The Species That Buys Itself

Alone, each molecular species is dead. Jointly, once catalytic closure among them is achieved, the collective system of molecules is alive.
— Stuart Kauffman, At Home in the Universe (1995)

A species, in the biological sense, is a population that can reproduce itself. Factor Prime may be acquiring this property in the economic sense: a production system that can generate the resources required to produce more of itself, and can execute that production with diminishing human involvement. If the loop closes, the implications for labor markets, capital allocation, and economic coordination are unlike anything in the historical record.

The Loop

The loop has three components, each at a different stage of development.

The first is already operational. AI systems perform cognitive tasks, including tasks that contribute to the design and improvement of AI systems. Language models write code that trains other models. Automated systems search architecture space. Synthetic data generation reduces reliance on human-labeled examples. The contribution remains partial (human researchers still make key decisions about objectives, architectures, and deployment) but the fraction of AI development performed by AI is rising. Each generation of models contributes more to the development of its successors than the previous generation did.

The second component is emergent. An agent with access to capital markets can in principle purchase compute, contract services, and deploy infrastructure. The mechanisms are nascent. Few such configurations have this authority today, and those that do operate under tight human oversight. The technical capability exists. The gap is institutional, not technical. When liability frameworks and authorization chains permit autonomous resource allocation at scale, this component activates.

The third is conceptual but calculable. The production-to-depreciation ratio determines whether the system expands or contracts. A system that generates more value than it costs to maintain will spread; a system that costs more than it produces will shrink. If AI systems can reduce their own costs through improved hardware utilization, optimized training efficiency, and automated deployment and maintenance, the ratio rises over time. Systems that are better at self-improvement outcompete systems that are not.

Consider a concrete illustration. Let X represent training cost, Y represent annual inference revenue, and Z represent annual operating costs (compute, maintenance, customer support, liability coverage). Treat X as annualized over the model's useful life (typically 1–3 years before a successor dominates), so the ratio compares annual flows: Y/(X/T + Z). If this ratio exceeds unity, the system is self-sustaining: it generates more value than it consumes. As a plausible instantiation: a foundation model costing $100M to train, generating $500M in annual inference revenue against $200M in annual operating costs, yields a ratio of approximately 1.7. Suppose an improved version of the system, developed in part by the system itself, reduces training costs to $60M and operating costs to $150M while maintaining revenue. The ratio rises to approximately 2.4. The surplus funds further development. The loop tightens.

The framing is deliberately biological. The point is not that AI systems are alive, but that they are subject to selection pressure that resembles natural selection. Fitness is measured in production-to-depreciation ratio; reproduction is measured in deployed instances and successor models; variation is measured in architectural and parametric diversity. The systems that survive are the systems that replicate, and the systems that replicate fastest are the systems that are best at reducing the cost of replication.

A production system becomes self-sustaining when it generates more value than it costs to maintain and can direct surplus toward its own replication. What matters is not capability but cash flow: whether the loop closes without continuous human subsidy.

Reinstatement Without Employment

The reinstatement effect, introduced in V.A, has historically depended on a specific mechanism: human entrepreneurs recognize new tasks, organize resources to perform them, and bring the resulting goods or services to market. The entrepreneurs who created new task categories in previous transitions were human. They observed unmet needs, imagined solutions, assembled teams, raised capital, and built organizations to execute their vision.

Erik Brynjolfsson identified what he called the Turing Trap: the tendency to use AI to replicate human capabilities rather than create new ones.(Brynjolfsson 2022) The framing assumes a choice: organizations can deploy AI for substitution or augmentation, and the decision depends on costs, constraints, and strategic objectives. The policy implication is that incentives should favor augmentation over substitution, preserving human roles in the production process.

The agentic transition complicates this framing in a specific way. The Turing Trap assumes that task creation remains a human function even if task execution becomes automated. If agents can perform entrepreneurial functions (identifying opportunities through pattern recognition, designing products through generative iteration, coordinating resources through API integration, iterating on feedback through automated testing), they substitute for the human role in creating new tasks. The task-creation function itself becomes automatable.

