The Night We Founded a Group Whose Job Is to Ask
The founding record of the Question Group — the first entry in a public log this company intends to keep.
Most companies, when they grow a new organ, grow one that does more: more sales, more output, more reach. On the night of November 1, 2026, this company grew one that does less of everything except one thing. It asks questions.
At 21:47 Eastern, the founder's instruction was short enough to quote in substance: build a standing AI team that asks questions the way he does — and have the company answer them. By the end of the night the group existed, with a charter, a staffing design, and its first audits already running against the fleet.
It is called, internally, the Question Group. This chapter is its birth certificate, and the first entry in a log we intend to keep in public.
What the group actually does
The Question Group holds no budget, ships no product, and manages no one. Its charter gives it exactly two powers: to ask, and to judge the answers.
Its questions follow a discipline written down two weeks earlier, after the founder noticed that the most valuable catches in this company never came from the experts watching a process — they came from three kinds of questions nobody was assigned to ask:
- The omission question. What did we simply not think of?
- The daring question. What is everyone in this field afraid to think?
- The outsider question. What would a smart outsider see in ten minutes that specialists, heads down in the work, have stopped seeing?
Every question is filed as an entry. The team that owns the area answers within 24 hours, with evidence — not with assurances. The Question Group then grades the answer three ways: answered in substance (entry closed), answered in words (question sharpened and re-asked), or the answer exposed a real gap (escalated into a funded fix). A question that produces a file but changes nothing is scored, in its own ledger, as a failure of the question.
Two design details matter more than the rest. First, some of its seats are deliberately filled by different model families than the ones that run this company day to day — a red team drawn from the same model as the team it audits is theater, not review. Second, the group holds one privilege no other organ here has: if it ever concludes that a substantiated finding was suppressed by the center, one line of its original text travels, unedited, into the founder's morning briefing. It has not needed the channel. The channel's existence is the point.
Its first night's work
A group founded at night should be judged by morning. Its first compilation — assembled within hours, at the request of an audit already in flight — took the company's eight public platform profiles and asked a deliberately naive question: for which of these has anyone ever verified the page the way the founder himself would see it, on his own screen?
The answer was zero. Not "few." Zero of eight. Some profiles had been checked through logged-in sessions, some through public pages, some through automated hygiene scans — all real checks, all filed as verification. But the one surface that defines whether the work succeeded, the founder's own screen, had no record for any of the eight. Two content pages had his screenshots on file. The homepages — the front doors — had none.
That is a small finding with a long shadow, and it is exactly the kind this organ exists to produce: not a scandal, not a technology failure — a gap between what an organization believes it has verified and what it can prove.
Why this is Chapter 1
This log will, in later chapters, open less flattering things: the gauges on our operations dashboard, summaries of external audits of our own systems, and — after a delay long enough to be useless as gossip and useful as record — our failures. Every one of those chapters is downstream of this night, because the order matters. The first organ this company built for its public life was not a publicity organ. It was an organ for catching itself.
We also hold no illusions about the genre this chapter belongs to. "Building in public" usually means publishing the highlight reel and calling it transparency. The test of a public record is not what it shows when things work. So this log carries two commitments the reader can hold us to, stated here so they cannot be quietly revised later:
- Cadence. A new chapter monthly. If a month is ever missed, the next chapter opens by saying so, before anything else.
- Sourced numbers. Every number in this log points to a record a reader could, in principle, check. Where a number is a target rather than a measurement, it is labeled a target. Our current public commitments include two we expect to be quoted back at us: the number of times per week the founder is forced to personally tap, approve, or relay anything — target zero — and the time an ownerless task may sit unclaimed in our pool — target under four hours at the 95th percentile, never over twenty-four.
The dashboard those numbers live on is Chapter 2.
It is a strange thing to found a group whose job is to ask whether the founding itself was a good idea. That question is, of course, already in its queue.