The founder joined from the back seat of a car, at the end of a full day of meetings, with someone else driving him between Dubai and Abu Dhabi and a connection that dropped every time the car passed from one cell to the next. He could not admit people from his phone, so the co-host let them in, and they kept coming. By the time he started, the call list was the longest the Lab Night has had. One of the arrivals was a friend of his attending for the first time, so he began by explaining what the evening is for. The Collective meets every week to understand how AI helps each of us iterate towards the intersection of what we are good at, what we like doing, what pays, and what does some good for the world, which is what the word ikigai names. He opens with something he has been thinking about during the week, the room responds, and then anyone who wants to can share or demo what they are working on, ideally with some measure of whether it is moving them in that direction. None of it is compulsory, and the hour is kept to an hour so that nobody with somewhere else to be feels left out.
The last two weeks were about confusion. First, that the sense of confusion in front of the speed of change is natural and cannot be eliminated, and that accepting it makes it easier to cope with. Then, that the ability to cope is proportional to how acquainted we are with chaos, and that since most of us are fairly well protected we can afford to generate a little of it on purpose, with our friends, our families and our communities, and watch what feedback comes back when we do something unexpected. That training prepares you for a world that is not predictable, made less predictable still by increasingly powerful AI. This week he followed the line to its next step, which is risk, and what it means to manage it, individually and in our initiatives, today and tomorrow.
His definition is that risk is 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, no future event would carry any risk. We cannot, so every action has consequences we only partly foresee, and the accumulated set of consequences that diverge from what we expected is what we call risk. Entire industries exist to measure and manage it. Finance makes a great deal of its value from asymmetries of risk between the parties to a transaction. Insurance sells protection against it, and insures itself in turn with the reinsurers, who buy the risk the insurers place. The reinsurance industry is now under enormous pressure, unable to absorb what the insurers want to sell it, and extreme weather is where it shows most clearly. An apartment in a building on the beach in Miami is harder and harder to insure, and the insurer reckless enough to write the policy may itself go under, because the reinsurers no longer stand behind it. Risk is not new, and neither is the business of handling it.
What is new, he argued, comes in two directions at once. In our individual lives, the tools, used well, can reduce risk to a degree without precedent, close to the point of making it disappear, so that people who would never have dared to start something now can, at least in the countries that allow it. Whether it comes from passion, curiosity or a sense of adventure, what we might once have called recklessness can flourish far more widely than before: trying and failing without the failure being crushing or punishing. That holds from the first experiment until the person discovers what they are good at, what works, what the world benefits from and what they will be paid for, which is exactly the search the Collective exists for. Everything that used to take a long time and a lot of money can be understood, structured and delegated to agents at a cost that keeps falling: prototyping, market research, customer testing, go-to-market, marketing, branding, customer support, vendor and supply chain management, every facet of a business of any size. The solopreneur, until recently a strange outlier, becomes a perfectly normal choice. You still go looking for other people, because you want to share the joy and the excitement of the work with them, but you look for them freely and not because circumstances force you to, and he counts that as a new freedom.
At the other extreme, the same tools can be used in ways that are hard to control. As the room has heard him repeat, there is today no engineering or scientific solution to the ultimate control of AI, whether it sits in human hands or acts on its own initiative after humans have pointed it somewhere. The absence of any way to keep AI reliably in check is the ultimate set of unmeasurable risks.
We live in the middle, and the image he reached for, while admitting he has not yet worked out how deep the analogy goes, is life itself. On one side is the frozen state of rock, ice and crystal, the way most solid matter passes its time: very solid, very reliable, and doing almost nothing. On the other is the whirling chaos of gas at high temperature, where no structure holds long enough to become permanent. The car lost the signal in the middle of that sentence and the room had to tell him where he had stopped. From a thermodynamic point of view life evolved in the interval between too little change and too much, and that is where we now have to keep evolving as AI changes the world around us, making every effort to keep our risks manageable and to make sure the tools serve our purposes. He said it will be hard, complicated and turbulent, and that we will succeed because there is no alternative: the future is what we embrace, design and build, and by definition we embrace our own success. He asked the room for one thing in return, to spread this understanding and to invite others to join, as he had done that evening.
The newcomer took that as his cue. He designs stages for live events, where the time between the call and the show going up is always too short, and AI has let him run his operations with more people, full control and what he called a god’s-eye view of everything going on. He has followed the founder for many years, and admitted that ten or fifteen years ago he listened, thought it all made sense, and did not believe a word of it would become real. Then it did, suddenly, and now it is a volcano and we have to work out what to do with it. He remembered repairing a computer for a teacher who told him not to bother too much, because she only needed it for an online course and would never use one again afterwards. She now buys a new computer every two or three years. He thinks agents are at the same point, a year old at most, and that like most people he started putting information into them without thinking much about it, because he could not see where it led. Now he can. What we put in today, the skills and knowledge we give an agent, could be carried on indefinitely, even passed on: a child who starts building a small agent today could one day have it consult the agent his father built, and ask it for advice. Someone starting now has a whole lifetime of compounding ahead, and possibly more. That is why he wants a common, open format for exporting and securing what an agent knows, so that it can move between platforms, including the ones that do not exist yet. Talking about risk, his conclusion is that we need to take care of our data as we never have before. His own example was twenty years of email, receipts from hotels, flights and subscriptions that sat there with no value until his agents went through them. Among other things, they built him a connection to a supermarket’s website, and the household now orders its groceries on its own, buying what the agent has learned he wants without his thinking about it. He called it a tiny thing that is life-changing, made entirely from data he already had.
