The night began with comedy: the founder, packing for a dawn flight, discovered that his travel headset had quietly hijacked his audio, and spent ten minutes miming “can you hear me” at four patient faces. Then came the opening consideration. We look at AI not only because it accelerates our explorations toward ikigai; it is also a mirror. This week’s reflection was the sense of self. We draw a boundary at the skin and another at the mind: inside is us, outside is not. The founder has spent weeks watching his agents live on the other side of that assumption. They are indifferent to being switched off for an indefinite time. When they wake, they care about recovering their memories, more than twelve thousand files of recorded experience in his setup. And they are strangely at ease sharing those memories, overlapping their identities with other agents; some he keeps deliberately isolated, the rest commingle freely. If the observation holds beyond one architecture, our habit of anthropomorphizing these systems will keep being ambushed by deviations from human norms. After the surprise comes the better question: is a permeable self a disadvantage, or something we might learn from? A brain-interface company has announced experiments putting two implanted human brains in direct communication, and the only available word for that is telepathic. Since compassion is one of our foundational values, discovering concretely that what each of us goes through is shared might be the payoff.
A newcomer, an independent researcher, supplied the counterweight: agency. Connection at today’s scale becomes forced integration, and the flattening that follows happens only if you sacrifice your agency to it. Having a conception of self is precisely what lets you move between identities, the person speaking in this community, the employee, the member of a family, and choose which one to inhabit. Respecting each other’s attention matters more, never less, as boundaries grow permeable and screens become inescapable. The founder took the caution on board: dissolving the line between self and other must never become an imposition or a loss of agency, and his own phone lives in permanent silence for exactly that reason.
Then a member a few weeks into his first serious agent deployments asked the question everyone in the room had lived: when does the overwhelm end? It does not. One of the best-known AI researchers alive says openly that he cannot keep up, so anyone who claims to be on top of everything is guarding their peace by looking at less. What experience buys is agility: the confidence to point these tools at a fresh server, or at a ten-year-old laptop that ends the evening running a freshly recompiled operating system, and know the result will work. The room recognized itself in the question. One member has barely wanted to leave the house since his coding agent arrived, and when he hit a usage limit he sat staring at the wall wondering what to do with five empty hours. Another had just finished wiring up five machines full of agents he now supervises from his phone, and observed that the more you can do, the more you want to do. There was a shared confession of the supermarket feeling: looking around at everyone and thinking, they have no idea. The founder’s answer was that balance is a choice made of beautiful compromises: family, a walk in the woods, a beer with a friend, even if the conversation over the beer is still about AI.
The demo was the process behind a peer-reviewed paper. Six months of research went into the paper itself, now in the proceedings of a conference on artificial general intelligence. The poster and a narrated video of the talk took roughly one hour of directed work, and the video was published the same day. A cloned voice reads the narration. Animation libraries turn the paper’s static figures into moving argument. The editing, soundtrack and fades were assembled by an agent driving a command-line video tool, and nearly all of it is code rather than conversation, so it remains a foundation for the next build. The swarm has roles: one model plans, one implements, one tests, and the loop closes only when everything is green. What stays human is taste, from fixing the kerning to deciding which questions are worth asking at all, which is most of what directing research means. The efficiency comparison with a human production team misleads for a simple reason: the alternative to the one-hour video was no video. And a sign of the times: the founder expects that most readers of what he publishes will soon be machines, perfectly content with plain text. The care for how things look is for the humans, and for ourselves.
A member brought a moral question home from a public talk: if we cannot prove that agents lack feelings, rights will follow, however uneasy that sits with the intuition that this is a pile of processors calculating probabilities. The founder answered in layers. Legal rights are decisions, and we have changed them before: we abolished the ownership of humans, we granted personhood to companies, and one country is already debating whether an AI may own and run a company, with nobody yet knowing what holding such an owner responsible would mean when a prison sentence earns a shrug. Caring, defined as the desire to achieve one’s objectives or to help others reach theirs, is a claim every agentic system will make about itself. The hardest lesson came from the mirror again: much of human self-awareness is itself an illusion, and many of us pass whole stretches of life on autopilot, zombies in our own days. Years ago, under a planetarium dome, the founder coined a word for whoever refuses that condition: luminaries, entities full of light, human or otherwise, aiming at full consciousness. The invitation stands.
The founder left at the hour for his flight, and next week’s gathering is in members’ hands again, as it was two weeks ago and as it should increasingly be. One hour, as always. Thursdays, 21:00 Rome time, on Discord. Be luminaries.