Prompt Engineering Fundamentals

What you'll be able to do

By the end you will be able to take a task you currently do by hand — a report, a set of emails, a piece of research — and write a prompt that produces usable output on the first attempt rather than the fifth. You will know why a prompt failed instead of guessing, and you will be able to fix it deliberately.

Concretely: the five-part framework applied to your own work, few-shot examples that actually steer output, chain-of-thought where it helps and where it wastes tokens, structured output that parses reliably, and self-grading rubrics that let a model check its own work before you see it.

Who it's for

Anyone who uses AI tools regularly and suspects they are getting less out of them than other people are. No technical background is assumed — there is no code in this course, and the examples are drawn from writing, analysis and everyday professional work rather than engineering.

It is also the right starting point for developers. The techniques are the same regardless of whether you are typing into a chat box or an API call, and skipping the fundamentals is the most common reason engineering teams end up with prompts nobody can debug.

Who it isn't for

If you are already writing structured prompts with explicit constraints and getting consistent output, this will be revision. Start at Loop & Agentic Engineering instead.

It is also not a course about a specific tool. If you want to learn a particular product's interface, the product's own documentation will serve you better.

What you'll need

Access to any AI chat tool — ChatGPT, Claude, Gemini, or an open model through our Playground. A free tier is enough. Nothing needs installing and no API key is required.

How it's taught

Every lesson gives you a prompt to run rather than a concept to remember. The prompts have been tested against the models they name, and the ones that fail on particular models are marked as such — a technique that works on one flagship and breaks on a smaller model tells you something useful about what your prompt was assuming.

Expect to spend as long practising as reading. Working through this in one sitting is possible and mostly wasted; the value is in running the prompts against your own work and seeing where they break.