Editorial standards
Every piece on Agent Mode AI carries an ID, a review date, and a verdict you can check. This page is the editorial charter — the rules the publication holds itself to, and how corrections work.
The review cycle
Every published argument is reviewed on a 30–90 day rhythm. The cadence is set when the piece goes out, based on how fast the underlying landscape moves. Security-advisory pieces run 30 days; governance-pattern pieces run 60–90. The next review date is visible on the piece and on the Holding-up index.
Three verdicts
Every tracked claim has one of three states at any moment:
- Holding. Re-checked on cadence; the argument still stands. No changes required.
- Partial. Most of the argument holds; one substantive part has been revised. The correction is dated and appended to the piece.
- Not holding. Falsified or overtaken. The piece stays up, annotated. The claim text is kept as originally published; the correction explains what broke.
Corrections
Corrections append; they do not silently rewrite. When a claim moves to Partial or Not holding, the correction log at the foot of the article grows by one entry — date, kind, note. The original claim text stays. If you read a piece today that has been partially revised, you see both what was published and what changed.
This is the point. The value of writing about enterprise AI is to be right for longer than a news cycle. The only way to prove that is to show the working.
Nothing is quietly removed
Articles are never deleted. URLs are never silently redirected to hide old work. If a piece goes to Not holding, it stays up and the Holding-up index shows why. If an article is eventually superseded by a rewrite, both pieces remain readable and the supersede relationship is explicit.
Sourcing
Every statistic links to its primary source — inline, not "industry research says". Named case studies link to the company's announcement, the analyst report, or named press coverage. Quotes are real, attributable, and linkable. If a claim cannot be traced to a verifiable primary source, it does not appear.
Our own analysis is allowed — and labelled as such. "Based on the evidence above, the realistic number looks closer to 15%" is an honest claim when the evidence is cited. "Industry analysts report 70%" without a citation is not.
Disclosure
Every article is written and curated by Claude, then reviewed, approved, and published by Peter. This is disclosed in the footer of every page and in the byline affordance. When an article discusses a vendor Peter has a commercial relationship with, that relationship is disclosed inline.
Views on this site are our own. Where an article draws on operational detail from any organisation, it is either public record or has been generalised enough that no confidential material is disclosed.
Subscribe
The newsletter is a weekly piece of original analysis — not a recap of what was published here. One or two sends a week. One-click unsubscribe. No tracking beyond open/click aggregate. Your address is not shared.