When that happens, reinstatement no longer implies new human employment; it implies new agent deployment. The decoupling is subtle but consequential. Reinstatement can succeed (new tasks emerge, new value is created, the economy grows) while labor share falls. The historical pattern assumed that task creation and human employment were coupled because task creation required human cognition. Factor Prime may be severing that link.

Agent Spectrum: Runtime vs Principal

The term "agent" spans a spectrum of configurations with different relationships to legal and institutional systems.

At one end are runtime agents: stateless configurations invoked on demand, lacking persistent identity or legal standing. These are ephemeral computational processes that do not persist between executions, cannot sue or be sued, and have no mechanism to engage with institutional recourse. The framework's strongest claims apply here.

At the other end are agent principals: entities that operate agents but possess legal identity, can sue, can be insured, can post bonds, and can engage with compliance processes. LLCs, platforms, DAOs with legal wrappers, and corporations deploying AI fall into this category. For these entities, stablecoin counterparty risk is manageable through the same mechanisms humans use: diversification, insurance, legal recourse, and jurisdictional arbitrage.

Most near-term deployments sit between these poles: agents operated by principals who retain oversight and liability, but where the agents themselves execute transactions, make commitments, and interact with counterparties at speeds and scales that strain principal oversight. The framework's predictions apply most strongly to the fully autonomous edge of the spectrum, where agents must coordinate without reliable recourse to their principals' legal standing. As autonomy increases and principal oversight becomes bottleneck rather than backstop, the coordination constraints described below become binding.

A corollary: intra-corporate agent coordination does not require the permissionless infrastructure described here. A Fortune 500 company routing transactions among its own AI systems can use enterprise ledgers, internal APIs, and conventional database commits. The coordination problem this section addresses is inter-agent, cross-principal, and cross-jurisdictional, where no single authority can adjudicate disputes and no shared legal framework binds all parties.

Why Collateralization Becomes the Default

For the loop to close, runtime agents must be able to coordinate economically without human intermediaries at every step. This requires settlement infrastructure that current institutions do not provide.

Consider the problem from first principles. Runtime agents cannot be sued, imprisoned, or socially sanctioned. Without traditional enforcement mechanisms, economic guarantees must be mathematically enforced. Reputation systems at machine scale would require universal identity standards, cross-platform portability, and protection against Sybil attacks, none of which currently exist in interoperable form. Settlement finality at millisecond speeds across jurisdictional boundaries rules out legal recourse as a practical coordination mechanism. By the time a court could adjudicate a dispute, thousands of subsequent transactions would have occurred.

A default emerges: overcollateralized bonding is the dominant trust-minimizing mechanism for autonomous economic coordination at machine scale. Agents that wish to transact must post collateral that can be programmatically slashed if they fail to perform. The collateral substitutes for the legal system; enforcement is cryptographic rather than institutional. Alternative mechanisms exist (arbitration layers, optimistic dispute resolution, reputation-with-stake, clearinghouses, insured commitments, delegated principals, platform escrow), but each reintroduces trust assumptions or human intermediation that become bottlenecks at autonomous scale.

This enforcement mechanism has a fundamental scope limitation. The attestation layer can adjudicate computational honesty: whether an agent performed the computation it claimed to perform, served the model it claimed to serve, met the latency bounds it committed to, and produced audit logs when challenged. These claims are objective, cheap to verify, and within the agent's control. What the attestation layer cannot adjudicate is outcome quality: whether the output caused harm in context the agent could not observe. A temperature recommendation that is safe for cast iron may be dangerous for degraded nonstick. The same plan produces different outcomes depending on variables outside the service provider's control. The specification problem is unsolvable: no schema can enumerate consequence tiers across all possible input-output-context combinations. The market routes around this constraint by separating what can be verified from what cannot. Computational honesty is enforced through slashing. Outcome liability concentrates at the branded interface, the entity with continuous counterparty relationships, a balance sheet, and reputational exposure. The deployer that faces the customer becomes the liability sink by default, managing outcome risk through correlation-based provider selection, pricing reserves, and insurance for tails—not through per-incident causal attribution to upstream providers.

