notes
writeups from the experiments.
-
Sep 10, 2026
shipping eval loops that fit in a lunch break
what we tried to keep agent regressions visible without building a platform first.
-
Sep 5, 2026
treat prompt changes like code
diffs, named versions, and rollbacks beat vibes when agents start to drift.
-
Aug 28, 2026
lab notebook over demo theater
why we write down what we tried — even when the experiment dies quietly.