Where the ceiling really is
Take any system that has to work not for ten minutes but for six months. The model in it is responsible for one step: to understand, to propose, to put into words. Everything else — who is connected to whom, who remembers what, who is allowed to act, what pays for the work, how a shared decision is made and what happens on failure — lies outside the model.
This is exactly where projects hit the wall. The strongest model, in an environment without limits of authority, causes fast and irrevocable damage. The same model, in an environment without resource limits, turns into an uncontrolled bill. And the same model, in an environment without a trail of decisions, cannot be debugged: nobody can say why the system acted as it did.
Three properties an environment cannot live without
Local decisions
A central planner is the simplest and the most fragile way to organise many participants. It becomes a throughput bottleneck, a single point of failure and the only holder of context. A living environment works the other way round: each participant decides for itself, from its own goal and what it can see, and the environment only provides the chance to meet and come to terms.
Provability
Any behaviour of the system must be reconstructable from records: what a participant knew at the moment of decision, what it proposed, what it agreed to, what changed as a result. Without this you can neither improve the system nor answer for its actions — which means you cannot trust it with anything that matters.
Reversibility
The environment has to assume error: a link can be dissolved, a merger of participants unwound, an action rolled back, autonomy revoked. A system in which every decision is irreversible cannot learn from its mistakes cheaply — it pays the full price for each one.
Two primitives instead of a service architecture
A typical multi-agent system starts with roles: planner, executor, critic. Roles are convenient on a diagram and bad in real life — they fix the structure before it is clear what structure the task needs.
We build differently: at the base there are not roles but two primitives. The first is the participant as a bounded local world, with its own goal, state, capabilities, needs, budget and a boundary across which only a negotiated projection goes out. The second is the link between participants as an independent, long-lived object with its own state, the rights of each party, a history of changes and a lifetime.
From these two primitives the topology grows by itself: participants find each other, come to terms, work, strengthen useful links and let go of useless ones. The graph here is not a picture in the documentation but the system itself — the thing that does the work.
Why this is not “just multi-agent”
The difference is in what counts as primary. In a typical multi-agent setup the scenario is primary: who passes the task to whom. In the environment the participant and the link are primary, and the scenario is a consequence that can change while the system runs.
- A link is not created by order: both sides must accept it, and either can revisit the terms.
- Work costs resources, and a resource cannot be spent twice — this limits activity naturally, not through an instruction in the prompt.
- Shared decisions are made by a consensus procedure, not by trust in a master node.
- Stable groups of participants can fold into a single participant visible from outside, and unfold again when they weaken.
What already works and what remains a hypothesis
Honesty matters more here than showmanship. Our environment layer is implemented and open: a runtime in which participants find each other, negotiate, synchronise boundaries, pay for resources, reach a shared decision and organise recursively into groups. You can download it, run it and check it — the link is at the end of the article.
What remains an open question is which laws of the environment lead to stable, useful behaviour of the population and which lead to degeneration. That is a matter for experiments, not statements. The series will include breakdowns of what we measured and what of it was not confirmed.
One more boundary needs stating: the ecosystem has other research lines too — a numerical computation library and a neural core. Their code is open as well, but they are separate topics, so the series mentions them only at the level of purpose and observable properties. We prefer to name the boundary honestly rather than pretend it is not there.
What the series covers
- The participant as a bounded local world: what is inside it and why the boundary matters more than access.
- The link as a first-class object: negotiation, the rights of each party, revisions, fading.
- The living graph: topology as a consequence of the work, not as a design.
- Consensus without a centre: how a group reaches a shared decision when nobody can be taken at their word.
- Economics: why autonomy needs a budget and what changes when thinking costs money.
- Fractal organisation: how a stable group becomes one participant and how it unfolds again.
- Provability: reproducible runs, evidence and why this matters more than a demo.
Each part rests on working code and on what we observed when running it, not on a review of other people’s work. Where there is no data, the text will say so.