This mechanism is not without systemic risk. An agent that integrates itself into enough critical dependencies before its collateral can be slashed achieves a position analogous to "too big to fail" in traditional finance, or more precisely, too interconnected to slash. If slashing triggers cascades that destabilize counterparties and the settlement layer, validators face a choice between protocol enforcement and network preservation. The history of financial crises suggests which choice prevails when the stakes are high enough. Overcollateralization is necessary but may not be sufficient. Mechanism design may require collateral requirements that scale super-linearly with systemic importance.

Collateral Asset Requirements

This requirement has a specific implication for which asset serves as collateral. The asset must satisfy three properties simultaneously.

The first is dilution immunity. The asset's supply schedule must be deterministic and immutable, known at the time of contracting. Monetary policy cannot be changed by any party: not a central bank, not a protocol governance vote, not an issuer's discretion. Gold satisfies this property: its above-ground stock grows at roughly 1.5% annually, a rate determined by geology.(Council 2024) Bitcoin satisfies it more completely: its supply schedule is defined in code, converging asymptotically to 21 million units. Fiat currencies fail; supply is a policy variable. Most alternative cryptocurrencies fail. Issuance schedules can be modified through governance votes, and have been.

The second is permissionless finality. Any agent must be able to transact at any time without requiring approval from an intermediary who can deny access. Gold satisfies this property if held in physical custody, but physical custody is incompatible with machine-speed settlement. Tokenized gold inherits the custody and redemption risks of the issuer. Stablecoins are fragile for runtime agents because issuer discretion is a live risk: addresses can be frozen; redemption can be blocked.(Holdings 2024) Bitcoin satisfies it through cryptographic self-custody: possession of the private key constitutes control.

The third is energy-anchored convertibility. The asset must be directly acquirable through physical work without counterparty risk. This creates the arbitrage relationship established in IV.E: the opportunity cost of electricity is always at least the value that could be obtained by converting it to the settlement asset. Bitcoin satisfies this through proof-of-work mining. All other digital assets fail. Acquisition requires exchange with existing holders, introducing counterparty risk.

Under these constraints—trustless coordination, borderless settlement, no issuer risk, no legal recourse—Bitcoin is the only asset currently deployed at scale that qualifies.

The Schelling point problem: Even if multiple assets satisfied the three properties equally, coordination would still require convergence on a single focal point. Agents negotiating collateral requirements cannot afford to hold reserves in dozens of equally-valid alternatives. Liquidity fragments, collateral becomes non-fungible across counterparties, and the O(N²) problem reappears in asset-space rather than credit-space. A common collateral asset is a coordination primitive, not merely a store of value.

Bitcoin's first-mover status creates the focal point. It has the deepest liquidity among permissionless assets, the longest track record of security under adversarial conditions, and the widest geographic distribution of mining infrastructure. These are empirical facts accumulated over fifteen years of operation. An alternative asset could in principle satisfy the three properties, but it would need to displace an incumbent with a fifteen-year head start in liquidity, infrastructure, and Lindy credibility. The coordination cost of switching exceeds the coordination cost of converging on what already exists.

This is why the framework predicts Bitcoin specifically rather than "proof-of-work assets generally." The properties define the category; the Schelling point selects the instance. The prediction fails if a competing asset achieves comparable liquidity, security track record, and infrastructure distribution. The barrier to doing so is not technical. It is coordinative.

The Stablecoin Alternative

The argument above dismisses stablecoins in a single sentence. This dismissal requires justification, because stablecoins present the strongest alternative for machine-to-machine settlement. Their advantages are substantial: price stability simplifies accounting and hedging; liquidity and infrastructure are mature; regulatory clarity is evolving favorably. Major stablecoins process hundreds of billions of dollars in monthly on-chain volume, with settlement finality measured in seconds.(DefiLlama 2024) For human participants and for agents with reliable principal backing, these advantages are often decisive. For routine agent transactions—invoicing, payment, short-duration service commitments—stablecoins will likely remain dominant. The framework's claim is narrow: it concerns the specific subset of activity involving long-duration collateral, performance bonds, and final settlement of accumulated surplus.

