Scratch

Live demo: A data analyst building a portfolio to attract senior roles

You are my career mentor with 25 years of experience building people's careers from scratch. I have [2 months] to put together [4 data analysis projects] with varied datasets and build a portfolio website to showcase them. I can spend [2 hours per day] with AI assistance. I have access to [Claude Pro]. My target: [senior IC role at a tech company + freelance pipeline]. Create a week-by-week plan that: (1) Picks projects that show range and depth, (2) Builds the portfolio site incrementally as projects complete, (3) Includes specific prompts I should use with AI at each stage.

Adding "specific prompts I should use at each stage" means the AI generates a meta-plan: a plan that includes its own prompts. This is how you compound AI assistance across a multi-week project.

Template 06 Developer

Context Setting for Large Codebases

From the workshop: "Dawning different hats, scanning large codebases, setting up context again and again"

You are a senior [language/framework] developer familiar with [your architecture: microservices, monolith, etc.]. I need to [add a feature / fix a bug / refactor module] in [module/system name]. Here's the relevant code context: [paste the key files, function signatures, or error logs] The current behavior is [what happens now]. The expected behavior is [what should happen]. Analyze the code, identify the root cause, and propose a fix with: (1) What's wrong, (2) The minimal change needed, (3) Any risks or edge cases I should test.

For large codebases, use Claude Code (the terminal tool) instead of the web UI. It reads your entire project context automatically, eliminating the need to paste files manually.

Template 07 Everyone

The Universal Prompt Framework

The core pattern taught in Workshop 1 that underpins all the templates above

You are a [role] with [years] years of experience in [domain]. I am a [your role] working on [your project/situation]. I need you to [specific output: write/analyze/plan/recommend/debug] [exact deliverable]. My constraints: - Tool stack: [tools you use] - Time: [how long you have] - Format: [output format: table, bullet points, code, slide outline] - Tone: [direct, formal, conversational, technical] - Avoid: [what you don't want: generic advice, jargon, vendor names] Ask me [2-3] clarifying questions, then deliver the output.

This is the one template to remember. Every other template in this cheatsheet is this framework applied to a specific role. Fill in the brackets, and you'll get better output 90% of the time compared to a generic prompt.

Get the cheatsheet