Design systems
B2B
Atlas
A design system and component library that unified a sprawling enterprise data platform across four product teams.
Role
Senior Product Designer
Timeline
8 months · 2022
Team
3 designers, platform eng
Platform
Web app
Overview
Atlas is an enterprise analytics platform that grew through acquisitions and fast shipping. Four product teams had built four visual languages — 60+ screens with inconsistent buttons, tables, and spacing — which slowed every team and quietly confused customers.
I led the creation of the shared design system, from audit through adoption, alongside two designers and platform engineers.
The problem
Four teams, four design languages, one confused product.
The same action looked different on every screen. Engineers rebuilt the same components again and again, and customers felt the inconsistency as friction — even when they couldn't name it.
The audit
I inventoried every screen and interviewed engineers from all four teams. Three findings shaped the approach:
01
Twelve button styles
The inventory found a dozen button variants all doing the same job in slightly different ways.
02
Tables were the real pain
Dense data tables were rebuilt per team, each with different sorting, density, and behavior.
03
Tokens before components
Teams needed shared color, type, and spacing tokens before any component would actually stick.
The messy middle
I started with one screen, not the whole system.
Designing a system in the abstract fails. I rebuilt a single high-traffic screen using proposed tokens and components, then let that real example prove the model and pull other teams in.
A — Big-bang library
Rejected: no adoption
B — Token-first
Promising: foundation
C — Per-team kits
Rejected: re-fragments
D — Pilot screen
Shipped basis
What I rejected
A full library shipped at once would have sat unused. Starting from one real screen gave teams proof and momentum they could see.
Key decisions
Three decisions, each with a real tradeoff.
Ship tokens before components
Shared color, type, spacing, and elevation tokens landed first, giving every component a common foundation.
Tradeoff
Tokens alone don't look like progress. I paired the rollout with one visibly upgraded screen to show value.
Make the data table the flagship
We invested most in the table — the platform's hardest and most-used component.
Tradeoff
It delayed simpler components. But solving the hardest one built real trust in the system.
Measure adoption, not completeness
Success was screens migrated to the system, not the number of components designed.
Tradeoff
The library stayed smaller for longer. It also actually got used.
The solution
Atlas is now the shared foundation for the platform: tokens, a documented component library, and migration guides teams actually follow.
Data table
Tokens & theming
Impact
60+
Screens migrated onto the system
2M
Users on Atlas-powered surfaces
−40%
UI build time for new features
1
Shared design language across four teams
What I'd do differently
I'd have invested in documentation earlier. Great components don't adopt themselves — the docs and migration guides turned out to be the real product.
Next project