Storybook as the bridge
An LLM reads component stories, not raw code, to pull props and variants and generate matching Figma components. Stories carry design intent.
Loading the case study
Case study 05 · CDK Global · Design Systems
One design system for every CDK product, from CRM to accounting, so design and engineering stop debating which button is correct.
Architect, Product Manager, Promoter, Advocate, Evangelist
6 years
Designers, developers, QA and UX architects; 50+ designers supported
The short version
Decades of acquisitions left CDK with 20+ tech stacks and as many opinions on what a dropdown is. Customers bought one company and experienced several.
The system as a product: roadmap, epics, releases, documentation and support. Also the quieter job of getting 50+ designers to actually use it.
Atomic design, WCAG 2.1 AA from the first component, and documentation tested as hard as the code. Support was staffed, because unsupported frameworks die quietly.
85% of core products rebuilt on the system in two years and 20+ stacks standardized. The UI now looks like one company, and the system is being taught to AI.
The long version · for the detail-inclined
01 — Strategy
Strategy · The Scope
The Scope We had an opportunity to create a design system For a platform that didn’t exist yet. CDK was literally redeveloping ALL applications, by which MOST would leverage this framework. The Challenges: • Had emerging uses cases • Had no real general implementation approach defined • Had evolving styling requirements and theming needed • Previous iterations had given people stability and quality perceptions
Strategy
Underlying CDK technologies data base was stood up using pre SQL technologies built on over the decades from a text based input utilized since the early 90s.
A majority of the main offerings were ON prem and had only begrudgingly made the business decision to design a modern interface and experience.
Strategy ·
Mistakes Were Made The organization initially started with a rigid, un themed, inflexible system based on Google material 1.
Strategy · The Problem Was….
How can you create a absolute standard design for ever evolving uses cases? The Problem Was…. Can there ever be ONE UI / interaction pattern that solves for everything?
Strategy
It was clear that flexibility would need to be factored in while large portions of the requirements were gathered from all product stacks to provide standard design patterns.
02 — Research
Research · A Solid Foundation in UCD Principals
A Solid Foundation in UCD Principals This framework was user centered in design from the very start with a absolute hard requirement to understand the 2 primary users of the RDS framework. The Archetype of the Designer and Developer. Designer Developer Is • Creative • Busy • Not necessarily experienced with design systems • Smart and looking for rapid and versatile creation • Mid to senior level with 5-20 years of experience Needs • Efficiencies in design system • Proper Figma libraries that are them able and can be used in any use case across all products • Needs to be shown how to create standard components • Needs flexibility and reliability in atoms and molecules • Needs • Coaching and problem solving • Proper documentation that is updated regularly • Needs to be shown how to create standard components • Lighting fast response and support initially Is • Varied skill sets across CDK products and acquisition • Not familiar with REACT or enterprise architecture • Self paced learning / may need coaching in problem solving within framework • Minimal experience with design systems • Smart and looking for pre-made examples for versatile creation • Mid to senior level with 5-20 years of experience
Research
This framework was user centered in design from the very start with a absolute hard requirement for WCAG 2.1 aa compatibility for every component, color contrast, font & button etc..
We made absolutely sure that in addition visual things we provided guidance and programatic attention to ARIA STATES , FOCUS ORDER & TAB ORDER
< alt= “description” >
Research
Designers and developers were interviewed as to technologies they relied on to ensure that the design system could be created using the most cutting edge and efficient technologies and that it could be maintained and built on for years to come.
03 — Analysis
Analysis
Analysis
Analysis
Like all best of class design systems
Analysis · Collaborative and Contribution
Collaborative and Contribution Build a composable solution that could be rebuilt by anyone ( almost) • While we were always building we encourage product team to create their own components and we would work with them as to how it would be come part of the official library • Due to the scale and pace of individual dev teams there was no way a small team could do EVERYTHING!
Analysis
We ensure that there were cycles spent on documentation and that the documentation system was ALSO tested and we got all the feedback from the users of the systems.
Analysis
Where frameworks and project dies, is when they are not supported!
Dedicated staff of 2 to ensure technical and design questions and issues were resolved quickly.
04 — Design
Design · Radial Design System Examples
Radial Design System Examples
Design
Support our brand and identity, designed to live side by side
Design
Design
Design
Design
Design
Design
Design
Edit Task
Design
Shared Services: Common Admin
Design
Services Dashboard
Design
Design
Design
Design
✦ — AI-native
Design · The system learns to talk to AI
As LLMs learned to read and write Figma and generate code, my team rebuilt Radial around that new way of working. The React component library became the hub, and every AI workflow, from a designer describing a screen in plain language to production handoff, connects back to it.
An LLM reads component stories, not raw code, to pull props and variants and generate matching Figma components. Stories carry design intent.
Every pull request compares stories against the Figma library and flags mismatches before merge, so design and code cannot quietly diverge.
Designer-controlled props live in Figma, developer defaults are pre-filled in code, and required developer decisions are flagged in Dev Mode. Handoff arrives complete.
AI-generated components go through the same release review for fidelity, accessibility and correctness. Prompts are versioned in git beside the React source.
Exploration stays freeform; production work runs through the system. Each product team gets an adoption path for both modes.
Figma read/write and complex prototypes work well. Organism-level layouts like multi-component navigation are still the stress test.
05 — Results
Results