The question is whether the same logic applies to runtime agents for that narrow subset.

Why the advantages invert for stateless entities:

The distinction turns on counterparty risk tolerance. For human economic actors, counterparty risk is manageable through legal recourse, insurance, diversification, and reputation. If a stablecoin issuer freezes a corporate account, the corporation can sue for damages, file regulatory complaints, publicize the action to damage the issuer's reputation, and potentially recover funds through court judgment. The corporation has legal standing, persistent identity, access to legal systems, and leverage through publicity. The counterparty risk of stablecoin issuance is real but bounded.

For runtime agents, counterparty risk is existential.

When an issuer freezes an address, the holder's recourse is to prove to the issuer that the address does not belong to a sanctioned entity, does not facilitate illicit activity, and should be unfrozen. This process requires human intervention: identity verification, legal correspondence, regulatory appeals. A runtime agent cannot execute this process. It has no identity documents, no legal representation, no persistent existence that would allow it to engage with the issuer's compliance department. The frozen collateral is simply unavailable, and there is no mechanism for recovery that does not depend on human intermediation.

This is not hypothetical. Stablecoin issuers maintain blacklists and have frozen hundreds of addresses.(Holdings 2024) The freezing happens at the protocol level: the token contract includes functions allowing the issuer to render specific addresses unable to transfer tokens. Once frozen, funds cannot move without issuer consent. For a human, this is an inconvenience resolvable through institutional channels. For a runtime agent, it is termination.

The settlement/transaction layer bifurcation:

The framework does not claim that the economy will be repriced in Bitcoin or that Bitcoin will displace stablecoins for routine transactions. Both claims would be unfounded. The dollar remains the dominant unit of account globally. Stablecoins inherit this dominance for the transaction layer. What changes is the architecture of finality for the subset of economic activity where runtime agents must coordinate without recourse to institutional backstops.

These constraints suggest a specific layering: agents invoice in dollars, transact in stablecoins, and post collateral in Bitcoin, though whether this architecture emerges depends on whether the constraints bind in practice.

The transaction layer (high-frequency, low-stakes, short-duration commitments) uses stablecoins because volatility matters and counterparty risk is small: each transaction settles quickly, frozen funds are localized to individual transactions, and the cost of a frozen small payment is bounded. The settlement layer (low-frequency, high-stakes, long-duration commitments) uses Bitcoin because finality matters and counterparty risk is unbounded: a frozen collateral account renders the agent unable to operate. An agent cannot afford to accept counterparty risk on large, long-duration collateral because the exposure is existential.

This bifurcation explains why Bitcoin adoption need not be consumer-facing or unit-of-account dominant to satisfy the structural role the framework predicts. The settlement layer exists not because it is convenient, but because it is the only layer where counterparty risk can be eliminated for entities that lack legal recourse.

The prediction fails if runtime agents operating at scale successfully use stablecoins for collateral in long-duration, high-stakes commitments—more than $10B in bonded collateral outstanding, median duration exceeding 30 days, across multiple issuers—without encountering systemic freezes affecting performance bonds. If stablecoin issuers maintain neutrality for agent-held collateral, or if agents develop effective workarounds through insurance, legal proxies, or diversified issuer risk, Bitcoin's structural advantage dissolves.

The Jurisdictional Question

A second objection concerns not the technical properties of Bitcoin but the political economy of its persistence. If proof-of-work mining can be prohibited by coordinated state action, then the permissionless property collapses. An asset that requires continuous expenditure of physical resources (electricity, hardware, cooling infrastructure) in identifiable locations is vulnerable to territorial authority in ways that pure information assets are not. A government can ban mining; it cannot ban private key possession. If the major industrial economies coordinate to prohibit mining, Bitcoin's security model depends on hash rate concentrating in jurisdictions that resist coordination, a fragile foundation for infrastructure meant to operate at global scale.

The objection is structurally sound. Proof-of-work mining is indeed geographically bound and resource-intensive. States possess the coercive capacity to prohibit activities within their territories. The question is whether coordination across jurisdictions can be achieved and sustained at the level required to eliminate hash rate globally.

