Design Systems
Scaling a Design System Without Losing Consistency
Every new feature stopped looking like it belonged to the same product as the team grew.
Token layer
The baseline inventory — color, type, spacing — necessary but insufficient on its own to keep a growing product consistent.
Pattern guidelines
The layer above tokens: documented rules for when a component applies, what it should never do, and how it relates to its neighbors.
Global pattern map
The system reframed as a set of patterns rather than a set of screens, so most new requests resolve to an existing rule instead of a new design problem.
AI-readable rules
The same guidelines written precisely enough for an AI tool to apply them correctly when generating a new screen, not just for a human designer to interpret.
Challenge
The system told designers what pieces existed, but not how they behaved together. Consistency ran on people remembering precedent, and it broke down as the product and the team grew.
Process
I rewrote every meaningful component and pattern to capture not just its variants, but when it applies, what breaks it, and how it relates to the patterns around it. Globally, not ticket by ticket. It took months, not the few weeks I'd planned.
Research
What I noticed going through the backlog was that most new design requests were a pattern we already had. There were plenty of components. What nobody had written down was the judgment about when to use which.
Design
Guidelines became the source of truth, and any single screen became one expression of them. Solid enough that other designers, engineers, and AI tools could apply the same judgment without working it out from scratch each time.
Outcome
Design reviews shifted from catching inconsistency to answering genuinely new questions. New designers and AI-generated drafts alike produced work that fit the system without first absorbing years of unwritten team knowledge.