Your agent ships code. Now it ships taste.
Give Claude Code a design director.
Something happened to AI-made interfaces. They all look the same.
-
01
We audit your product
We score your key flows the way the director will, and agree on what “good” means for your product.
You get
- A baseline scorecard of your key flows
- 10–15 test briefs from your roadmap, run on your current setup
- A ranked list of what to fix first
Your partA kick-off call and access to the product. Strategy workshop optional.
-
02
We set the direction
We turn the audit into a design guide your AI can follow: type, colour, spacing, motion and voice.
You get
- A design guide and tokens built for your product
- Good screens to copy and labelled bad ones to avoid
- A one-page cheat sheet of the design terms that matter for your product
Your partOne review session. You sign off the direction before we install anything.
-
03
We install the director
Everything goes into your repo as one pull request. Then we run the same test briefs again and compare.
You get
- The director connected to Cursor or Claude Code
- Components themed to your tokens
- Before and after results on the same briefs
Your partReview and merge one pull request.
-
04 · Optional
We build with you
You keep prompting the way you always do. The director checks the work, and one of our designers reviews what ships.
You get
- A scored fix list on every request
- Designer review on what ships
- Rules tightened whenever something slips through
Your partKeep building. Ask in your own words.
-
05 · Optional
You run it, we stay close
The setup is yours to use. We keep it sharp as your product grows.
You get
- Help when something feels off
- Test briefs run again when models or tools change
- Your guide updated as you add features
Your partTell us when something doesn’t look right.
- Websites
Same screen. Same AI. Now with a director.
A bet slip from our dogfood project, before and after one crit loop.