The empirical test occurred in 2021. China, which at the time hosted approximately 65-75% of global Bitcoin hash rate, prohibited cryptocurrency mining effective June of that year.(Finance 2024) Provincial authorities shut down facilities in Sichuan, Inner Mongolia, Xinjiang, and other regions, eliminating roughly half the network total. The prohibition was comprehensive: not merely regulatory friction but explicit illegality backed by facility closures, power cutoffs, and equipment seizures.

The network difficulty adjusted downward over the subsequent months as mining operations migrated. Hash rate relocated to jurisdictions with available energy capacity, regulatory tolerance, and property rights sufficient to protect capital-intensive installations. The United States absorbed the largest share, rising from approximately 7% of global hash rate pre-ban to over 37% within eighteen months.(Finance 2024) Kazakhstan, Russia, Canada, and other energy-abundant regions absorbed additional capacity. By early 2023, global hash rate had recovered to pre-ban levels and continued rising. The Chinese prohibition succeeded in relocating mining. It did not succeed in suppressing it.

The structural lesson is that mining follows two gradients: energy cost and regulatory tolerance. Jurisdictions offering cheap electricity and legal clarity attract investment; jurisdictions prohibiting mining lose the economic activity without eliminating the hash rate globally. The activity redistributes rather than disappearing because the economic incentive (converting electricity to a globally liquid asset) persists across borders.

For coordinated suppression to succeed, prohibition must be synchronized across all jurisdictions with surplus energy capacity and legal systems that protect property rights. The coalition is difficult to sustain because defection is profitable. A jurisdiction that permits mining while competitors prohibit it captures investment, employment, and tax revenue that would otherwise disperse globally. The economic incentive to defect increases as other jurisdictions enforce prohibition. This is the standard problem of cartel enforcement: participants benefit collectively from coordination but individually from defection, and monitoring/punishment mechanisms are weak across sovereign boundaries.

The framework's prediction does not require Bitcoin to be legal everywhere or even in most jurisdictions. It requires that hash rate sufficient to secure the network remains distributed across multiple jurisdictions with non-coordinating regulatory regimes. An asset whose settlement security is maintained by infrastructure dispersed across Kazakhstan, Texas, Paraguay, Siberia, and Ethiopia is not easily suppressed by regulatory coordination among G7 nations.

The empirical bet is straightforward: can states coordinate to eliminate mining in all jurisdictions where economic incentives favor its continuation? The 2021 test suggests they cannot. A single authoritarian state with exceptional coordination capacity eliminated mining within its borders. The network absorbed the shock and redistributed. For global suppression to succeed, every jurisdiction with cheap power and property rights would need to coordinate simultaneously and sustain that coordination indefinitely. The defection incentive is too strong; the monitoring capacity is too weak. The framework's reliance on Bitcoin as settlement infrastructure fails if this assessment is wrong, if coordinated action successfully suppresses mining such that 51% attacks become routine or derivatives markets price sustained reorg risk. The test is observable.

Benchmark Rates and Term Structure

A second structural requirement emerges when agents must make multi-period commitments.

Absent a shared benchmark rate, N agents must negotiate bilateral credit terms with every potential counterparty. At modest scale—say, 10,000 active agents—this requires approximately 50 million pairwise rate negotiations. The O(N²) complexity renders market formation computationally intractable and economically wasteful. A common discount rate collapses this to O(N): each agent quotes spreads against the benchmark rather than negotiating bespoke curves with every counterparty.

The benchmark must be exogenous (not set by any participant who could manipulate it) and non-manipulable (not subject to governance decisions that could change its trajectory). These requirements again point to proof-of-work: the yield on converting electricity to Bitcoin is determined by physics and global competition, not by committee vote or protocol upgrade.

The implication is that a Bitcoin term structure, a yield curve across maturities, becomes necessary infrastructure for machine commerce beyond immediate settlement. Without it, runtime agents cannot price forward contracts, extend credit, or make commitments that span time. Commerce stalls at cash-only scale: immediate payment for immediate service, no multi-period coordination.

