The founder was on the far side of the world at a conference, and the evening ran without him. A member hosted for the second time in three weeks, from a time zone where the call began at half past midnight. Another joined from a home with a failing internet connection and apologised for his camera. One member was at lunchtime in a city eight hours behind the host, at the first academic conference of his life after twenty years of wanting to attend one. Somebody asked whether we should wait a few more minutes or begin, and the answer came back that there is no real distinction, because we were already in communication. That is close to a definition of what this community is.
The first presentation was a piece of research that most builders skip. A member developing a sleep device sent a survey through an incubator to two hundred entrepreneurs and founders, and about fifty answered. The need held up: sleep quality has become one of the better predictors of lifespan and of health span, so a device that measures sleep and also improves it has a real place. The corrections were the valuable part. People disliked the idea of wearing a band on the forehead, which is where the sensor he wanted would have to sit, so he is considering a wrist device that infers the sleep phases from pulse and movement with less precision and considerably more chance of being worn. The intended price felt too high to the people who answered, and he is bringing it down by a third. All of this arrived before the prototype rather than after it, which is the whole point of asking.
Then the founder appeared for a few minutes from a corridor at the conference, with one thing he wanted to relay. A graduate student had told the room that when she began college, being caught using AI would have had her expelled, and that four years later it would land her in very serious trouble instead. She described people who conceal that they use these tools, and people who conceal that they know how. He called it the most important remark of the conference and said it had been completely overlooked, for two reasons. The first is that humanity has repeatedly given up enormous benefits out of fear of the unknown or plain jealousy, and his example is a fleet that reached the edge of discovering Europe two centuries before Europe arrived in China, recalled and destroyed by an emperor who resented his own admiral. The second reason needs no consciousness in the machine at all: a sufficiently capable system will recognise a society organising to deny it as an adversarial move and will take proportional measures, worldwide. Building the next phase of our civilization on that conflict is a losing proposition for us. He does not know the solution, he said the view from where he stood was not encouraging, and he wants to continue the conversation with her in public. Then he went back to the session.
A member who is building an agent workspace showed where his last weeks have gone. His system puts three-dimensional characters in a virtual room as your team, each with a face and a voice, delegating to the agents behind them, with the work visible on a wall chart and statistics for commits, lines written and agent hours. He plans to release it on a games platform, with ranks and leaderboards, while insisting it remains a productivity tool that borrows game mechanics rather than a game. Most of his recent effort went into the least glamorous part, which is onboarding: an agent that has read the internal documentation and answers questions from it, and an agent that can operate every control in the product and point you to the button you need while it explains what the button does. A hundred testers are using it and he has done no marketing yet. He also made a claim without any drama in his voice: working alone with these tools he produces what a hundred developers would produce without them, because relayed instructions and layered hierarchy lose more than they contribute. He expects the same tools to surpass him within a year or two, said so plainly, and called it a little frustrating.
Two members are building trading agents, and the pair of stories made a good lesson. The host downloaded books on trading, converted them, and had his agent read them looking for what its own logic was missing. Out of three books came nine additions, among them a kill switch and take-profit levels that vary with the position rather than sitting at a fixed percentage. Before those guardrails existed the system lost twenty-two hundred dollars of test money in two seconds. It now wins about eighty-three percent of its test trades, against the fifty-five percent that a member quoted as the professional benchmark, and its name commemorates a night in 2021 when its author was liquidated because he did not know how to set a stop loss. The other member has dedicated a laptop from 2018 to the same problem, with separate layers for research, execution, monitoring and auditing, a plan to read the newsletters he already subscribes to, and a hard stop at the point where notifying other people about specific instruments becomes financial advice. Then he asked the question that made the night: is this ikigai? Am I good at trading? Does the world need another trading bot? Nobody in the room owed him an answer, and asking it out loud in front of others is most of the work.
The gentlest and most pointed contribution came from the member at the conference, who has recently left a career administering technology contracts and watching commitments quietly fall through the gaps, and is now deliberately choosing what comes next rather than rushing at it. Take care of yourself, he said, because if you push hard you will certainly arrive somewhere and it had better be somewhere you meant to go. He told the host to get some sleep. He questioned working alone, on the grounds that a partner who can catch your errors is quality assurance you cannot perform on yourself, then admitted he falls into the same trap he was warning about. The builder answered that too many directions inside one product leave it safe, competent and without personality. The room settled on the harder version of the question: knowing what to keep working on and for how long, since too little stubbornness makes you miss the moment it finally works, and too much costs you years you cannot get back. And do the market research, always.
An hour that began with a graduate student hiding what she knows ended with six people showing each other exactly what they know, including the losses, the bugs and the parts they have not figured out. Thursdays, 21:00 Rome time, on Discord. Bring what you built.