How I built this.
AI agents did the research and built these pages. Here is how the work was run, what was checked, and where the AI was wrong.
The workflow
MeSet the briefPitch the role with a plan and a working prototype. Read the company as its customer would first. Report what turns up, privately.
AIRead everything publicThe lead agent read OpenAI: 111 pages saved, two documentation indexes and 201 changelog entries searched. A second agent read six rival providers: 99 pages.
AICheck each findingEvery OpenAI finding was checked against the saved page text. One was rewritten and one was dropped.
AIBuildThe plan, the delivery health board and a private findings report.
MeRead and stand behind itI read every page of this site before sending it.
The lead agent was Claude. It did the OpenAI research, ran the rival research agent, did the checking and wrote the pages. One working day, November 22, 2026.
The rules the agents worked under
- Every claim carries its source.A page address and where on the page. The page text is saved, so any claim can be checked again.
- “Not found” is never rewritten as “missing”.The agents had to say where they looked. Only a statement by OpenAI itself counts as confirmed absent.
- Read only.No account, no login, no form, no testing of any system. Nothing behind a sign-in was opened.
- A separate agent for rivals.So what one provider says about another never stood in for that provider’s own page.
Where the AI was wrong or thin
- One finding was wrong on first read.The agent wrote that OpenAI’s newer capacity product states no service level. Its own page says the service level of the tier in use applies. The finding was rewritten.
- A search summary passed on a forum member’s claim as fact.It said one capacity product had been renamed to another. OpenAI’s pages show two products. Dropped.
- The agent’s own memory of OpenAI was out of date.Model names, tiers and whole products had changed since its training. Nothing here comes from memory. Every statement comes from a page saved on November 22, 2026.
- The rival rows are thinner than the OpenAI row.Eight rival facts were spot-checked. Several rival trust portals and partner lists could not be read at all, and the evidence page says so.
- The prototype’s numbers are samples.The only real facts on the board are the ones marked as OpenAI public pages, each with a link.
Where my judgment came in
The agents did the reading, the checking and the building. These five calls were mine.
- Pitch the role with a plan and a working prototype.The posting asks for a delivery plan and a clear view of delivery health. So I made both, not a letter about them.
- Read the company as its customer would, before pitching anything.A delivery lead’s first job is to see the deployment from the customer’s chair.
- Report what turned up, privately.If I find holes, I bring them up. The findings went into a report for OpenAI, not onto this site.
- Say what I cannot see.I have no view inside OpenAI. The plan says which parts are my proposal and what would change it.
- Stand behind the output.I read every page of this site before sending it. If something here is wrong, that is mine, not the agent’s.