# Field notes
Engineering, written down.
How we design, build, and ship production AI on compressed timelines. Practical notes from the pod: models, infrastructure, evals, and the decisions behind them. First principles, measured results, no filler.
# All notes
02Soon
Evals that actually catch regressions before they ship
A custom eval harness on real traffic catches the 14% accuracy drop that a generic benchmark sails right past. How we build them, and why every engagement gets one.
03Soon
Anatomy of a 12-day vertical agent
Day by day, what it actually takes to put a production agent in front of real users in under two weeks, and the shortcuts that are never worth it.