Six-week structured cohort on agent architecture — planner-executor patterns, memory models, guardrails, human-in-the-loop design. I've been building these patterns in the lab for a year; now I'm pressure-testing what I taught myself against how it's being formalized. Week two. It's tough. That's the point.

Inside the enterprise, the current work is building composable AI skills — real primitives that combine, not recorded macros that replay. The lesson so far: most of what gets called a "skill" is RPA wearing a new badge. Getting the primitives right before adding multi-agent complexity is slower and correct.

Just published: The Dashboard is the QA Layer — what happens to reporting when agents do the work the reports used to coordinate. Several more behind it. One post every couple of weeks is the pace; this is a workbench, not a content operation.

Two autonomous pipelines in production in the home lab: Morning Wire runs its daily 5am triage-and-synthesis loop, and Tailwater & Freestone keeps pulling gauge data and synthesizing fishing conditions across nine rivers. The most useful thing about both at this point is watching what breaks when nobody's looking.