Practice · Zocdoc · 2021–2024
As a design-systems engineer at Zocdoc, I owned the header migration under a company-wide accessibility mandate. What made it land was product judgment, not more code: change the size of the reviewable unit, coordinate everyone who depends on it, and let evidence confirm each step. The same move as the Klarna call—three years earlier, and in code.
Zocdoc was working toward a fully WCAG-compliant site. Mezzanine—the TypeScript/React design system many teams built their products on—was the mechanism: fix a component once, and every product surface consuming it inherits the fix. The header was the highest-reach instance of that leverage: present on every page, on a legacy API that had to be replaced rather than wrapped, and consumed across four to five independently owned codebases.
Partway through, the feedback was to push more code to production. But the components were built and correct—they sat in review, in branches that drifted while product teams kept editing the codebases underneath them. The constraint was not how much I produced; it was how large and illegible each unit of work had become by the time someone had to review it.
Blocked by review, not by code. Changing the shape of the work—not the amount—is what moved it.
Identify engineering ownership, bring stakeholders in early, and line up QA ahead of each change—so every dependent team advances together. I audited the legacy header and then read how downstream teams actually called it, because real usage is what would break.
The design-system team's first frontend-component experiment: a gradual rollout, measured across browsers and mobile, with test/control analysis. A load-bearing change, released like a product rather than shipped in one cut.
A measured, confident migration the dependent teams advanced on together.
Smaller, well-scoped tickets made the work legible and moved it faster.
A PR merge template that changed how the team shipped, not just my own tickets.
Coordinate everyone who depends on it, size the unit of change so it stays reviewable, and let the evidence confirm the change is good. That is the same instinct as the Klarna decision—name the boundary a strong headline hides, and move only as far as the evidence supports. Three years earlier, and in code.