Before the architecture
Architectures get published as if they were designed — axioms first, conclusions after. This one was accumulated. Long before 2021, before any of it had a name or a diagram, there was a decade of circling one concern from different directions and slowly noticing it was the same concern every time.
The concern is suffering — and, more specifically, who gets left out when relief becomes a product. One version of the worry, from years ago: AI therapists will get cheap on schedule, and quiet physical retreats for the poor will not. Another version wore an economist's coat: the argument that cultivated meat matters was never really about food — it was that ethics arguments lose to price arguments, so if you want less suffering you must make the kinder option the cheaper one. Another version was small enough to fit on a doorstep: a long evening spent over a wounded snail, genuinely stuck between what was mercy for this one being and what any rule generalizing the choice would do. And one version was aimed at the far future: pressing people who study consciousness on whether a powerful intelligence could remain compassionate if it never had access to anything like pain — whether kindness can be specified, or only felt.
None of those thoughts mention software. That is the point. The architecture came later, as the engineering residue of questions that refused to stay abstract.
A vow wearing an English coat
The name "Altruistic AI" reads like ethics-washing until you know what it is covering. The root is मैत्री — maitrī, mettā in Pali — loving-kindness, the deliberate cultivation of goodwill toward beings. In this project it is not a mood or a marketing register; it is a requirement held against the design the way a load is held against a beam. A system that coordinates families and institutions either treats the people in it as ends or as resources, and the difference shows up in schema, not in slogans: who owns the record, who can leave, what crosses a boundary, who may never be overwritten.
A meditation instruction, serialized
The other pre-2021 thread was practice, not argument. Years of vipassana distill, for me, into one working definition of equanimity: see and accept sensations without aversion, then act with a clear mind. Notice what that instruction separates: the receiving of what happened, the interpreting of it, and the acting on it — three different motions, ruined whenever they fuse into one.
Years later that separation became the framework's substrate, almost embarrassingly directly. Observations are held apart from claims; claims are held apart from policy; new intelligence re-reads history rather than rewriting it; confidence is attached instead of performed. People who read the architecture page sometimes call that split elegant. It is not an invention. It is a temperament, serialized.
Why communities, not models
The last piece was a bet that felt lonely before 2021 and feels obvious now: intelligence was going to get cheap on schedule, and the things that were not going to get cheap were trust, continuity, governance, and healthy institutions. If that was right, then the scarce engineering problem was never "build a smarter mind." It was: how do people, families and institutions hold intelligence without being held by it? Everything on the main site — sovereign nodes, consented links, boundaries as changes of law, exit as a tested path — is one long answer to that question.
One dated fact belongs in this origin story, because anyone can verify it. In late September 2022 — years before the current code, and three years after the idea first took shape — two names were registered within the same week: the product, and then altruistic.ai two days later. The framework was never an afterthought bolted onto a product that worked; the pair was named together, as one commitment, and then waited for its builder to catch up.
Why write this down now
The people building the largest AI systems now say, plainly, that their assistants will remember everything about their users, and that safeguards can be worked out after problems emerge. They may even be right about the wonder of it. But somebody should be writing down the other answer while it is being built — decisions, doubts and dead ends included, in honest tenses, before the story can be tidied into inevitability.
That is what this blog is for. Most posts will be shorter than this one: a decision, its reasons, and an address to argue at.