You can ship an interface now. That's the problem.
Building a working product screen used to take a designer, a frontend developer and a week. Now you describe it and it exists in ninety seconds. It runs. It has a sidebar, a table, a modal, states you didn’t ask for. It looks like software.
And most of the time it’s wrong in a way that’s genuinely hard to see. Not broken, not ugly — wrong. The total at the top of the dashboard doesn’t match the rows underneath it. The landing page opens with a sentence that would fit four thousand other companies. The payment form asks for a card number before it says what you’re buying. Nobody will file a bug for any of it. The screen just quietly fails to do its job.
This course is about seeing that, and fixing it.
The model fills gaps with the average.
Ask for a dashboard and it invents plausible numbers. Ask for a landing page and it invents plausible content — the gradient, the three icon cards, elevate your workflow, a testimonial from a person who doesn’t exist. It isn’t being lazy. You asked a question, it answered the question, and everything you left unspecified was filled in with the most likely thing.
The average is not a design. It’s the absence of one.
So the skill this course teaches isn’t drawing. It’s specification — knowing which decisions matter, making them on purpose, and being able to look at generated output and say precisely what’s wrong with it. That was always what separated designers from decorators. It’s now the whole job for everyone else too.
Who this is for
If you’re not a designer
You prompt, you get a screen, something feels off, and you can’t name it. You don’t need Figma and you don’t need design theory — you need vocabulary for the problems and a habit of reviewing. Every unit opens with the symptom you’d actually notice, and no term appears before you’ve felt the thing it names.
If you are a designer
You already know why the screen is wrong. What’s new is the medium: you’re not drawing screens any more, you’re writing the constraints that generate them, then reviewing output at a volume no design process was built for. Parts 1 to 5 will be partly familiar. Parts 6 and 7 won’t be.
Both of you are doing the same thing: getting a machine that is very good at plausible to produce something correct.
How it's built
Three layers, used differently.
Lessons
One idea each, in order: a symptom, two versions of the same screen, a short explanation, a prompt you can copy, and a rule to keep. Each takes about ninety seconds. Read them front to back, once.
Reference
One short entry per component: what it’s for, when to use it instead of the one next to it, which states it needs. It answers which one do I use, not how do I build one. Don’t read it — search it.
Recipes
Whole patterns, where the sequence is the knowledge. A payment form isn’t hard because inputs are hard. It’s hard because the order, the trust signals and the decline path are load-bearing, and no component library will tell you that.
Rules
Every unit ends with one line. Those lines are the actual product of this course.
Collect them. Put them in the rules file your AI tool reads — AGENTS.md, CLAUDE.md, .cursorrules, whatever yours is called — and they apply to every screen you generate from then on, without you having to remember them. The rules index has all of them on one page, ready to copy.
A prompt fixes one screen. A rule fixes every screen after it. That difference is most of the value here.
Before you dive in
You need none of this to start. But if a unit refers to something you’ve never met, one of these will fill the gap faster than a lesson can. Read all of it, or none of it — the units don’t depend on it.
If you’ve never studied design at all
If you’re already a designer
Tools worth keeping open
A contrast checker — generated colour pairs fail contrast constantly, and it’s the one design error with a legal dimension. A type scale generator, so your sizes are a system rather than six guesses. And your browser’s device toolbar: most generated layouts break at 390px wide and nobody looks.
How to use this
Bring a real screen — yours, or one you generate as you read. Every unit ends with a checklist you run against your own work, and that’s where the learning happens, not in the reading.
Read Part 1 in order; it’s short and everything else assumes it. After that, go to Part 2 if you build marketing sites, or Part 3 if you build product screens. Parts 6 and 7 are worth reading whatever you build.
The course is published as it’s written, so some units are marked planned. Those pages still carry the idea and the rule, and you can ask to be told when the full unit lands.
One note on language: this is written in English only, and there’s no French edition planned. The prompts and rules are meant to be pasted exactly as written, and translating them would change what they do.
Add up the numbers on your own dashboard