Design Playground
Luxury Presence
A sandbox repo that took designers from never having coded to opening pull requests in the product.
- Role
- Design Engineer
- Duration
- Spring 2026, ongoing
- Skills
- Developer experience, Design systems, AI workflows, Enablement
- Tools
- Claude Code, VS Code, GitHub Desktop, LUX design system
- of designers who had never coded started building in 30 days
- 50%+
- pull requests in under a month, for one designer
- 0 → 19
- pull requests and 265 commits across company repos
- 55
- active designer contributors
- 4 → 7
The problem
Our designers had AI tools and still could not contribute code. Not because they lacked ideas, but because every path in had a catch.
- No code tools abstract too much. You get a screen, not an understanding of what is underneath it, and the result has to be rebuilt before it can ship.
- The real codebase is intimidating. Without a safe place to break things, most people never take the first step.
- Contributing to an existing product is a different skill from prototyping from zero, and nobody had a way to practice it.
The idea: a repo inside the repo
I built Design Playground as one shared repository with a base that is already set up, and a folder for every designer. The base carries the design system, the packages and the company's guardrails. Each personal folder works almost like its own repository: install what you want, build what you want, add your own skills and commands, without touching anyone else's space.
design-playground/├─ shared/ design system, packages, guardrails├─ .claude/ skills and commands everyone gets└─ designers/ ├─ you/ your space: install, build, break ├─ … └─ …
The tooling is deliberately ordinary. VS Code and GitHub Desktop instead of a terminal first workflow, and Claude Code as the pair. The point was never to hide the code. It was to make the real workflow approachable: install dependencies, run the project locally, open a branch, ask for a review. Unlike tools that hand you a finished screen, you understand what happened, and that is what transfers to the product.
What shipped
- The playground itself, with the design system preinstalled so a prototype uses the real components, not look alikes. I review and approve its pull requests.
- A design toolkit plugin on the company's Claude plugin marketplace, with five skills for research planning, call processing, synthesis and a token audit, so the workflows travel beyond the playground.
- Setup guides that keep themselves current, through an automated workflow that updates them when the setup changes.
- It kept growing after launch: a project hub, version history, comments on prototypes and a skill that prepares an engineering handoff.
Results
Within 30 days, more than half of the designers who had never written code were building. One went from zero to 19 pull requests in under a month. By June the team had 265 commits across 55 pull requests in company repositories, four designers had shipped their first production code, and active contributors went from four to seven.
The result I care most about is the change in behavior: designers started opening pull requests in the main product, not only in the sandbox. I presented the work at PM3 Builder Night, an event with more than 550 attendees, and its numbers were part of our internal all hands.
What did not work
The playground is not the answer for everything. When the website builder team needed production grade prototypes, they chose to prototype inside the product repository instead, because anything built in the playground would have to be rebuilt to ship. They were right. The playground is where you learn the workflow; the product repo is where a prototype becomes the product. Making that handoff explicit, prototyping against production code without a rebuild, is now a goal of the design system itself.
Reflection
The hard part of getting designers into code is not the code. It is the first safe step and the second one after it. A guided setup with real freedom on top, real components inside it, and a short path to a real pull request did more than any tool that promised to skip the learning.