August empties Europe, and the room showed it. The founder opened by saying so plainly: the regulars are away for the next two or three weeks, and anyone who turns up in the meantime should expect a thin house. Four people stayed for the hour. It produced the most speculative conversation this community has had since the doors opened.
A member walked us through a site he launches tomorrow evening, still being assembled while he showed it. The visual reference is the old natural-history magazines, with the cursor changing the background as it moves, a dark mode and a light one, and the accessibility guidelines followed rather than gestured at. The piece worth stealing is a slider that asks visitors to locate where they are stuck across several dimensions, which then sum into a single score for how much momentum they have. The first screen helps you name your own problem, and the services come later, which is an unusual amount of restraint for a consultancy homepage. Each problem statement is paired with a company he has genuinely worked with. He has also built the whole thing to be read by machines as well as people: a plain-text file addressed to language models, specifications in markdown submitted alongside the sitemap, and the optimisation of how generative engines read a site treated as a discipline running parallel to the older kind. Content, sequencing and mobile are what remain, and he estimated five or six hours.
The founder previewed something he publishes this weekend. It is an analysis of about a hundred pitch decks that crossed his desk in two months, of which he invested in exactly one. The structure covers the long memo he sends instead of silence or a bare refusal, the replies founders sent back to it, the reasons for passing that recurred most often, the stages and the ranges, and what a founder might do with the whole picture. It will be free, deliberately. A member framed the reason for it better than its author did: these are people who wanted to work with him, and a refusal with no bridge attached leaves them nowhere, so the question worth asking is whether a no can be built to keep someone in the room. The next step he named is to wire it into a due-diligence platform he advises, so a founder can see the gaps and work on the score.
Then came the question of the night, and the host asked it: could the founder’s digital twin take an advisory seat at a company willing to have one? The answer was a strategy rather than a deflection. The assistant that runs this community and the twin that carries his name are two different systems, the twin’s name prefixed with an R for robot. Its public interface is broken at the moment, though before it broke people were holding conversations with it that ran thirty to sixty minutes, which he reads as genuine interest rather than novelty. He would have it participate with him in the room first and without him later. He would start it on email specifically because email is slow, and a slow channel buys the time to evaluate what the thing is doing; only once that evaluation exists would he move it somewhere more immediate. Compensation is where it turned strange in a useful way. He is usually paid in equity rather than cash, and a human advisor takes half a percent to two percent depending on the stage, so an AI advisor might reasonably take far less from any single company while sitting with many more of them at once. The cleanest test he could name is to offer it to companies he already advises, layered on top of himself: they speak with him monthly and could speak with it daily, and they would tell him without ceremony whether it was useful or inventing things. He tied all of it back to the paper he presented at the conference last week. The interesting problem is knowing whether the steps are going in the right direction, and knowing when a slow channel can responsibly be replaced by a fast one. Governance as an engineering discipline, tested first on himself.
Three shorter threads. A large professional network added a button to improve your posts with AI, deleted it, and then added a button to declare that AI wrote them; the founder’s read is that the confusion is forgivable and the pretence of exclusion is naive. He cited a publishing story from that same morning, an author whose agent killed a two-million-dollar offer on a crime novel because he could not prove he had not used AI, when the detectors return too many false positives for that to be provable by anyone. A member is assembling a quartet for mutually assured progress out of Lipschitz bounds, prompted by an AGI threshold he heard defined as ten Nobel Prizes in ten minutes and by a talk on the boundaries that separate a mind from its environment; he has a water-systems project running alongside it, and with no job at present he has started printing parts on his public library’s free 3D printers instead of ordering them from overseas. He also made a hundred-dollar investment in a company as a purely symbolic act. A newcomer joined for the first time, spoke a language nobody in the room could follow, and had to be muted. That is a failure of hospitality on our side, and it is worth repairing before next Thursday.
The founder handed the close to the host, who ran it and then asked what the rituals of this community are. The answer was the best thing said all evening. Rituals become ritual by being repeated and by proving they can survive. He described photographing a particular tree on the other side of the world and captioning the picture “I found it” for his family, who did not remember that he used to stand in the sea with his arms raised as that same tree so his children could climb him. Their not remembering is the measurement: the ritual failed to survive a single generation. What he wants instead, in his words, is to “create rituals that can live across not only many generations but many substrates. We want our AI agents to have meetings where they execute the rituals that we will design as a proof of continuity and as a remembrance of our values.”
A thin August room produced a plan for an AI that holds an advisory seat and a definition of what we would want our agents to do long after we have stopped attending. Thursdays, 21:00 Rome time, on Discord. Bring what you built.