Field notes, build notes, and teaching notes on AI systems.https://benfaircloth.com/https://benfaircloth.com/fieldnotes/004-explainable-ai-is-not-the-end-state/https://benfaircloth.com/fieldnotes/004-explainable-ai-is-not-the-end-state/Understanding why an AI system behaves as it does is important. The ability to systematically improve that behavior is what ultimately matters.Fri, 23 Oct 2026 00:00:00 GMThttps://benfaircloth.com/buildnotes/bn-002-the-agent-that-had-to-ask-first/https://benfaircloth.com/buildnotes/bn-002-the-agent-that-had-to-ask-first/What started as a household agent became an experiment in whether models can learn where their authority ends. Supervised adaptation improved escalation under familiar conditions, but the behavior transferred poorly under structural variation.Mon, 19 Oct 2026 00:00:00 GMThttps://benfaircloth.com/teaching/tn-002-from-measurements-to-decisions/https://benfaircloth.com/teaching/tn-002-from-measurements-to-decisions/Evidence does not come with a stopping point. A system needs a requirement for deciding when what it knows is enough to act. Teaching Note 002.Fri, 09 Oct 2026 00:00:00 GMThttps://benfaircloth.com/fieldnotes/003-build-your-party/https://benfaircloth.com/fieldnotes/003-build-your-party/An aggregate evaluation score is a projection. It collapses a model's capability profile into a number that makes comparison possible and loses the shape that deployment actually cares about.Wed, 07 Oct 2026 00:00:00 GMThttps://benfaircloth.com/fieldnotes/002-the-benchmark-ends-at-deployment/https://benfaircloth.com/fieldnotes/002-the-benchmark-ends-at-deployment/A benchmark can tell you what a model can do under evaluation conditions. Deployment asks whether the system built around it works under yours.Fri, 02 Oct 2026 00:00:00 GMThttps://benfaircloth.com/teaching/tn-001-constraint-state-decision-response/https://benfaircloth.com/teaching/tn-001-constraint-state-decision-response/A practical model for reasoning about applications built with language models. Teaching Note 001.Wed, 30 Sep 2026 00:00:00 GMThttps://benfaircloth.com/fieldnotes/001-small-is-not-open/https://benfaircloth.com/fieldnotes/001-small-is-not-open/Model capability, openness, operational control, and evidence answer different questions. Treating them as a single continuum leads to poor AI architecture decisions.Tue, 29 Sep 2026 00:00:00 GMThttps://benfaircloth.com/buildnotes/bn-001-separating-measurement-from-interpretation/https://benfaircloth.com/buildnotes/bn-001-separating-measurement-from-interpretation/Most AI coaching demos ask a model to look at something and tell you what it sees. xCycleForm enforces a harder boundary: the model interprets evidence but never manufactures it.Tue, 29 Sep 2026 00:00:00 GMT