Not faster.
Finally possible.

Every design system accumulates work everyone knows needs doing — but the time can never be justified against the roadmap. So it gets deferred. Workarounds get built. The debt grows quietly.

At Pushpay: components that shipped without documentation. Variable structures that needed auditing across 17,000 tokens. A library cross-referenced against Storybook and GitHub to find inconsistencies. Weeks of focused effort from someone who had neither the weeks nor the focus.

AI didn't make this work faster. It made it possible for the first time — by replacing hours that were never design decisions to begin with.

The key isn't any single tool — it's connecting them. Claude becomes the connective tissue: context stays alive across tools, and outputs feed directly into the next step.

Some of this work got faster. The more interesting category is work that wasn't happening at all — because the cost could never be justified against other priorities.

The approach isn't to use AI for everything — it's to use it precisely where human judgment isn't the bottleneck. Test it. Validate it. Run it once a human is no longer needed in the loop.

Design decisions still require a designer. What components exist, how they should behave, what the token architecture should mean — these aren't AI questions. What AI replaces is execution overhead: migration, documentation formatting, audit transcription, repetitive find-and-replace work that fills hours without requiring thought.