Experiments in Flight
What I'm actively building, testing and sometimes abandoning. Each entry records the finding, not just the finished thing — the failures are usually the more useful half.
3 entries
One repository, many agents, independent deploys — path-filtered pipelines over a shared base image, so editing one agent doesn't rebuild the rest.
Measuring where the six seconds actually go on a scale-to-zero agent container, and which mitigations are worth their monthly cost.
Regression testing for prompts — a graded golden set built from real production queries, with per-category scoring so an aggregate average can't hide a broken edge case.