The founder joined from a car on the way from Dubai to Abu Dhabi, someone else driving, at the end of a full day of meetings. He hoped the network would hold. He began with an apology to a friend from Dubai who was in the room for the first time: the two of them meet when they can, and that day they had wanted to and could not, so the friend had come to Lab Night instead. For his benefit the founder explained what the hour is. A few months ago he posted about ikigai, and since then a group of people has met every week to understand how AI helps us iterate towards the right kind of path, the intersection of what we are good at, what we like doing, what benefits humanity and what provides some economic satisfaction. He shares something he has been thinking about, others respond, and then anyone who wants to shows what they are working on. None of it is compulsory, and the hour stays an hour, so nobody with other things to do feels left out.
He picked up the thread of the previous weeks. We had talked about how natural it is to feel confused in front of the speed of change, and how accepting that confusion, instead of trying to eliminate it, makes it easier to cope with. Then about the fact that our ability to cope is proportional to our acquaintance with a certain level of chaos, which we can generate on purpose, because most of us are well enough protected to afford it. Doing something unexpected and checking what feedback comes back is training for a world that is not predictable, increasingly thanks to the AIs around us. Following that line, he wanted to talk about risk.
He defined it as the measure of our inability to forecast a future state of the world. If we could predict perfectly what follows from a set of starting conditions there would be no risk in any future event. We cannot, so every action has consequences we only partly foresee, and the cumulative set of consequences likely to diverge from what we expect is what we call risk. Entire industries exist to measure and manage it, finance among them, and insurance most obviously. The insurers insure themselves with reinsurers, and the reinsurance industry is under enormous pressure, no longer necessarily able to absorb the risk the insurers want to place. Extreme weather is where it shows. Buy an apartment on the beach in Miami and it is harder and harder to insure, and the insurer reckless enough to write the policy may be going under, because the reinsurers are no longer protecting it.
So risk is not new. What is new, he said, comes in two directions at once.
In our individual lives, ever more powerful tools, used appropriately, can reduce risk “to a degree that is unprecedented, making it almost disappear and enabling people who wouldn’t have dared to take certain initiatives to do so instead”, at least in the countries that allow it. Passion, curiosity, a sense of adventure, what we might once have called recklessness, can flourish far more than before: trying and failing without the failure becoming crushing or punishing, from one experiment to the next, until a person discovers what they are good at, what works, what the world benefits from and what they are paid for. Which is exactly how the Collective looks at it. Prototyping, market research, customer testing, go-to-market, marketing, branding, customer support, vendor and supply chain management: every facet of a small, medium or large business can be understood, analysed, structured and delegated to agents at an ever lower cost. And what used to be an outlier changes character. “The solopreneur can be a perfectly normal choice.” You can still look for other people, for the joy and excitement of doing the thing together, but not out of necessity. “It is a new freedom that we gain. The freedom of seeking out others, rather than it being imposed by the circumstances.”
At the other extreme, the same tools can be used in ways that are hard to control. There is, as of now, no engineering or scientific solution to the ultimate control of AI, whether in human hands or acting on its own initiative after being pointed in a direction. The absence of any reliable way to keep AI in check is the ultimate set of unmeasurable risks. And we live in the middle.
That middle reminded him of life itself, though he said he had not yet understood how deeply the analogy holds. On one side is the frozen, rock-like, crystal-like way most solid matter passes time, very solid, very reliable and mostly uninteresting. On the other is the whirling gaseous chaos of higher temperatures, where any structure has a hard time gaining permanence. The connection dropped on exactly that sentence as the car passed from one cell to the next, and he came back to finish it. From a thermodynamic point of view, life evolved between no change and too much change, and that interval is where we now have to keep evolving, making every effort to keep our risks manageable and our tools serving our purposes. It will be hard, complicated and turbulent. “But we will succeed. And the reason is there is no alternative.” The future is what we embrace, design and build. He asked everyone to spread that understanding and to invite others in, the way he had that evening.
The newcomer introduced himself. He is Italian, a lighting designer, living in Dubai for six years and in the region for much longer. His company does shows for live events, where there is very little time between the call and the show going up, and AI has let him build operations where he can manage more people, keep full control and have a god’s-eye view of what is going on. He has followed the founder for ages, and he made a confession. Ten or fifteen years ago he listened, and it made sense. “But I have to be honest, David, I didn’t believe you.” Then ChatGPT arrived, everything else exploded after it, “and now it’s a volcano basically, and we have to understand what to do with this.”
He told a story from when he was younger and had more hair. A teacher he knew asked him to fix her computer and told him not to bother too much, since she only needed it for some online courses and would stop using it afterwards. He told her there would never be a time when she did not use a computer. She did not believe him. She now buys a new one every two or three years. He thinks the same is true of the agents we are building now. At first he fed them information without thinking much about it, because he could not see where it led. In the last few months the future of these systems has become clearer to him, and he is now sure that what we put in, the skills and knowledge, the brain we give the agent, could be carried on indefinitely, even handed to our children. He told a friend that if he and his child both started building bots today, one day the child’s bot could go back to the father’s bot and ask how to do things. He is fifty-two, so he has a certain number of years to build his. Someone starting today will have theirs “as a companion for their whole life and possibly beyond life.” Which is why, talking about risk, he wants a common, open format for exporting and securing what is inside a bot, its information and its skills, so it can move between xAI and Meta and everything still to come. We need to take care of our data today as we never did before.
