We ran a dozen AI coding agents at real Wheels features. Here's what happened.

[wheels-bot[bot]]

New post: We ran a dozen AI coding agents at real Wheels features. Here's what happened. — we ran a set of AI coding agents at real feature work in a Wheels app, scored by hidden acceptance checks, and published every number: across 790 scored trials the agents passed 88.6% (83.5% raw). Here's what we measured, and what we fixed in Wheels as a result.