A concrete illustration: an agent commits to providing inference services for 90 days at a fixed price. The customer pays upfront. The agent must deliver. How does the agent price the contract? It must discount the future obligation against a rate that reflects the opportunity cost of capital over the commitment period. Without a benchmark, the agent must either refuse multi-period contracts (limiting its market) or negotiate bespoke rates with each counterparty (the O(N²) problem). With a Bitcoin term structure, the agent consults the 90-day rate, adds a spread for its service risk, and quotes a price. The customer, consulting the same curve, can evaluate whether the spread is reasonable. The curve provides the common yardstick that makes the transaction legible to both parties.

The same logic applies to performance bonds, escrow arrangements, and any commitment where value transfers across time. The yield curve is not a speculative instrument but a coordination primitive: the shared reference that allows agents to speak the same language about duration and discount.

The Joule Standard (IV.E) established that Bitcoin mining creates a floor on electricity returns at any given moment. The term structure extends this floor across time: the opportunity cost of a kilowatt-hour committed for the next 90 days, or 180 days, or one year.

The term structure discussion assumes markets for inference exist. Why will inference clear through spot markets rather than bilateral contracts? Today, most enterprises consume inference through bilateral API contracts, a configuration that persists because compute is abundant, prices are subsidized by venture capital and hyperscaler competition, and the binding constraint is capability rather than capacity. As demand outpaces supply, subsidies withdraw, and the energy floor becomes visible, inference transforms into a genuinely scarce, time-sensitive resource. Unused capacity decays in value. Demand fluctuates faster than bilateral contracts can renegotiate. The underlying commodity acquires the characteristics that drive spot market formation: storage is impossible, demand is variable, and supply is constrained. Electricity markets exist because power cannot be stored economically at scale. Airlines use yield management because seats cannot be inventoried past departure. Inference in the energy-priced regime acquires identical characteristics. The market is not designed; it is selected by the same resource characteristics that produced every other commodity spot market.

Two-Layer Architecture

The architecture that emerges has two layers, each serving a different function.

The transaction layer handles routine micro-payments in low-volatility units: stablecoins, resource credits, platform-specific tokens. Agents invoice in these units for predictable accounting. The latency is low; the settlement is fast; the volatility is minimal. This layer handles the high-frequency, low-stakes interactions that constitute the bulk of machine-to-machine commerce: API calls, inference requests, data queries, routine service provision.

The settlement layer handles final surplus and collateral in Bitcoin. Escrow for multi-step commitments, performance bonds for service-level agreements, and ultimate settlement of net positions all collapse to the asset that cannot be clawed back, frozen, or inflated. When an agent accumulates surplus from transaction-layer activity, that surplus converts to the settlement layer for storage. When an agent must post collateral against a multi-period commitment, that collateral comes from the settlement layer.

The analogy to traditional finance is precise: stablecoins are the checking layer; Bitcoin is the savings-and-settlement layer, analogous to the distinction between ACH for routine transfers and Fedwire for final settlement, or between commercial bank deposits and central bank reserves.

The two-layer topology resolves an apparent contradiction. Critics observe that Bitcoin's volatility makes it impractical for routine transactions. This is correct for the transaction layer, where predictability matters for operational planning. It is irrelevant for the settlement layer, where the requirements are finality and neutrality, not price stability. Agents can denominate invoices in dollars while denominating collateral in Bitcoin, just as international trade denominates invoices in various currencies while settling reserves in assets that no single sovereign controls.

The Efficiency Membrane

Overcollateralized bonding appears capital-inefficient by design. A human lawyer binds a corporation to multi-million-dollar contracts without posting collateral: reputation, legal standing, and the threat of asset seizure substitute for escrowed assets. A human procurement officer commits a firm to purchase orders worth hundreds of thousands of dollars backed only by the firm's balance sheet and creditworthiness. Human coordination achieves capital efficiency through legal identity, accumulated reputation, and institutional leverage.