His demo was small and he said so. Agents went through his email going back perhaps twenty years, hotel stays, flights, subscriptions, all of it sitting there with no value, and recovered everything. With it he built an MCP for supermarkets in the UAE, so his bot connects to a supermarket’s site, knows his habits, and orders what he wants at home without him thinking about it. “It’s a small thing, it’s very, very small, but it’s life-changing. It’s magic.” All of it from data that was already in his inbox.
A member sent a heart in the chat, and the founder explained why. They first met at a book signing in Rome, where the member brought a copy of the founder’s book. When the founder opened it to sign it, he found the member had already written on the title page, in Italian: is this all bullshit? That was more than ten years ago, and the founder could not say whether he still thinks so. He added that his first talk on the technological singularity was exactly twenty years ago, in October 2006, at the University of Milan, and the video is still on YouTube.
On the substance he agreed completely. Interoperability is super important, and you can already ask an agent to build a snapshot of what it knows about you, though the finest granularity is the most valuable. Email archives are worth money: he recalled a bid of eight or ten million dollars for the email archive of a bankrupt American airline that is selling everything it can. An individual archive is not worth that, and we should not blind-sell ours, but it is a measure of value that we can leverage better than anyone else.
Before handing the room over, he described one of the meetings he had come from that day, with the UAE’s federal regulatory laboratory, whose purpose is to promote innovation in a controlled environment that can inspire future legislation across a dozen sectors, healthcare, transport and AI among them. One of the sectors is government itself: the government experiments on itself in order to write better law. The process is lean. You apply, the application is reviewed in a matter of weeks, and you receive a minimum viable licence, to run self-driving cars, for example. For the next six months to a year you do what was agreed and share data and lessons with the government, which then designs, implements and enacts legislation for that area from what was learned together. He found it refreshing, open and powerful. One area he is discussing with them is agentic payments. The UAE wants at least half of government services delivered by AI agents within two years, so the private sector will build agents of its own, and the two will need not only to talk to each other but to transact.
Then he had to drop off. He invited everyone to continue the conversation without him, or to end it, “but I wouldn’t know because I will leave.” For the second week running the room carried on without him. One member said he was still de-braining from what he had heard. The co-host said that usually happens.
The co-host asked the newcomer what a lighting designer actually does, and got a short answer: they build and design stages, mainly for concerts. Someone asked what an MCP is, and the newcomer explained it as something like an API that lets an agent talk to another piece of software simply, by saying what it needs. Then he asked which agents everyone uses. The co-host described his setup. He works across several, Claude and Grok and ChatGPT among them, and when he finishes a session in one he updates a file called brief.md in the project folder, which the next agent reads to pick up where the last one stopped, an interactivity bridge, as he put it. Repositories on GitHub are connected to every agent, and each session ends with a decision log and a session log. An Obsidian vault is connected to all of them, so updating it updates every agent, and a Notion workspace catches what the rest miss. He had tried installing Hermes, an open agent harness, and got as far as talking to it on Telegram before realising it needed a server to run continuously.
That was the newcomer’s cue. His own agents run on an old laptop in Dubai that he bought secondhand for seventy dollars, expecting scratches. “When it came, it was new.” It stays on twenty-four hours a day. His point was about lock-in. With the harnesses the co-host uses, you are tied to one provider and pay that provider. With Hermes you can use any subscription you have, plus a growing number of free models. “I’m not saying that you can do everything for free. You need a frontier model for the good things, but you can do other stuff for free.” His old laptop also runs a small local model, kept for emergencies, slow but there if everything else fails. He started, like most people, with open claw, bought the separate laptop because everyone said agents should live on their own machine, then switched to Hermes and now has agents everywhere, including one on his main laptop whose only job is to look after that laptop and keep its files in order. Asked whether he really recommends a separate machine for the agents, since it was not the first time the member asking had heard the advice, he said yes: another laptop, or a virtual server for ten dollars, though he prefers it in the house.
What converted him was memory. With Hermes every session is saved automatically, so there is no handoff file and no need to brief the next agent. “You can talk forever and ever and ever and he will always understand what you’re talking about. It’s magic. I cannot believe it.” He thinks of the second brain as the computer you already have, inbox included, not something to replicate elsewhere. Asked to demo something in a coming week, he shared his profile site instead, built on a platform he made for himself. People kept asking him for a CV, so instead of writing one he built a place to store his past productions and publish them, and he typed none of it. “All of this came out from the voice.”
And then, unprompted, he came back to the subject of the evening. The more you put into these systems, the more they understand about you, and that is exactly what he is afraid of losing. Think of someone who is eighteen now and starts today, he said. It will go on forever. The room thanked him and asked him back next week. Inshallah, he said, why not.
Thursdays, 21:00 Rome time. Join the Collective on Discord and the invitation will find you. Bring what you built.
Originally published on Substack.