Field notes
Field notes
Each field note reads the public record of a real AI rollout through one friendly lens: cost per accepted change. The goal isn't to second-guess the teams involved — they shipped at real scale under real pressure, and we've all learned from what they shared in the open. It's to show where the metrics everyone reaches for measure velocity, and where one more number can help measure value kept. We're all figuring this out together; this is one tool that makes the figuring easier.
The Copilot productivity paradox, and the number that resolves it
GitHub measured a 55% speed-up. Independent studies measured 41% more bugs, rising code churn, and experienced developers running 19% slower while feeling 20% faster. Every one of those findings is a different face of the same missing number.
Uber measures almost everything — here's the one number even they don't roll up to
Uber is one of the most thoroughly instrumented engineering orgs in public — and refreshingly honest that the AI value "link is not there yet." The one dial even great instrumentation doesn't quite produce.
Klarna, AI customer service, and the resolutions that stayed
Klarna leaned hard into AI customer service, then thoughtfully adjusted course when quality slipped. The generalized metric — cost per accepted resolution — is the part the deflection-rate dashboard couldn’t show.
AI-first mandates, and the number that would tell you they're working
A run of "AI-first" mandates announced on adoption and headcount metrics — and IBM's quieter, wiser exception. What a kept-value denominator adds to each.
Want a company analyzed, or have public data that sharpens one of these? Open an issue on GitHub.