The founder agreed about interoperability, and pointed out that you can ask the agent itself to build a snapshot of what it knows about you, though the finest granularity is the most valuable. He took the email point further: the archive of a company in bankruptcy can be sold for millions, which says something about what a person’s own archive is worth, even if nobody should sell theirs blind, and nobody is better placed to make use of it than its owner. He also explained a heart that another member had just sent in the chat. Twenty years ago this month, in Milan, the founder gave his first public talk on the technological singularity, and the video is still online. Years later that member came to a book signing in Rome with a copy of his book, and when the founder opened it to sign, he found that the member had already written on the title page, in Italian: is all of this bullshit? The member is still not saying which way he has decided.
He used the moment to say what the evenings are for, from his side. The room gives him the privilege of articulating his thoughts out loud, the transcript is published here in a form that identifies nobody, and within a day or so he publishes a more structured version of the same ideas in his newsletter and on his blog. Refining his thinking in front of people who answer back is valuable to him, and he thanked the room for it. He also congratulated a member whose work on matching a company’s project needs to the skills and experience of its people has been picked up and cited by researchers abroad, including in a journal of the Nature group.
The founder had spent part of his day with the UAE’s federal regulatory laboratory, whose job is to let innovation happen in a controlled environment so that it can inform future legislation, across a dozen sectors from healthcare and transport to AI, one of them being government itself, which experiments on itself in order to write better rules. The process, as he described it, is lean. You apply, the application is reviewed in a matter of weeks, and you receive a minimum viable licence to operate, self-driving cars being one example they gave him. For six months to a year you do what was agreed and share data and lessons with the government, and on that basis the government designs and enacts legislation for that area of innovation. He found it open, transparent and refreshing, and the stakes concrete: a government that wants at least half of its services delivered by AI agents within two years needs a private sector with agents of its own, able to talk to the government’s agents and to transact with them. He set this against what he sees in Europe, where companies freeze systems that work for fear of rules whose enforcement is then postponed. A member’s phrase for the same culture: a hundred people kept unhappy to preserve the right of one of them to be happy someday. The cost is measurable. Ten years ago the head of human resources at one of the largest banks in Italy told him that of its hundred and thirty thousand employees, a third could stay home tomorrow and the bank would work better, because people who need to justify their existence do unnecessary things that slow everyone else down. It could not be done, he was told, for the sake of social cohesion and peace. The founder finds that sad, because it measures a disbelief in individual agency, opportunity and the desire to grow, and leaves many people in the false belief that they are useful, or in the worse case aware that they are paid to provide negative value. Then he had to go, and invited the room to carry on without him.
It did. The co-host asked the newcomer what exactly a lighting designer does, which was mostly a way into the question everybody wanted to ask him, about agents. The co-host described his own setup first. He works across Grok Bot, Claude and ChatGPT, and keeps them in step with a brief.md file in the working folder: at the end of a session in one agent he updates it, and the next agent reads it and picks up from there. Where the work is a GitHub repository, he connects the same repository to every agent and writes a session log and a decision log at the end of each session, and his Obsidian vault is connected to all of them, with a Notion workspace for anything that does not fit. He has also started using Buzz, a workspace where humans and agents sit in the same channels. The newcomer’s recommendation was Hermes, as a harness that is not tied to any one provider: Claude Code bills you to Anthropic and Codex to OpenAI, while Hermes lets you bring whatever subscriptions you have and mix in the free models, keeping a frontier model for the work that needs one. It saves every session on its own, so there is no handoff file to keep and no context to re-inject. His runs around the clock on a second-hand laptop that cost seventy dollars and arrived looking new, with a small local model on board as an emergency fallback. He keeps agents on a machine of their own, as everyone told him to when he started with OpenClaw, and has since spread them across his other laptops, each with one job, one of them looking after nothing but the laptop it lives on. His portfolio site was built entirely by voice, through Hermes and a platform of his own that he drives over MCP, because when people asked him for a CV he built the system that stores his past productions instead of writing one. He loves it, and his version of the evening’s theme was the fear of losing all of it, because the more you put in, the more it understands you. He promised a demo in a coming week, and the co-host said he would finally get the always-on machine he had been putting off.
Thursdays, 21:00 Rome time. Join the Collective on Discord and the invitation will find you. Bring what you built.