How do you know an ai-generated interface is actually good?
Learn to see what's off in an ai-generated interface, name it, and fix it. Built for people shipping real product with ai tools.
Taught by Ryan Cuff
Building an interface that looks good is still hard.
The generating part is solved. Describe a screen, it exists four seconds later. Most of the time it's fine.
Then sometimes it isn't. Something in it is off, and you can't say what. No word, no reason. Just the sense that someone with taste would wince at this.
So you prompt around it. Make it cleaner. Make it more modern. You get back a different screen with a different thing wrong. Now you're negotiating with a model that has no idea what you're pointing at. Neither do you.
You run out of words before you run out of problems. So you settle, and you ship it, and you know.
It doesn't look right and you can't say why.
The left button was built with our skill file, so it gets real press feedback, a dedicated CSS class, and a pointer cursor. The right button looks the same but has none of that. That gap is what the course teaches you to close.
What if you could name it?
If you knew what a good interface is made of, you'd see it fast: which element is too heavy, which spacing is doing nothing, which two things are competing for the same attention.
You could stop typing make it cleaner and start typing the third heading is the same weight as the first, drop it a step. A model can act on that. It cannot act on cleaner.
You'd stop guessing whether something is off. You'd know what is off, and you'd have the words to hand it over.
That's a skill. In short supply, and in high demand.
The course where you learn to see it
Untrained Eye teaches you to read an interface the way an editor reads a draft. You learn the vocabulary, you practice on real screens, and you leave with a set of instructions that stop your AI from making the same mistake twice.
Every foundation comes with a small coding exercise, plain CSS, so you're adjusting real values and watching what changes, not just reading about it.
No theory for its own sake. Every idea arrives attached to a screen you're looking at.
The left toggle was built with our skill file. It only animates when you actually click it, and it settles with a slight overshoot instead of stopping dead. The right one snaps instantly and never gives you anything to feel. Same switch, same colors, different amount of care.
The loop
Everything in the course runs through the same three steps, until it stops being steps and starts being reflex.
Slow down in front of the screen you just generated. Find the thing your eye keeps snagging on before you touch a prompt. Most people skip this part and go straight to the fix, which is exactly why the fix never sticks. The look is the whole game.
Hierarchy, spacing, contrast, motion, restraint. Five words that turn a vague feeling into a defect you can point at. Once you can name what's wrong, you're not describing a feeling anymore. You're describing a fact the model can actually act on.
Rebuild the piece by hand, then encode the fix as a reusable AI instruction. You fix it once, it stays fixed. The next time the model makes the same mistake, you already have the words ready, and you hand them over instead of starting from scratch.
What you get
Three layers, each built on the last. You'll move from seeing to diagnosing to directing, until you can trust your own judgment.
Module 1
See it
The vocabulary. Hierarchy, spacing, contrast, motion basics, the words you've been missing.
You'll learn to read a screen the way an editor reads a sentence, spotting exactly which element is doing too much and which one isn't doing enough.
Each concept comes with a short coding exercise, plain CSS, so it turns into muscle memory instead of just vocabulary.
Module 2
Diagnose it
Real interface teardowns. Before and after, every decision narrated out loud.
You'll watch the same screen go from generic to considered, one small, deliberate call at a time, until the pattern starts to feel obvious instead of mysterious.
Every teardown ends with the actual code change that fixed it, so the diagnosis turns into a diff, not just a description.
Module 3
Direct it
How to train your own eye and direct AI like a design partner, not a vending machine.
By the end you're not asking the model to guess what you mean. You're telling it exactly what to change and why, and you can defend the call.
You'll practice turning a diagnosis into a written instruction an AI can implement, then checking its code against what you actually meant.
Who this is for
You build real product with AI tools, and something about what it hands back feels off before you can say why.
You'll start with the basics: hierarchy, spacing, contrast, motion, the same vocabulary designers spend their first year on. That's what lets you actually see a screen instead of just reacting to it.
From there you learn to diagnose real interfaces and direct the model instead of negotiating with it.
Hey, I'm Ryan
Three years of product engineering, most of it as an engineer at Genesis Systems, and the last stretch spent learning design and animation in public. Badly at first, then less badly. I'm also the creator of loomui.design.
No design degree, no theory to defend. What I have is the path from this looks wrong and I don't know why, to naming the defect and handing the model a fix. I'm teaching it the way I learned it: direct, concrete, in front of real interfaces.
One step ahead of where you are, which is the distance that makes it useful.
Reserve your spot
Sign up now and lock in the launch price, $79. I'll email you a private link when the course opens.
- See it, diagnose it, direct it
- Real interface teardowns
- Reusable AI skill files
- Direct email access to me
- Every future update, included
- 14-day money-back guarantee
- Lifetime access
- Built by someone one step ahead
Course
Engineers, mainly. If you've got some React and CSS under you and you're shipping real interfaces with AI tools, this is built for you, no design background required.
Not yet. If enough people want one, that's the next thing I build.
Short, narrated teardowns and written breakdowns you work through at your own pace, alongside real screens, not slides.
Yes. It's being built in public right now, and everyone who buys in gets every update as it ships.
No. The code you write is plain CSS and React, just enough to prove the diagnosis, not a library tutorial. Bring whatever stack you already use.
No. The proof is the skill file you leave with, the thing that actually changes how your AI builds for you.
Purchasing
No. One payment, $79 at the launch price, covers every update as the course grows.
Not yet, it's an individual license for now. Email me if you need several.
No, it's included in the $139.
Email me within 14 days and I'll refund you, no questions asked.
Yes, email me and I'll send one.
Not yet. If the price is the barrier, email me directly.
Not yet, but email me if you're a student, happy to work something out.
Help
Yes. Email me directly, I read everything during the pilot.
Email me and I'll answer directly.