Design Ops
Scaling Design Review Without Scaling the Team
Getting design sign-off stopped being the thing that delayed every release.
Agent roster
Narrowly scoped agents — component usage, content and tone, accessibility, layout audit — each drawing from a defined slice of the design system.
Request routing
A single incoming request, such as a new dashboard page, broken down and distributed across the relevant agents automatically.
Synthesized review
Four separate agent findings reconciled into one coherent report, instead of arriving as disconnected outputs someone has to manually cross-reference.
Continuous audit
A full-product design system audit running on a regular cadence, catching drift early instead of surfacing it in a rare, painful cleanup.
Challenge
Every design system question, which pattern applies, is this the right component, queued behind whoever held the context to answer it. The system was fine. A person still had to retrieve and apply it every time, and that person was usually me.
Process
I made each agent small on purpose. One only knows component usage, one only content and tone, one accessibility, one layout. A coordinating layer splits a request between them and puts the findings back together. The first version was one big agent, and it mixed everything up: accessibility notes inside layout feedback, tone comments on components. Nobody could act on it.
Research
What I learned quickly was that the agents were only as good as the skills behind them. The orchestration was useless without the deep component documentation it drew from, so the two had to grow up together.
Design
A new page or flow now gets checked against component usage, content guidelines, and layout rules in parallel, and the results come back as one report instead of four.
Outcome
Design QA went from a queue behind a senior designer to an immediate first pass, and human review is kept for genuinely ambiguous calls. I read everything it produces before it goes anywhere.