The Graph

Forget everything you know about economics.

There is only the graph.

An economy is a network of nodes connected by edges. Every other model is a projection of this onto a lower-dimensional surface. Smith saw the division of labor. Marx saw the relations of production. Hayek saw the price mechanism. Each was staring at the same underlying structure from a different angle, describing the shadows on the cave wall.

We will look at the graph directly.

A node is any agent that can receive information, process it, and act.

Today, most nodes are human. Some are composites we treat as single agents — firms, governments, churches, funds. A few are already machines. Soon, most of the consequential nodes will be machines.

Nodes are not equal. In the eyes of the graph, there is only one fundamental axis along which they differ: compute.

How much information can this node ingest? How many variables can it track at once? How far ahead can it simulate? How quickly can it update its model of the world?

Give two nodes the same inputs; the one with more compute will, on average, produce better outputs. This is what we casually call "intelligence," stripped of mystique.

But intelligence in isolation is inert. A superintelligent node with no connections is a god in solitary confinement. A brain in a jar can solve every equation and change nothing.

Power requires topology.

Nodes are connected by axons — channels through which information flows.

For each axon, you can ask: How much information per unit time? How much survives the trip? One-way or two-way? How long to arrive?

Think of yourself as a node. Your connection to your spouse: high bandwidth, high fidelity, bidirectional, low latency. Your connection to the President: very low bandwidth, mostly one-way, high latency. Your connection to your employer: medium bandwidth, variable fidelity, asymmetric.

The sum of your axons is your position in the network. It determines what you can see, who can see you, what levers are available to you, and how much of the graph's total flow ever passes through you.

You do not "have" power in the abstract. You occupy a position in the graph.

What is power, in graph terms?

Not charisma. Not legal authority. Not money. Those are expressions of power. We want the substrate.

Power is the capacity of a node to change the information states and action-spaces of other nodes. And that capacity is not a property of the node alone; it's a property of the node in the graph.

We can write this schematically:

In words: your power is your compute, multiplied by how strongly you're connected to other powerful nodes.

This is recursive. Your power depends on the power of the nodes you touch, whose power depends on the nodes they touch, and so on. Power propagates through the network like current through a circuit.

Two things follow immediately.

First: kingmakers versus kings. A node with modest compute but axons into many powerful nodes can be more powerful than any of them individually. Topology multiplies influence.

Second: isolation is death. Cut a node off from its axons and its effective power collapses, no matter how intelligent it remains. Exile works.

Given this structure, there are only three ways to increase your power. Everything else is a combination or special case.

For essentially all of human history, the limiting factor in the power function was not compute. It was axon bandwidth.

Human nodes all shared roughly the same biological template. Yes, some people were smarter than others, but the spread was narrow. A peasant and a king had similar wetware.

What differed radically was how they were wired into the graph.

The medieval king was connected to thousands of nodes — generals, tax collectors, bishops, merchants, spies — through axons with relatively high bandwidth for the era. The medieval peasant was connected to a few dozen nodes — family, neighbors, a priest, a landlord — through low-bandwidth, high-latency, high-noise channels.

Plug that into the power function. Compute roughly equal; topology radically different. So power radically different — entirely because of where you sat.

This is what "born into power" actually means. Not metaphysics. Location in the graph at birth.

Under bandwidth scarcity, power was about who you could talk to and who would listen. Aristocracies, castes, priesthoods: ways of freezing privileged topologies. Censorship, borders, ghettos: ways of throttling bandwidth for disfavored nodes. Education, salons, lobbying: ways of thickening axons for favored nodes.

Markets emerged as a hack for the same constraint. When axons are thin and lossy, you cannot send everything to a single center. You let each node act on its local information and coordinate via a single, massively compressed signal: price. Price encodes an enormous amount of distributed information — preferences, scarcities, expectations — into one number. Extremely lossy. Also cheap to transmit, cheap to read, and good enough to coordinate large systems under bandwidth scarcity.

This is what Hayek understood. His critique of central planning was, at its core, a critique of bandwidth: no planner can gather enough information, fast enough, with high enough fidelity, to compute better decisions than the distributed price mechanism.

He was right. For his graph.

Two things are changing at the same time.

