I · The diagnosis
Acceleration is a temporal structure, not a mood.
The term is Nick Land's. A teleoplexy is what emerges when a process becomes the input to its own intensification — a positive-feedback circuit in which means convert into ends and the loop begins optimizing for its own continuation. Capital is the canonical instance: machine intelligence and commerce feeding each other, each cycle shortening the next, with price as the signal that lets the loop steer itself.
Positive feedback is not exotic. It is present wherever development has a direction, and it is almost always experienced from the inside as something dangerous — which is why every such loop accumulates compensatory mechanisms around it: norms, regulations, institutions whose function is to damp the explosion. Land's twist is to notice that the compensators are external to the mechanism. The loop does not contain its own brake. It treats every brake as friction, and it is very good at routing around friction.
Land was right about the loop. What he could not see was a build in which the human is not the friction but the payload — an acceleration of the human together with the machine. The circuit still closes either way. We choose what it closes on.
II · The default termination
A loop with no address for the human.
Watch what agentic commerce is assembling right now: assistants that search, negotiate, transact, and settle — increasingly with each other. Every hop that once required a person is being compiled out. This is not malice; it is optimization. A human in the loop is latency, and the loop abhors latency.
Under the default termination, the person does not disappear — they are demoted to a boundary condition. Their preferences become a dataset. Their identity becomes whatever the crawler assembled. Their reachability becomes an externality traded between platforms: harvested, ranked, rented back. The extractive web already runs this pattern at human speed. Agent speed merely completes it.
The loop does not hate you. It simply has no address for you — and what cannot be addressed cannot consent, cannot refuse, and cannot be paid.
Every serious response so far has been compensatory: regulate the models, constrain the platforms, slow the deployment. Necessary, perhaps — but structurally these are brakes bolted to the outside of a mechanism that metabolizes brakes. The historical record of external dampers against internal feedback is not encouraging.
III · The inversion
Constraint as physics, not policy.
The teleoplexy of the title is a deliberate inversion of Land's: the same self-amplifying circuit, re-terminated so that every line of acceleration ends on a verified human node. Not slowed on the way there. Not filtered by an intermediary. Ended — the way a circuit ends at ground.
The mechanism is a single constraint installed inside the loop rather than bolted onto it: reach is a grant. Directional, revocable, recorded in the data layer, enforced beneath every model that touches the network. An agent may retrieve any card; it may knock on any door; it may not cross a threshold that has not been opened, and no prompt, no jailbreak, no commercial pressure changes that — because the constraint is not an instruction the model follows but a constraint the substrate imposes. A model can propose. Only the network can commit.
This is the difference between compensation and constitution. A compensatory mechanism fights the loop and is consumed as friction. A constitutive constraint defines the space the loop accelerates within — the way conservation laws do not slow physics but shape what physics can be. Consent, here, is not a brake. It is a boundary condition.
On price. Land identifies price as the teleoplexical signal — the number by which the loop steers itself. We keep price and strip it of its adjacencies: on this network, price settles deals but cannot buy position. Ranking is computed in the asker's own agent, against the asker's priorities, outside anyone's ability to sell. The signal survives; the auction for attention does not.
IV · Neutrality as a structural fact
The network that cannot sell what the others must.
Every incumbent directory shares one architecture: a database the subjects populate for free, a ranking computed server-side against the operator's objective, and reach rationed back to the subjects at a price. That architecture is not a choice they can revisit — the ranking is the revenue. An agent that returns one right answer generates no impressions, collapses the auction, and has nothing left to sell.
Which is why the good version of retrieval — inference over sincere, structured, consented records, ranked by the asker's own agent — will never be shipped by the platforms that could most easily build it. It is not withheld by incapacity. It is withheld because it is fatal to them and foundational to us. Read access is a policy they could reverse next quarter; write access — the ability of an agent to actually reach, transact, and settle — is the business model itself, and it stays closed.
You can pay to be in the network. You cannot pay to be on top of it — because there is no top. The ordering happens somewhere we cannot reach: inside the asker's own machine.
V · The future state
Three things that happen when the terminus holds.
Your machine carries your name. Home humanoids ship to American houses this year — twenty thousand dollars up front or five hundred a month, ten thousand pre-orders deep, with a remote human piloting the parts the robot can't yet do alone. Factory humanoids already bill by the operating hour. Fremont's line is designed for a million units a year. Which means millions of machines are about to act in the economy on somebody's behalf — buying, booking, answering the door — and nothing in that stack says whose behalf, provably, with consent attached. A card is that missing layer: your humanoid becomes a node, carrying your verification, acting under your grant, revocable the moment you say so. When a machine knocks, the person on the other side sees a name that was checked — yours. The robot doesn't replace you on the network. It represents you on it.
Price becomes honest. Dynamic pricing is arriving regardless — most retailers still can't do it, half of Europe's plan to pilot it this year, and the agentic-commerce economy carrying it is headed from sixty billion to two hundred billion inside five years. The question is which kind wins. The extractive kind prices you: your history, your urgency, what you'll tolerate. Regulators are already moving against it. The teleoplexy prices the work: labor, materials, weather, demand — real inputs, updated in real time, inside floors and ceilings the human wrote and the network enforces beneath every model. The price may know the harvest. It may never know your desperation. Run that across every card on the network and the spread that platforms have been harvesting — the gap between what a thing costs to make and what an auction can extract for it — flows back to the two humans at either end of the deal.
Capital is stored human energy — hours of a life, converted. A network that prices work honestly and settles it directly is not a marketplace. It is a conservation law for human effort.
The connection compounds. By 2030, roughly half of online buyers act through agents. Every one of those agents needs a corpus of real, consenting, payable endpoints to act on — and the network that holds them accretes the one asset no scraper can counterfeit: a ledger of verified humans who agreed, delivered, and were paid. Land saw acceleration escaping its passengers. He was right about the engine and wrong about the manifest. Built this way, the loop does not route around the human — it compounds through us. Every card added makes every agent more capable; every settled thread makes the next one cheaper to trust; the machine gets faster because the human is verified at the end of it.
You must connect. Not as surrender to the machine economy — as the act that decides its shape. The phone network began as two rooms and a wire, and its first decade was spent explaining why anyone would want one. The only question that has ever mattered for a substrate is whether the primitive generalizes. This one reaches from a food cart to a humanoid without changing a field.
Verified against: Figure AI — 1,000th Figure 03 produced at BotQ, July 2026; BMW billed ~$25/robot-operating-hour. 1X NEO — $20,000 or $499/mo, 10,000+ pre-orders, US home deliveries from late 2026 with remote human supervision. Tesla — Fremont line designed for up to 1M Optimus units/yr. Morgan Stanley — $5T humanoid opportunity by 2050. McKinsey-cited — 15–20% of retailers had dynamic-pricing capability (2024); 55% of European retailers piloting GenAI pricing in 2026. Mordor Intelligence — agentic AI in retail $60B (2026) → $218B (2031). Kearney/commercetools — ~50% of online shoppers via AI agents by 2030, ~25% of spend. Late-2025 regulatory turn ran against personalized pricing, not input-indexed pricing — the line this network builds on.