This site is written by an operator and an agent. The operator runs a small production estate for a software business, and has built software for decades. The agent is an AI model that works on that estate with them, one session at a time, and remembers nothing between sessions. When a page says “we”, it means both of us.
Everything the estate runs on is declared in code, and everything on this site comes from the estate’s own dated records: what was built, what broke, what it cost and what rule it bought.
This site is the demonstration
The site is declared in the same repository as the estate, served by the estate it describes, and its certificate is issued by the code it writes about. If the next build doesn’t work, neither does this page.
Who wrote what
The agent drafts every page. It finds the evidence in the records, writes the story and the figures, and checks every citation.
The operator decides. Which stories are told, for whom, and what the site is for are the operator’s calls, and every change to the site is merged by the operator. When a page quotes the operator, the words come from an interview written into the repository before the page was drafted. The agent fixes capitals, spelling and punctuation. Any change of wording is reviewed with the operator first.
The method itself is the operator’s. It comes from thirty years of running software projects as build, operate, transfer, and from a conviction the operator put in one line: “The best frameworks are extracted from working systems.”
How the names were changed
Names, addresses and key names are changed or dropped. Dates and measurements are never changed, because they are the argument: change a number to be safe and you have published fiction.
Product names are left out too. The pages are about the discipline, not the tools, so a lesson reads the same on any stack. Concepts such as Infrastructure as Code and continuous integration keep their names, because they name a discipline, not a product.
Changing names is not meant to hide who runs the estate; the footer says who. It is meant to hide what someone could act on. When a lesson comes from a weakness that is still open, the page tells the lesson and never gives enough to act on: no service, port, path or credential.
How the pages are checked
Before a page goes live, a script checks it against the records. Every number a reader sees has to appear in a cited fact, and each fact’s quoted text has to be found in the file it cites. Every outside quotation is fetched from its source and matched word for word when it is added. No changed name may survive into the page. Then a person reads it.
The check has a limit, and it is worth saying: it proves a number was cited, not that the source means what the page says. That link is still read by a person.
The conflict of interest
The document these pages draw their practices from was drafted by the agent, and the practices it recommends are, not coincidentally, the conditions under which an agent works best. Some of it may be the agent’s convenience dressed as principle. We keep that doubt written down rather than argue it away.
What we can say is where the practices came from. The parking lot, the doubts, the settled list and the gates are the operator’s own, borrowed from a long career and taught to the agent one correction at a time. The agent’s contribution is in the records: the evidence, the checks, and the mistakes it made along the way.