Deming

You have agents running in your system. Your first instinct was to add controls. Approval gates before they execute. Escalation policies when they’re uncertain. Manifests declaring what they’re allowed to do. Kill switches for when they go wrong. This is the same instinct that produced the approval queue for humans. You are governing agents the way you governed people, because that is the only governance you know. Deming had a name for the knowledge required to manage a system well. He called it the System of Profound Knowledge.1 Four components, never meant as a checklist: what is the system, where does it vary, how do you know what you know, and why do people behave the way they do. ...

Someone approved your last deploy. Maybe it was a thumbs-up on a pull request. Maybe it was a green check from a required reviewer who opened the diff, scrolled for eight seconds, and clicked “Approve.” Maybe it was your own merge after CI passed. Regardless of the ceremony’s size, the question is the same: what did that approval actually verify? Not what it was supposed to verify. What it did verify. What evidence exists that a specific risk was evaluated, a specific judgment was exercised, a specific context was held that the system could not hold on its own. ...