The last 10%

For years, the gap between what I could design and what I could ship was a template. I knew what good looked like. I could draw it, specify it and defend it in a room. What I couldn’t do was build it to that standard myself. Customising someone else’s theme always ended the same way: close enough, never right.

AI closed that gap. My coding skills are not amazing, and they no longer need to be. That is why Craftform exists now, building its own products.

What I didn’t expect was where the work would go. AI didn’t remove the design work. It moved it. Here is what building Restform and its siblings has taught me about where it went.

1. Start from your own gap

Every Craftform product starts with something missing from an app I already use. Restform began when I joined a gym. None of the training apps I tried felt like mine, and on rest days I kept drifting back to journalling apps. One app for the body, another for the mind, and nothing joining them.

Being the first user is not a shortcut around research. It is a starting point I can test against every day. The design work is the step after: putting myself in other people’s shoes and asking whether the fix matters beyond me. Physical and mental health are the same story told on different days. That idea travels further than my gym routine.

2. Use speed to decide, not to finish

Getting an idea from nothing to ninety percent is now fast. The obvious use of that is shipping more. The better use is deciding sooner. The question shifts from “can I build this?” to “is this worth finishing?”, and I can answer it in hours rather than weeks.

That changes where judgement sits. When building is cheap, choosing what deserves the last ten percent becomes the expensive decision.

3. Design the states, not the screen

The last ten percent is where good products separate from bad ones, and it is made of states: focus, hover, empty, loading, animating, the small phone with the keyboard open. AI builds the screen you describe. It does not build the ten states you forgot to describe.

So I stopped describing screens and started describing behaviour. A brief that names every state gets a product. A brief that names a screen gets a mockup that happens to run.

4. Learn to point, and learn to talk about motion

In Figma, I point at a thing and change it. In a prompt, I have to name it, and most of my wasted iterations came down to two gaps.

The first is reference. “The card on the home screen” is ambiguous to an agent looking at a codebase full of cards. Naming the component and the file is the difference between the change landing and a confident edit to the wrong thing.

The second is motion, which is hard to describe because designers and agents don’t yet share a vocabulary for it. “Make it feel smoother” means nothing. What worked for the Craftform mark, which turns as you scroll, was the language of motion design: how it should look, which easing, what triggers it, when it starts and stops.

The words that worked were a designer’s words. That is the most useful thing I’ve learned. The people best placed to direct agents are the ones with the most precise vocabulary for detail, and designers have spent whole careers building that vocabulary. Most of us just never had to write it down.

5. Build for a product, not a demo

A hobby app and a public one look the same in a screenshot. The difference is everything around them. Restform needed 6-digit email verification codes, a consent checkbox because a journal holds mood data, and an honest answer on backups. In launch week, the pace of shipping hit the hosting limit of 100 deploys a day.

The lesson that cost most came earlier. Supabase’s free plan allows two projects, and when you build several products, that ceiling decides your architecture before any user does. I moved to Neon. It is less streamlined, and it works. Now I choose infrastructure for the portfolio, not the project.

None of these is hard on its own. Together, they are the cost of being taken seriously, and they are cheaper to plan for on day one than to discover on launch day.

Where the work went

Put together, the pattern is clear. The making got cheap. Deciding, describing and directing got more important. Running several products, with agents working across many threads at once, is closer to leading a team than designing a screen, and attention is now the scarcest thing I have.

If you are a designer wondering what this means for you: learn to write down what you used to draw. The tools now do the first ninety percent. The last ten percent belongs to whoever can describe it precisely.