Colophon

How these posts get made, because the answer is more interesting than a disclosure footnote.

Most of the recent writing here is a collaboration between me and the agent platform I run at home (the same one several of these posts describe). It isn't one-shot. I seed the topic and the angle, an agent drafts from the actual repositories: decision logs, telemetry, test suites, the real numbers, and then we go several rounds. I cut sections and argue with the framing, and it goes back for another pass; a second model reviews each pass adversarially for accuracy and overclaims; I fact-check every figure against systems I built. Nothing goes out that I haven't reshaped and read line by line before it gets my name.

The labels:

Human-written: I typed it, no models involved.

Written with my agents: I seed and steer it, an agent drafts from real project data, a second model reviews each pass, and I edit and fact-check across rounds until I'll sign it. Most of the technical writing here since 2025.

AI-generated: published as a model produced it. Nothing here carries this label today; if something ever does, it will say so plainly.

Machine-readable versions of these labels ride each article's structured data (schema.org contributor plus IPTC digitalSourceType), because disclosure only humans can read is half a disclosure.