Christian Catalini

The economics of AI

The Economics of AI: Verification as the New Scarcity

When intelligence becomes cheap, verification becomes the binding constraint — a unified economic theory of the AGI transition.

Updated July 2026

For three hundred thousand years, human cognition was the main engine of progress on Earth. That is now changing. We are decoupling intelligence from biology — bootstrapping a second, alien form of cognition trained not by the friction of survival, but by compressing, predicting, and recombining the sum total of digitized human thought. As the marginal cost of generating an answer falls toward zero, the binding constraint on the economy shifts: the scarce resource is no longer intelligence — it is trust.

“Some Simple Economics of AGI” presents a unified economic theory of this transition. Its central claim is that the dividing line in the agentic economy is no longer routine versus non-routine work, but measurable versus non-measurable. Any task that can be reduced to a metric can be industrialized — regardless of the prestige or complexity once attached to it. What stays scarce, and therefore valuable, is the human capacity to verify, underwrite, and stand behind what machines produce.

From scarce intelligence to scarce trust

Standard models treat AI as a substitute for labor or a complement to human skill — a “bicycle for the mind” in which human judgment is the essential complement, and therefore the primary limit on value. But as agents acquire the ability to plan, act, and learn autonomously, they internalize the very judgment those models reserve for humans. In an agentic economy — where autonomous agents operate with broad agency rather than narrow instruction — the binding constraint on growth is no longer the scarcity of intelligence, but the scarcity of trust. Can we verify that an agent executed the right action safely? Can we know what it did, why it did it, and whether its output deserves to be relied upon?

This also explains the pattern of early adoption. The first widely used systems clustered around domains where humans can verify outputs in seconds — chat, images, short bursts of code — because the cost to verify was negligible relative to the value created. As agents take on longer-horizon, higher-stakes tasks, verification becomes the scarce resource, and competitive pressure pushes the market toward unverified deployment — letting hidden systemic risk accumulate in the gap between what agents can execute and what humans can confidently validate.

The Measurability Gap

The automation frontier is not a static technological boundary; it is the collision of two racing cost curves. The Cost to Automate is driven relentlessly downward by compute and accumulated knowledge. The Cost to Verify is bounded by the biological limits of human cognition and feedback latency. The widening distance between them — the Measurability Gap — determines the verifiable share of the economy: the threshold separating genuinely productive agentic execution from merely plausible output.

This geometry yields a strict typology of work in four regimes: the Safe Industrial Zone (cheap to automate, affordable to verify), the Runaway Risk Zone (cheap to automate, unaffordable to verify), the Human Artisan Zone (hard to automate, verifiable), and the Pure Tacit Zone (neither automatable nor verifiable). The structural danger lives in the long-horizon tail — a growing domain where automation is virtually free, yet verification remains fundamentally infeasible.

Why human oversight quietly erodes

The “human-in-the-loop” equilibrium is dynamically unstable, eroded simultaneously from below and from within. From below, the Missing Junior Loop severs the apprenticeship pipeline: human expertise is a stock accumulated through the friction of routine execution, so automating entry-level cognitive work destroys our capacity to build future verifiers. From within, the Codifier’s Curse hollows out existing expertise, as senior professionals rationally mine their own tacit knowledge to create the proprietary ground truth that trains their replacements.

Together these forces push firms toward a fragile “sandwich” topology: humans define intent, machines execute, and a shrinking, over-leveraged layer of humans underwrites outcomes. Economic rents shift accordingly — away from those with the deepest formal education and toward those who operate in unmeasured domains and those willing to absorb the liability of standing behind machine output. Tasks that are highly complex but purely measurable face rapid wage compression toward the marginal cost of compute.

The Hollow Economy

Driven by the imperative to scale, unverified deployment becomes privately rational. Agents consume real resources to produce output that satisfies measurable proxies while violating unmeasured intent — counterfeit utility. As this hidden debt accumulates, the system drifts toward a Hollow Economy: explosive measured activity, but collapsing realized utility and hollowed-out human control. The tempting shortcut of using AI to verify AI manufactures false confidence, because the agent and its synthetic auditor share the same priors and the same blind spots — synthetic validation becomes abundant while true ground truth stays scarce.

The model identifies three countervailing forces. Observability compresses high-dimensional agent behavior into signals experts can reliably process, and — coupled with cryptographic provenance — makes agentic actions mathematically auditable. Accelerated Mastery rebuilds experience stocks through human augmentation and AI-driven synthetic practice when traditional apprenticeship collapses. Graceful Degradation invests in base alignment so that when oversight falters inside the Measurability Gap, systems revert to safe baseline policies rather than optimizing aggressively in unverifiable regimes.

What stays human

Human comparative advantage distills into three roles, each powerful and each precarious: verifying and underwriting agentic labor, using embodied experience to validate high-stakes outcomes; intent arbitration, resolving the value conflicts that objective functions cannot capture; and operating at the Knightian frontier where the map still fails. Mirroring this, the economy bifurcates. In the measurable domain, the price of execution races toward the marginal cost of compute. Capital instead gravitates toward what is not yet measurable — deep tech, long-horizon R&D, and the status economy anchored in human consensus and meaning — and toward the verifiable economy of ground truth, cryptographic provenance, and liability.

The dominant revenue model shifts from monetizing software access (Software-as-a-Service) to monetizing outcomes (“Software-as-Labor”), and firms are valued primarily on their capacity to absorb tail risk through “Liability-as-a-Service.” Execution is now infinitely scalable; the legal and financial capacity to absorb its inevitable failures is the new bottleneck.

The choice ahead

Beneath the market failure lies an irreversible structural divergence. Do we deliberately govern agents as scalable levers for human intent, or allow them to operate as a successor species that eventually displaces us as the economic apex predator? Our institutions will decide whether we build an Augmented Economy — systematically scaling human cognition to remain peers with our creation — or tacitly accept a succession event, ceding stewardship of a civilization we no longer understand.

The paper translates this into a practical playbook. Individuals should rebuild expertise faster than the market rate and move up the intent-and-underwriting stack. Companies should anchor scale to verified throughput, treating verification as a core production technology. Investors should pivot from funding commoditized execution toward what is not yet measurable and the trust complements that make deployment insurable. Policymakers should treat verification infrastructure and ground truth as public goods, enforcing the preconditions of insurability. The frontier of automation is not an exogenous force — it is a reflection of our institutional choices.

Read & cite the work

See all digital assets & AI writing →

Related: essays & columns · papers · all ideas