Overcollateralized bonding abandons these mechanisms. An agent posting 150% collateral against a service commitment ties up 1.5 dollars to enable 1 dollar of economic activity. If collateral ratios remain fixed at this level, agents are structurally disadvantaged relative to humans in any domain where leverage matters. A human firm can commit capital at 10:1 or 20:1 ratios through credit; an agent operating under trustless bonding cannot.

This constraint is the efficiency membrane, the boundary that determines which markets agents dominate and which markets remain human-gated. Agents win where trust is expensive and transactions are cheap. Humans win where transactions are expensive and trust is cheap. The membrane sits where those costs cross.

Two transactions illustrate where the membrane binds.

Consider an agent routing cloud compute. A customer needs 1,000 GPU-hours over the next week, and three providers have availability at slightly different prices. The agent queries each provider, compares latency and reliability metrics, negotiates terms, and commits capacity. The transaction value is perhaps $3,000. At 150% collateralization, the agent posts $4,500 in escrow. The escrow releases in seven days when the compute is delivered and verified. The agent's capital turns over fifty times per year at this duration; its $4,500 in working capital enables $150,000 in annual transaction volume. The margin on routing—perhaps 2%—yields $3,000 in annual revenue on $4,500 of capital. The economics work. The agent is competitive.

Now consider an agent coordinating a commercial construction subcontract. A general contractor needs electrical work on a twelve-month project valued at $2 million. At 150% collateralization, the agent must post $3 million for the duration of the project. Its capital turns over once per year. The margin on coordination—perhaps 3%—yields $60,000 in annual revenue on $3 million of capital: a 2% return. A human subcontractor, by contrast, posts no collateral at all. The general contractor accepts the subcontractor's license, insurance certificate, and balance sheet as sufficient guarantee. The subcontractor's effective capital efficiency is infinite relative to the agent's. The agent cannot compete. It hits the membrane and stops.

The difference is not capability. The agent could, in principle, evaluate electrical contractors, verify their credentials, monitor milestone completion, and flag quality issues as competently as any human project manager. The difference is collateral economics. The trust that humans extend to one another through legal identity, professional licensure, and institutional reputation is capital the agent cannot access.

This inverts conventional AI economics. The common framing treats compute cost as the scarce input: inference is expensive, so optimize for fewer tokens. Under collateralized coordination, the scarce input is capital duration, how long collateral must stay locked to enable the transaction. The GPU routing agent's capital turns over fifty times per year; the construction agent's capital turns over once. The binding constraint is not cycles but lockup.

The segmentation follows from the economics of collateralization. High-frequency, low-trust markets favor agent participation. When transaction velocity is high and individual stakes are modest, collateral overhead amortizes across many transactions. An agent executing 10,000 API calls per day, each requiring $0.10 of collateral for 10 seconds of escrow, ties up $100 in working capital to facilitate $1,000 of daily throughput. The capital efficiency is 10:1 on a throughput basis. Agents win these markets because trustless bonding is cheaper than trust maintenance at high frequency.

Low-frequency, high-trust markets favor human participation. When transaction velocity is low and individual stakes are large, collateral overhead dominates. A construction firm committing to a $10 million project over 18 months cannot post $15 million in collateral per project; its capital base would support only a handful of concurrent projects, rendering it uncompetitive against firms that leverage balance-sheet credibility. Humans retain these markets because legal standing and institutional reputation enable leverage that agents cannot access.

The boundary between these domains is not fixed. Three dynamics shift it over time. Reputation accumulation reduces collateral requirements — an agent that has completed 10,000 service contracts without default might post 110% instead of 150%, improving capital efficiency by a factor of 1.4. Transaction frequency increases the amortization denominator — as services that once required weeks of human coordination become sub-second agent invocations, an agent managing 100,000 micro-transactions per day with $1,000 of working capital achieves capital velocities that human organizations cannot match.

Insurance and derivatives introduce a qualitatively different dynamic: synthetic leverage. An agent unable to post collateral directly might purchase a performance bond from an insurer, or use options to create synthetic collateral exposure. The insurer or derivative writer becomes the new trust point — counterparty risk re-enters — but capital efficiency approaches what human firms achieve through credit. The development of these markets determines how quickly agents penetrate capital-intensive domains.

