Evidence
Our evidence standard.
PathPatron teaches leaders to evaluate AI claims critically, so our own claims should be easy to audit. Every briefing in the AI-Enabled Teams & Agent Ecosystems series — and increasingly everything we publish — stands on one of four kinds of evidence, and the prose tells you which.
Published evidence
Cited, with its maturity stated in the sentence, not a footnote. A proof of concept is called a proof of concept. A pilot is called a pilot. A company's own engineering blog is called self-reported, because it is. We do not cite vendor case studies about their own customers, advertorials, or aggregator blogs.
Our own practice
PathPatron itself is run with AI agents — gated changesets, tiered approval, verification checklists, and dated revision notes when we correct something. Where this series draws on that experience, the artefacts exist and can be shown, failures included.
Professional observation
Passages marked "in my work" draw on two decades inside enterprise transformation. They are always patterns seen across many organisations, never one firm's story in disguise, and they carry recognition rather than proof: their job is that you have been in that meeting too.
Prescription
Where the field has no published evidence yet — and for agent teams outside the software industry, as of September 2026, it largely does not — we say so and author the standard openly, in normative language, versioned and dated. The Agent Team Blueprint ships as v1.0 with the sections most likely to be revised marked as such. When practice proves a part of it wrong, the revision is dated and explained, the same way we correct everything else.
Where the evidence comes from next
Tell us what broke.
If you run any of this in your own organisation, tell us what broke. Anonymised reports from practice are how the missing evidence gets built.