What a transaction costs before it happens
Every exchange carries costs that have nothing to do with the price agreed. Before a deal is struck, someone has to find a counterparty, work out what the going rate even is, negotiate terms that anticipate the ways things could go wrong, write those terms into a contract precise enough to be enforced, then watch performance and be ready to litigate if it fails. None of that is the transaction itself. All of it is the cost of being able to have the transaction at all.
The temptation is to treat these costs as friction — real, but incidental, the sort of thing that gets smaller as institutions mature and disappears in the analysis. Ronald Coase's contribution was to refuse that move. Transaction costs are not noise around the allocation of resources; they are what decides how resources get allocated in the first place. Where the cost of using the market — finding, pricing, negotiating, monitoring, enforcing — exceeds the cost of simply directing the activity yourself, the activity gets pulled inside an organisation and run by command. Where market contracting is cheap relative to internal direction, the activity stays outside, coordinated by price. The boundary of the firm, on this view, is not a legal accident or a matter of taste. It sits at the point where these two costs balance.
This is a claim about structure, not sentiment. It says nothing about efficiency in the abstract, only about which mechanism is cheaper for coordinating a given piece of activity, given what can currently be known and enforced. Change what can be known, and the balance point moves.
Coase's question, and what came after it
Coase asked the question in 1937, aged twenty-six, in a paper called "The Nature of the Firm." The puzzle he set himself was almost embarrassingly basic for an economist trained to believe in the price mechanism: if markets allocate resources so well, why does so much production happen inside firms, under orders, insulated from price signals entirely? Why does a factory not simply subcontract every task to the lowest bidder at every hour? His answer was that using the price mechanism has a cost, and that cost is sometimes higher than the cost of hierarchy. That single move — costing the market itself, rather than assuming it is free — founded transaction cost economics.
Oliver Williamson extended the framework from the 1970s onward, giving it teeth. He identified asset specificity, opportunism and bounded rationality as the conditions under which contracting becomes expensive enough to justify integration, and turned Coase's insight into testable predictions about when firms make and when they buy. Coase won the Nobel prize in 1991, Williamson in 2009. Neither was writing about technology. They were writing about why General Motors owned Fisher Body, why the corner shop is not a subsidiary of the flour mill. The framework was built to explain institutional boundaries observed in the twentieth century, full stop.
Where the observation goes
Look closely at what most transaction cost actually consists of, and a pattern appears. A large share of it is the cost of not knowing something. Did the steel meet spec. Did the driver brake hard on the motorway or coast into the junction. Did the tenant run the boiler at thirty degrees all winter or twelve. Contracts are elaborate — full of warranties, penalty clauses, inspection rights, arbitration procedures — precisely because finding out the truth of a matter after the fact is expensive. Firms exist, in part, for the opposite reason: inside a firm, you can generally just look. A supervisor can walk the floor. Observation is cheap inside the boundary and costly across it, and that asymmetry does real work in deciding where the boundary sits.
This is where the intake axis becomes relevant, and it is worth being careful about how it becomes relevant, because the connection is not decoration.
A Large Language Model works from a corpus frozen at some cutoff. It can price only what has already been written down, published, and archived. Ask it to arbitrate a live dispute about whether a shipment met spec last Tuesday, and it has nothing: the record it holds ended before Tuesday existed. A Large World Model changes this by sensing a scene while the scene is present. Point sensors at the loading bay and you can verify, at the moment of exchange, whether the pallet count matches the manifest. That collapses a great deal of inspection cost — but only at the moment of exchange. The Large World Model goes dark once the scene ends, and most contractual risk is not at the handshake. It is in the months after, when the boiler is running unattended and nobody is watching.
A Large Universe Model is defined by holding many streams open with no stopping point: revisable beliefs, each carrying provenance, decaying in confidence as they age rather than vanishing at a cutoff. Applied to a commercial relationship, this means the thing contracts exist to simulate — ongoing verification of performance — becomes directly available. Ongoing verifiability is exactly what a penalty clause, a warranty, an inspection right, is a substitute for. Once ongoing verifiability is cheap, the substitute has less work to do.
