A Design System People Trust

We had a design system. It had just stopped being the source of truth. Developers shipped small, ad-hoc fixes, each one simple and harmless on its own, and over time those little decisions drifted out of sync until nobody trusted the library anymore. The big-ticket widgets held up fine. The small stuff fell apart, the bits and pieces the original system had never really accounted for. I found the shape of it by auditing what we already had. Our entire library lived on one sprawling board, hard to parse, disorganized, missing the very components the site ran on every day. Everything we needed sat crammed into a single screenshot. The evidence pointed past another patch, toward a new foundation.

• What drifted: the disclosure control, every label (filter pills, search-result labels, campus and location labels, the accepting-patients label, the distance indicator), the single card block, the condition-search result card, plus drop shadows, strokes and borders, accents, and text highlighting.
• What held: the large widgets stayed consistent, provider cards, location cards, CTA blocks, and stories.
• The gap: we needed a system comprehensive enough to govern the small pieces of the site, not just the big-ticket widgets.

The University of Rochester Medical Center runs as a 30,000-person academic health system, serving 1.2 million patients across the region. When the system drifts and the two teams stop speaking the same language, the cost lands on patients.

Three steps, two of them mine and one I drove with engineering. The first was executive buy-in, and I made that case in the terms leadership weighs: money and time. Every week, teams burned hours reconciling work built from different sources of truth, with a standing two-hour sync running every other week just to catch the drift. Framed as efficiency and cost, the rebuild got its green light.

1. 01

   Problem Framing

   I owned

   I started with the developers rather than the components. They showed me why the old library had failed: it covered the big widgets but never the small pieces teams reached for, so trust in it eroded. One comprehensive source of truth would fix that; a longer component list wouldn't.

2. 02

   Token Architecture

   I owned

   I rebuilt the foundation and defined the full token set, from type and color through grid, spacing, iconography, animation, borders, and shadows. Drift stops at the root now: the small pieces inherit from one source instead of getting hand-styled on every page.

3. 03

   Engineering Partnership

   Collaboration

   I built every component alongside the developers, never across a handoff wall. When something couldn't be built cleanly, the design gave way. That habit earned the system enough trust to be adopted, and it cut review time once work reached handoff.

Most design systems document the big components and stop there. This one documents every widget the same way, down to the small pieces that used to drift, so developers and content editors can move without pulling each other off real work. Each widget answers the same six questions. That consistency turned the library from a liability into a tool the team reached for on its own, and it shifted reviews from "I don't like the blue" to "does this serve what the patient needs?" The payoff showed up in the review queue. Once developers and content editors had what they needed up front, their comments on the design docs dropped 40 to 50 percent, a number I got by counting the threads on the Figma files before and after, alongside the meeting hours we won back. The questions that remained were sharper, builds moved faster, and my best estimate is that a typical component went from spec to merged in about a third less calendar time.

Once the core library had earned trust, I extended it to the medical school as a bespoke pattern library, one that sharpened student storytelling to draw applications while holding clinical brand coherence across very different audiences. The same foundation now serves patient-facing care and academic recruitment alike.