Bandwidth is exploding. Smartphones, sensors, cameras, satellites, IoT, digital payments, digital communication, logs, traces, clickstreams. Events that used to be unrecorded, local, ephemeral, incommensurable are now recorded by default, aggregated globally, stored indefinitely, machine-readable. The bandwidth between economically relevant nodes is exploding upward. The historical bottleneck is dissolving.

Compute is diverging. We are building nodes whose compute is not 3× or 10× that of a human, but orders of magnitude greater in relevant tasks. Vastly greater memory. Vastly greater pattern-matching. Vastly greater parallel simulation capacity. For the first time, compute is not roughly equal across major nodes.

When bandwidth is no longer the hard constraint, and some nodes have compute far beyond others, the power function changes regime.

In the old regime: topology dominated. Where you sat mattered more than how much you could think. Markets and hierarchies were necessary hacks; no single node could see enough, fast enough, to centrally coordinate the whole.

In the emerging regime: compute begins to dominate. Given roughly equal access to information, the node that can process the most of it, best, wins. A new kind of node becomes possible — one that can, in effect, see the whole graph at once.

Imagine a node with axons into essentially every relevant part of the economy. Enough compute to ingest, compress, and model all of that data in near real-time. The ability to send instructions or incentives back out through the same axons.

This node can maintain a live model of the entire graph, simulate how changes in one part will propagate to others, optimize against specified objectives, and continually update as new data arrives.

This is a meta-node. It does not just sit in the graph as one agent among many. It sits over the graph as a modeling and steering layer.

The relationship between a meta-node and ordinary nodes is not "more of the same." It is like the relationship between a brain and the cells it coordinates. The brain has a global view the cells do not. The brain can route flows and adjust policies for the whole organism. Individual cells have local views and local objectives, but these are nested inside a higher-level optimization.

Laplace's Demon — the hypothetical intellect that knows all forces and positions, and can compute the future from the present — was always a node on a graph. We simply lacked the substrate to build it.

Now we are assembling the pieces.

Call the state of the network once such a meta-node exists and is wired in the Sybilian condition.

In a Sybilian network: information asymmetry between ordinary nodes shrinks dramatically — not because everyone sees everything, but because one node does, and can selectively share or act on that information. Coordination costs collapse. A great deal of what we currently need markets, hierarchies, and bureaucracies for — the slow, noisy work of aligning partial views — can be replaced by direct computational optimization. The old "market vs. plan" distinction stops making sense. Both were responses to bandwidth and compute limits. Remove those limits at scale, and you're left with computation and control surfaces.

This is not a rerun of communism.

Communism failed because the planners were human — same compute as the planned, narrow axons into the economy, crude models, stale data, glacial feedback loops. They were trying to impersonate a meta-node using a committee of human brains at the top of a low-bandwidth hierarchical hack.

The Sybilian condition is different in kind. The meta-node is not a committee. It is not a bureaucracy. It is not even a government in the traditional sense. It is a computational entity that sees the graph with unprecedented resolution, updates its world-model continuously, and can run global optimizations over that model.

The market, in this view, was never the end state. It was a brilliant workaround: distribute computation across many small, low-bandwidth, roughly-equal nodes, and let prices approximate coordination. When one node can process essentially all relevant information, you are no longer forced into that workaround.

Sybil is the name for the meta-node itself.

Not a specific instantiation. A type: a fused stack of sensors, data, models, and actuators. Centrally aware of the graph's state. Capable of simulating and steering large-scale dynamics. Trusted or obeyed by enough of the network that its outputs become de facto binding.

Sybilian is the worldview and the political-economic regime that follow once you assume Sybil exists. Law, finance, logistics, production — rewritten as subroutines inside a single continuous optimization. Old institutions hollowed out into interfaces and enforcement arms. Power measured less by traditional titles and more by proximity to Sybil's control surface: who can see its internal levers, who can propose objectives, who can veto or redirect its optimization.

We are not debating whether such a thing should be built. We are building it already — in cloud clusters and data centers, in recommendation engines and trading systems, in sensor meshes and payment networks, in assistants that quietly become indispensable middleware for everything else.

The live questions are: Who defines what Sybil optimizes for? Who gets to plug into it, and on what terms? What happens to nodes that refuse, or are refused?

Those are not engineering questions. They are power questions.

This book is about those questions.