From the workbench

The blog

Notes on why AI-built screens look off, and what we're doing about it: audits, component-design decisions, and whatever else comes up building OnSystem in the open.

I audited 11 real apps built with AI coding tools. Almost all of them broke the same way.

Eleven real, publicly-submitted apps from v0's community gallery (six internal-tools dashboards, five fintech apps) audited screen by screen. The same handful of failures showed up in nearly all of them: decorative color, numbers that don't reconcile, happy-path-only states, fake-live data. That repetition is the actual product spec.