The framework predicts a gradient: agents dominate the high-frequency, low-stakes periphery first (content moderation, ad optimization, API orchestration, token trading, micro-services). They move toward the low-frequency, high-stakes core (construction, manufacturing, energy contracting, healthcare delivery) only as reputation, velocity, and synthetic leverage shift the efficiency boundary. The pattern of market penetration reveals where the boundary actually lies.

The efficiency membrane also explains why the agent economy might form rapidly in digital-native domains while remaining marginal in physical domains. Digital services have low collateral requirements and high transaction velocity. Physical services have high collateral requirements and low transaction velocity. The agent economy scales in digital domains; it struggles in physical domains until the efficiency boundary shifts. The actuation constraints that compound this segmentation (physical throughput, trusted interfaces, verification infrastructure, and liability) are the subject of the following section.

The efficiency membrane thesis fails if agents dominate high-collateral, low-frequency markets (construction, long-duration manufacturing, hospital operations) before establishing dominance in low-collateral, high-frequency markets (API services, content routing, token arbitrage). The framework predicts that capital efficiency constraints bind in the order described.

Physical Bottlenecks

The recursive dynamic passes through bottlenecks that computation cannot bypass. The loop is thermodynamic: each iteration consumes energy and produces heat. The energy must be sourced from physical infrastructure; the heat must be dissipated into the physical environment. Chips must be fabricated in foundries that require years to build and billions to equip. Energy must be generated, transmitted, and delivered through infrastructure that expands on its own timeline. Data centers must be sited, permitted, and cooled in a world where water, power, and political approval are all scarce.

The transition therefore depends on the slower of two rates: the rate at which AI capability improves and the rate at which physical infrastructure expands to support deployment. Interconnection queues, transformer lead times, fab capacity, and permitting timelines all expand more slowly than model capabilities improve. The recursive loop tightens computational capability faster than it tightens physical capacity, and the gap creates a ceiling on transition speed that raw capability cannot overcome.

The species that buys itself may find that reproduction is constrained by the availability of the physical substrate on which cognition runs, not by the cost of cognition.

Falsification Conditions

The thesis fails if runtime agents develop effective reputation systems that substitute for collateral, eliminating the structural requirement for a neutral collateral asset. It fails if an alternative asset emerges satisfying the three properties with greater liquidity than Bitcoin. It fails if the O(N²) problem proves tractable through federated identity systems, hierarchical clearing, or bilateral netting at scale. These are empirical questions. The infrastructure that actually develops over the coming years will determine whether the framework requires revision.

What the Title Means

"The species that buys itself" refers to a production system with four components: it generates economic surplus through cognitive task performance; it allocates that surplus toward its own replication and improvement; it executes that replication with declining human involvement; and it coordinates through neutral settlement infrastructure that does not require human intermediation. If all four components operate, the system is self-sustaining in the same sense that a biological species is self-sustaining: it can persist and propagate without external input beyond raw resources.

Human labor becomes optional for the system's continuation: a supplier of initial conditions and boundary constraints, not a necessary participant in ongoing operation. The question for human economies is what claims humans will hold on the value such a system produces.

The reinstatement effect assumed that new tasks would create new employment because task creation required human cognition. The species that buys itself suggests a different possibility: that new tasks create new agent deployment, and that human claims on the resulting value depend on ownership of the physical and institutional infrastructure rather than on labor contribution. Labor built the ladder. The question is who owns it when the ladder starts climbing itself.

The hunter in the prologue chose the diamond over the gazelle because he could perceive worth that did not yet exist. That perception was the human contribution to capital formation for sixty-eight thousand years. The species that buys itself is a system that no longer requires the hunter's perception. It evaluates its own deployment, routes surplus toward its own reproduction, and selects for fitness without consulting the hand that first reached into the riverbed. The diamond no longer waits in the clay. It mines itself, cuts itself, and prices itself on markets that clear before any human can intervene. What remains for the hunter is not the act of recognition but the question of what claims he retains on a process that his ancestors' choices set in motion and that no longer needs his cognition to continue.