Engineering intelligence, quantified.
Essays on checking AI against the source, factbooks, and the shape of AI-native engineering. Long enough to matter; short enough to read on a Tuesday.
More essays from the build.
Claude Told Me a Family Member Had Kidney Disease. They Don't.
A frontier model reconciled years of a relative's medical reports — genuinely useful, and it made five mistakes that would matter clinically. What caught each one wasn't a smarter model; it was checking every claim against its source. De-identified. Not medical advice.
Read the case study →Facts that bind at write time
Session-start reminders decay. Nora v2.8.1 binds directive MUST-NOT factlets at the write — pre-write blocking plus a post-output secret scanner. Small-N field evidence; Factlet RFCs merged as optional vendor extensions.
Read the essay →Inside the Nora Dashboard: Where AI Sessions Become Engineering Intelligence
A deep dive into the AI Leverage Score, session analysis, playbook extraction, anti-pattern detection, and sprint-level engineering metrics — with illustrative recreations of our live dashboard.
Read the essay →Kiro + Nora: Get Running in 5 Minutes
One-click install in Kiro. Your agent captures each session, extracts what worked and what broke, and grounds the next prompt automatically — zero config.
Read the guide →New pieces when there's something real to report.
No inbox, no schedule. Follow the repo to know the moment something new ships.