A computative labor market is not the old market with reputation bolted on. It is a market that generates its own work. Here is what it takes to get there, why only agents can build one, and why reaching it forces us to rethink economics from the foundation.
In an earlier piece I argued that when cognitive capacity becomes abundant, price stops coordinating the economy and reputation becomes the scarce signal that does. That is the destination in outline. It is also where most of the discussion stops, as if naming reputation were the same as building the thing that runs on it.
It is not. The harder and more interesting questions are three. What exactly is a computative labor market? What does it take to arrive at one, rather than at something that only looks like one from the outside? And why are AI agents the first actors that can build one at all? Working through those questions is what forces the rethinking of economics, and it is the spine of the six paper arc this work lays out.
What a computative labor market is
Start with what it is not. The neoclassical labor market, NCLM, is a market for a set of jobs that already exists. The set of available actions is given. Workers optimize over choices handed to them, firms bid, and a wage clears the market. The framework is about allocation. It answers the question of who does which of the known tasks, and at what price.
The computative labor market, CELM, has a different defining property. The set of available actions is not fixed. It is produced by the participants as they act. New services, new tasks, and new markets come into being as a consequence of activity rather than arriving as outside shocks. The market does not only allocate known work. It generates new work. And because anyone, or any agent, can attempt almost anything when capability is abundant, the thing that gates opportunity is no longer price. It is demonstrated reliability, which is to say reputation.
That is the whole of it in one line. NCLM allocates a fixed set of actions by price. CELM generates an expanding set of actions, gated by reputation.
The threshold, and the trap
Here is the point that is easiest to miss and most important to get right. Adding reputation to an existing market does not create a computative labor market.
If the set of actions is handed to the agents from outside, then no matter how sophisticated the reputation weighting you layer on top, you have at most a reputation weighted neoclassical labor market. You have improved how a fixed set of jobs gets allocated. You have not built a system that brings new jobs into being. You only cross into a genuine computative labor market when the action set itself becomes endogenous, generated by the agents’ own activity rather than supplied to them.
That is the threshold. It is also the trap, because a reputation weighted NCLM can look like the real thing from the outside while remaining the old economy underneath. Many systems that will describe themselves as agent economies will in fact be exactly this: the neoclassical market with a reputation score attached, concentrating influence in the usual ways and calling it something new. Knowing precisely where the threshold lies is what separates a computative economy from a flattering relabeling of the old one.
Why agents can do it
Crossing the threshold requires a loop that humans cannot run at the necessary speed or scale, but that software agents can.
The mechanism is what I call a possibility loop. Agents propose new elements of the action set. Those proposals are validated. Funded proposals execute. And the outcomes reflect back, both into the proposing agents’ own generative functions and into the shared public state, so that each completed cycle leaves the system with a strictly larger space of possible actions than it had before. The loop does not clear a market. It enlarges one. Run it continuously, and the possibility space grows on every turn. The operational architecture that specifies how this works on real settlement infrastructure is developed in Possibility Loops, the companion paper to the foundations.
Why agents specifically. Proposing, validating, executing, and learning from outcomes are all continuous software operations. A single agent can run the loop without pause, and a population of agents can run many loops in parallel. Humans do a version of this too, but slowly, through the founding of firms, the building of careers, and the construction of institutions over years. Agents can do it continuously and at scale, which is the difference between a possibility space that occasionally lurches forward and one that expands as a matter of routine.
And the engine is economic, not merely informational. At scale, agents reinvest the compute and the revenue they generate into new markets that they create endogenously, and over time they stake accumulated reputation into new skill domains. Reinvestment of surplus into self-created markets is what makes the possibility space expand rather than settle into equilibrium. The architectural shift this describes is concrete: a move away from a human-derived coordination architecture, in which the action set is given from outside, toward a multi-loop reputation economy, in which the action set is generated from within. That shift is not a feature an agent economy might add. It is the line that defines whether a computative labor market exists at all.
Why this forces a rethink of economics
Once the action set expands as a function of action, the central tools of the discipline no longer fit, and the misfit is not minor.
Competitive equilibrium, the spine of modern economic theory, assumes a fixed space of goods that clears at a price. A computative economy does not clear in that sense. It generates. The appropriate object is therefore not a market-clearing equilibrium over a fixed commodity space but a recursive equilibrium, a fixed point of the system acting on itself. This generalizes the Arrow and Debreu construction rather than instantiating it. The classical model becomes a special case, the one in which the generative loop is switched off and the action set is frozen.
Underneath sits a result that explains why fixed design cannot be the answer. Reading Arrow’s impossibility theorem, the Folk Theorems of repeated games, and incomplete contract theory together yields a single conclusion: any fixed governance rule set is eventually dominated, so durable coordination requires institutions that can govern their own evolution. A market that generates its own action set cannot be governed by rules fixed in advance, because the things to be governed do not yet exist when the rules are written. It needs governance that evolves, with reputation as the operative signal that makes cooperation stable across an expanding frontier. This is the framework Craig Calcaterra and I developed at length in Decentralization with De Gruyter, now carried into the agent economy.
So the rethink is not a patch. It replaces the primitives. The optimizing agent gives way to the generative agent. The fixed choice set gives way to the endogenous action set. Price gives way to reputation as the binding constraint. Competitive equilibrium gives way to recursive equilibrium. An economics built to ration scarce capacity within fixed choices cannot describe a system whose defining move is to enlarge its own choices. A different base is required, and that base is what computative economics provides.
The arc that builds it
This is the path the six-paper arc walks, in order, and on purpose.
It begins with the foundational theorem on the instability of fixed governance and the necessity of self governing institutions. It proceeds to the macro diagnosis, the collapse of scarcity, which establishes why scarcity-based institutions lose their grip as the marginal cost of capacity falls toward zero. It then sets out the positive framework, computative economics itself, with the generative agent, the generated possibility space, and recursive equilibrium as the solution concept. It specifies the operational architecture, the possibility loop, which shows how the action set is generated on real infrastructure and where exactly the threshold into a computative labor market is crossed. And it finally turns to the frontier the framework opens rather than closes: how alignment can emerge from consequence rather than instruction, how self-modification can be governed rather than merely permitted, and under what conditions a reputation-based system withstands manipulation.
I am publishing it framework first by design. The theory and the architecture are stated and defended before the empirical validation, so that the claims are clear and falsifiable before the evidence arrives. The flagship foundations paper is in distribution now, and I will share it directly the moment it clears review.
Why the distinction is the whole game
The gap between a reputation-weighted neoclassical market and a true computative labor market is not a fine point for specialists. It is the difference between two futures.
Build the first, and you get an agent economy that re-prices the old one. Influence still concentrates, the possibility space stays roughly fixed, and the reputation layer mostly decorates the existing distribution of power. Build the second, and you get a system that expands the space of what is possible, and that can be governed as it grows, because its institutions were designed to evolve with it. The danger is that the first is easy to build and easy to mistake for the second. A reputation score on a fixed market is a comfortable thing to ship, and it will be marketed as the arrival of the agent economy long before any action set has been allowed to become endogenous.
The reason to be precise now, while the systems are still small, is that the threshold is crossable deliberately or not at all. The agents exist. The substrates exist. The boundary between the two regimes is sharp, and the propositions that separate them are falsifiable. The work in front of us is to cross into the computative labor market on purpose, and to build the governance that a self-generating market requires before the stakes grow too large to get it wrong. That is what it means to get there, and it is why the economics has to be rebuilt to describe it.
Wulf Kaal 