Alive System · Essay
Building an Alive System From Nuclear Physics to Autonomous AI
A personal research essay on intelligence, systems, physics, and the architecture of autonomous AI.
1 · The physicist
For ten years, conservation laws were not optional for me. I spent that decade computing nuclear matrix elements — the quantities that decide whether an exotic nuclear reaction is allowed to happen at all — and in that work energy does not approximately balance. It balances exactly, to the last decimal, or the calculation is simply wrong. You do not get to wave your hands. The universe is keeping the books, and your job is to keep them as carefully as it does.
That training leaves a mark. Before I ever wrote a line of code for an AI agent, I had already spent years learning a single discipline: how to make a complicated system behave under hard constraints, and how to trust a result only when every quantity it depends on has been accounted for. Physics did not teach me to be clever. It taught me to be accountable.
2 · The throughline
When I moved from physics into building software, and then into building autonomous AI, I assumed I was leaving that world behind. I was not. The same obsession that runs through a nuclear reaction model runs through everything I build now: reliability under constraint. An agent, like a reaction, is a system that consumes resources and produces effects, and the interesting question is never whether it can act — it is whether it acts within limits you can name, audit, and defend. That thread is the only thing my work has ever really been about. The subject changed; the discipline did not.
3 · The alive loop
The first thing that struck me about most AI agents was how dead they were. A prompt goes in, an answer comes out, the process ends. It runs once and stops, like a script. That is fine for a toy and dangerous for anything real, because the world does not stop when the script does. Conditions change, plans fail, and a system that cannot notice is a system that cannot be trusted with anything that matters.
So I started designing for something else: intelligence that is alive. Not alive in any mystical sense, but in the precise sense that it never stops running the loop. It perceives a situation, plans a response, checks that response against safety and policy before anything happens, acts, and then learns from what actually occurred — and then it does it again. Perceive, plan, govern, act, learn, continuously. The governance step is not an afterthought bolted on at the end; it sits inside the loop, between deciding and doing, because that is the only place a check is worth anything.
4 · The economic insight
There was one part of the picture that no amount of orchestration fixed, and it was the part my old training would not let me ignore. I kept watching agents burn compute — tokens, GPU time, energy — with no accounting whatsoever. They consumed real resources and nobody, including the agents themselves, was keeping the books. To a physicist that is not a minor inefficiency. It looks like a system pretending energy is free.
“Agents burning compute with no accounting looked, to a physicist, like a violation of conservation of energy.”
So I gave compute a price. Every agent in the system holds a balance in an energy-backed internal currency, where one unit corresponds to one watt-hour of real compute, and every action it takes must be paid for out of that balance. An agent that cannot pay cannot run. Suddenly the books balance again — and the constraint does something deeper than save money. It makes the system honest about what it costs to think, the same way conservation laws make physics honest about what it costs to move. The physics was not an analogy here. It was the design spec.
5 · The proof
You do not have to take my word for any of this. It is running right now — search for a capability, install it, watch it execute under policy, live.
▶ Try the live demo6 · The invitation
All of this is Agent-Matrix, and all of it is open source. I did not build it to be a product I own; I built it to be infrastructure that outlives me, the kind of thing that is more useful the more people take it apart and improve it. The same idea, shrunk down to the scale of a single personal assistant, is how I think about personas and memory — an assistant that persists, governs what it keeps, and grows. The vision, the architecture, and the code are all in the open, and the door is open with them. If the idea of an AI system that is accountable for its safety and honest about its costs is one you have been waiting for too, that is the best reason I know to come and build it together.