Engagespot was shipping faster than its interface could stay consistent — a different button here, a slightly-off pink there. This is how I turned that inconsistency tax into infrastructure.
Audit before you design. Build infrastructure, not decoration. Validate every decision with engineering before it ships.
I audited every live screen and catalogued every button, card, and input already in use — more than anyone expected. Engineering was rebuilding the same components from scratch, onboarding took longer without a shared reference, and several colors failed WCAG contrast outright.
Structure the system atomically — atoms to pages — so it scales without rework every sprint. And run it as a Lean UX loop, validating each foundational decision with engineering before it ships, instead of designing in isolation. Buttons, cards, and templates went first because the audit showed they were rebuilt most often; everything else followed in order of reuse, not completeness.
Notification templates and channel management became full components with real property variants — state, avatar, action buttons — built for both light and dark mode, and validated against how engineering would actually implement them.
A component library that only lives in Figma is still a translation problem — someone has to eyeball a token and type a hex code into the codebase by hand. I connected the entire system to an MCP server so engineers and AI coding agents can query it directly: exact colors, spacing, component states, and accessibility ratings, pulled straight from the source file instead of guessed from a screenshot.
It's the same principle behind how I build products today: treat the design system as structured context an agent can reason over, not a picture to copy. Paired with the component work, it's a real part of why that 30% speed-up holds — one live source of truth, not just tidier files.
Development velocity increased 0%faster after launch. Engineering stopped rebuilding the same button, and handoff got faster with one place to check. This system also became the ground the Notification System Revamp was built on — a parallel project that cut support tickets 85% using these same components.
A system only pays off if people use it — next time I'd pair the library with an adoption plan from day one, not treat adoption as automatic. And I'd audit before building again: every inconsistency caught early was one less thing to reconcile after launch.
Want to talk about design, ideas, tech, coffee, building — or just anything in general? Feel free to hit me up!
Send message