Free Course · 5 lessons · ~3 hours · The highest-leverage course in our library
The Art of Prompt Engineering
Writing prompts is the most consequential AI skill most people will ever learn. The same model produces mediocre results for a bad prompt and excellent results for a good one. This course is about closing that gap — for any tool, any task, any model.
Who this is for
You should take this course if you use AI chatbots — Claude, ChatGPT, Gemini, Copilot, or any of their cousins — and have noticed that the quality of the answer depends enormously on how you ask the question. You may be a complete beginner. You may have been using AI tools daily for two years. Either way, the principles and patterns in this course will give you a meaningful step-change in what you get out of these tools.
You do not need any technical background. The course works as well for a teacher and a marketer as it does for a software engineer. We will use Claude.ai or ChatGPT (free tiers are enough) for hands-on exercises, but every principle transfers to whichever tool you use.
Why this course exists
Prompt engineering is the one AI skill that compounds across every tool you will ever use. Someone who reads our other seven courses but cannot write a clear prompt will get less from each of them. Someone with good prompting instincts will get more from every AI tool, every future model, every assistant they have not yet met. The course has the broadest payoff per hour invested of anything in our free library.
What you will learn
What a prompt actually is
How a language model reads what you type. The three layers most people don't know exist (system prompt, conversation history, current turn). Why "just say it clearly" is not the whole answer.
The five principles of a good prompt
Specificity. Context. Role. Format. Constraints. Before-and-after examples for each, applied to a single recurring task we improve throughout the lesson.
The patterns that work
Few-shot examples. Chain-of-thought. Structured output (JSON, tables, lists). Decomposition into steps. Negative examples. When each one helps — and when it backfires.
Iteration and evaluation
What to do when the output is wrong. How to A/B test two prompts. Building a small personal prompt library you can reuse. The 80/20 of prompt debugging.
From prompts to systems
System prompts (the constitution of any custom bot). Long-context prompts (when you have a whole document). Prompts for agents that take actions. The next step.
What you will need
- About three hours of total time. Each lesson is 25 to 45 minutes.
- A browser, and a free-tier account with either Claude.ai or ChatGPT. The exercises will work with either.
- A notebook — paper or digital. The course works much better when you write your prompts and their results down as you go. Pattern recognition is a written practice.
What this course is not
This course is not a list of "prompt templates that will 10x your productivity". It is also not a treatise on the mathematical inner workings of language models. It sits in the middle: the principles a thoughtful adult can apply to any task, with enough understanding of what is happening under the hood to know why the principles work.
The templates you find online are useful for the specific task they were designed for. The principles you build in this course are useful for the next thousand tasks you have not yet encountered.
How this fits with the full programme
This is a cross-cutting course — it sits between Tier 1 — The Foundation (where critical thinking and AI history live) and Tier 2 — Core AI (where the deeper material on generative AI, large language models, and human-AI interaction lives). Every learner in the Integrated AI Program takes this material in more depth across multiple modules. The full programme launches with the December 2026 cohort.
You can finish this free course without ever applying. We hope it is genuinely useful regardless of what you do next.