The operating model
How this business runs.
Deep Dive Reports publishes research on applying AI inside Testing, Inspection, and Certification firms. It would be a poor advertisement for that if the business itself were run any other way.
This page sets out the operating model, in the same form the reports use. It is here to be checked, not admired. If you are considering the same thing for your own firm, this is what one version of it actually looks like.
The work is divided across four agents, orchestrated by one person. An agent here means what it means in the reports: a scoped process with defined inputs, defined outputs, a stated autonomy limit, and a handoff. Not a member of staff. There is one person in this business, and that does not change.
Exhibit
Operating model, as at 28 August 2026.
| Agent | What it does | What it may settle on its own | What comes to a person |
|---|---|---|---|
| Commercial and go-to-market | Drafts outreach and positioning, prepares account and market notes, keeps the catalogue copy current | Wording, structure, sequencing of drafts | What is sold, at what price, to whom, and every claim made in public |
| Technology and operations | Builds and maintains the site, the delivery gate, the release tooling and the backups | Implementation detail, refactoring, tooling choices | Anything that touches money, personal data, or the paid file |
| Research and product | Gathers and cross-checks sources, drafts chapters, prepares exhibits and the research log | Structure, ordering, first drafts | Every figure, checked against its original source, and every conclusion |
| Legal, finance and privacy | Prepares terms, tracks obligations and deadlines, flags conflicts between what a page promises and what the software does | Nothing that binds the business | Every commitment, every published policy, and anything a professional should review |
The handoffs are the system
Not four assistants working in parallel.
The value is in the order they run in, and what each one is required to hand the next.
Research produces a draft with sources attached; it cannot publish. Legal reads the draft against what the site and the terms already promise, and flags contradictions rather than fixing them. Technology only builds what has passed that check. Commercial can describe only what has actually shipped.
That last constraint is the one that does the most work, and it is the one most often missing when a firm adopts AI. Marketing that can outrun delivery will.
What this is not
One business, one operator.
A sample of one is not a benchmark. Nothing here is a claim about how much time or money this arrangement saves, because that is not a figure that could be sourced.
The model changes as the work changes. It is dated for that reason, and earlier versions are kept.