The claim, stated precisely
The intake axis terminates because the reasons to contract rather than observe are finite, and each one is a limit on observation. You contract instead of watching when watching is unaffordable, when it is too slow to matter, or when nobody can be held to what was seen. A system that keeps every relevant stream running, never stops watching, and attaches provenance to every belief it forms has removed all three limits at once. Nothing goes unobserved. Nothing waits for a cutoff. Nothing is anonymous testimony with no source.
You can certainly imagine that observation getting cheaper, faster, better calibrated, more widely trusted. What is harder to imagine is a fourth category of evidence that is not reducible to more of it, faster, with a better audit trail. The categories transaction cost economics inherited — the market's price, the record, the certificate, the inspection — were carved out by what could not be watched continuously. Remove that constraint and the remaining work is not conceptual. It is scale, calibration, and the slower business of institutions agreeing to trust a stream. Coase's boundary keeps moving after that point. It is the menu of ways to move it that stops growing.
Three objections, taken straight
Perfect monitoring does not remove bargaining power. A supplier who has sunk capital into relationship-specific plant can still be held up, however well you observe them.
This is correct, and it narrows the claim rather than merely qualifying it. Continuous observation dissolves the measurement branch of transaction cost — the not-knowing. It does nothing to the bargaining branch — the leverage that comes from sunk, specific investment. Hold-up survives omniscience. What does change is the interaction between the two: hold-up is most damaging when paired with informational asymmetry, because the victim cannot tell a genuine cost shock from an invented one. Remove the asymmetry and hold-up becomes a naked distributional fight, which can be priced contingently ex ante. A smaller problem, differently shaped. The argument here is about intake, not about power, and it was never meant to be about power.
Cheap observation raises transaction costs before it lowers them. Every new stream is a new thing to dispute — whether the sensor was calibrated, whether the vendor's reading is admissible. Telematics litigation in freight and insurance shows measurement multiplying arguments, not settling them.
This is the strongest empirical objection and the short-run evidence supports it. But it locates the failure correctly: contestation rises where provenance is thin — a single vendor's number, no chain of custody, no corroboration. That is exactly what "revisable beliefs with provenance and decay" is built to answer. The claim is not that raw observation ends disputes; a raw reading asserted without a source is just a new claim to fight over. The claim is that only observation, properly sourced, can end disputes, and that provenance is the whole difference between data and evidence.
Every prior information technology — the telegraph, the shipping container, EDI, ERP — was said to redraw Coase's boundary permanently, and each time a further class of observable turned up. Calling this one terminal repeats a known mistake.
A fair warning. But each of those technologies removed one specific constraint — distance, latency, incompatible formats — and left the others standing. The claim made here is structural, not enthusiastic: intake is bounded by what is observed, when observation stops, and whether it can be attributed to a source. "Everything, continuously, with provenance" saturates all three simultaneously. The test is straightforward — name a fourth class of evidence that is not just more, faster, or better-trusted streams. No earlier generation, including this one's own predecessors on the axis, could make that offer honestly.
The misreading to disown
The weak version of this argument says cheap observation dissolves the firm: everyone becomes a spot-market contractor, hierarchy withers, Coase's answer to his own question stops applying. That is wrong twice over. Observation lowers the cost of contracting and the cost of internal direction at the same time, and Williamson's own data suggest large organisations often integrate more, not less, when monitoring becomes cheap — visibility makes coordination at scale more attractive, not less. And measurement was never the only term in the equation; bargaining power, sunk relationship-specific investment and ordinary political control over an organisation survive perfect visibility intact. The boundary moves. It does not vanish.
What this does and does not establish
Continuous, provenanced observation removes the measurement component of transaction cost as a limiting factor. It does not remove bargaining power, sunk cost, or the plain fact that someone still has to decide who reports to whom. It shifts disputes from "what happened" toward "who has standing to say so" and "how much is this worth," which are older, harder arguments that observation alone was never going to settle. What it establishes is narrower and, for that reason, more defensible: among the three things that have historically forced activity into a contract instead of a glance — unaffordable observation, delayed observation, unattributable observation — a system built on continuous, provenanced streams has no fourth constraint left standing. That is a claim about the ceiling of one axis. It is not a claim about the end of bargaining, or of firms, or of argument.