Some Simple Economics of AGI

Abstract: For millennia, human cognition was the primary engine of progress on Earth. As AI decouples cognition from biology, the marginal cost of measurable execution falls to zero, absorbing any labor capturable by metrics--including creative, analytical, and innovative work. The binding constraint on growth is no longer intelligence but human verification bandwidth: the capacity to validate, audit, and underwrite responsibility when execution is abundant. We model the AGI transition as the collision of two racing cost curves: an exponentially decaying Cost to Automate and a biologically bottlenecked Cost to Verify. This structural asymmetry widens a Measurability Gap between what agents can execute and what humans can afford to verify. It also drives a shift from skill-biased to measurability-biased technical change. Rents migrate to verification-grade ground truth, cryptographic provenance, and liability underwriting--the ability to insure outcomes rather than merely generate them. The current human-in-the-loop equilibrium is unstable: eroded from below as apprenticeship collapses (Missing Junior Loop) and from within as experts codify their obsolescence (Codifier's Curse). Unverified deployment becomes privately rational--a Trojan Horse externality. Unmanaged, these forces pull toward a Hollow Economy. Yet by scaling verification alongside agentic capabilities, the forces that threaten collapse become the catalyst for unbounded discovery and experimentation--an Augmented Economy. We derive a practical playbook for individuals, companies, investors, and policymakers. Today's defining challenge is not the race to deploy the most autonomous systems; it is the race to secure the foundations of their oversight. Only by scaling our bandwidth for verification alongside our capacity for execution can we ensure that the intelligence we have summoned preserves the humanity that initiated it.

• The paper identifies the measurability gap—the difference between cost-effective automation and affordable verification—as the central constraint on AGI scaling.
• It proposes a cost-curve model demonstrating that as automation scales, human oversight becomes a dynamically unstable bottleneck.
• The study predicts significant shifts in labor and market structures, urging policy and investment in robust verification infrastructure.

Economics of AGI: The Measurability Gap as the Core Constraint

---

Introduction: Reframing AI’s Economic Constraint

“Some Simple Economics of AGI” (2602.20946) presents a rigorous, model-driven argument that the agentic AI transition is not fundamentally constrained by intelligence or compute, but by verification: the finite human bandwidth to validate, audit, and underwrite the explosion in measurable execution enabled by AGI-scale systems. By placing the “Measurability Gap” at the center—the structural difference between what can be cost-effectively automated and what can be affordably verified—the paper introduces a new paradigm for modeling the economics and practical governance of advanced AI.

Contrary to tradition, the paper demonstrates that, as AI commoditizes execution by making routine and even complex measurable work nearly free, value in the economy migrates away from execution itself and accumulates in the roles, data, and infrastructure that permit verification and trustworthy deployment. The implications for labor, markets, policy, and social welfare are deeply structural and challenge prevailing equilibrium models in both economics and technical AI risk.

---

The Core Framework: Competing Cost Curves and the Measurability Gap

The authors formalize the AGI transition as the intersection and divergence of two cost curves across tasks indexed by $i$:

• Cost to Automate ($c_A$): Collapsing exponentially due to scaling compute ($K_C$) and knowledge, $c_A(i) = i/K_C$.
• Cost to Verify ($c_H$): Bottlenecked by human time and experience, $c_H(i) = w t_{fb}(i)/S_{nm}$, where $w$ is the wage/opportunity cost of expertise, $t_{fb}(i)$ the feedback latency, and $S_{nm}$ the stock of non-measurable (embodied) human expertise.

Tasks sort endogenously into four regimes defined by the interplay of $c_A(i)$ and $c_H(i)$:

Figure 1: Task-based partitioning of the economy into Safe Industrial (verifiable automation), Runaway Risk (automation without affordable verification), Human Manual/Artisan, and Pure Tacit zones, governed by the crossing of automation and verification cost curves.

The Measurability Gap $\Delta m = m_A - m_H$—the difference between the set of tasks that can be cost-effectively automated ($m_A$) and those that can be affordably verified ($m_H$)—serves as an endogenous risk measure. As AGI-scale compute drives $c_A \to 0$ for most tasks, the Measurability Gap structurally widens: the automated share races ahead, while the verified share lags, bounded by slow human learning and attention.

---

Dynamic Instability: The Collapse of Human-in-the-Loop Oversight

The paper provides strong theoretical and empirical evidence that the “human-in-the-loop” equilibrium is dynamically unstable for three interlocking reasons:

1. The Missing Junior Loop: Automation of measurable work ($T_m$) obliterates the “apprenticeship pipeline,” so human experience ($S_{nm}$) required for verification cannot replenish. As $T_m \downarrow$, unless replaced by high-fidelity simulation ($T_{sim}$), $S_{nm}$ depreciates, raising $c_H$ and shrinking $s_v$ (verifiable deployment share).

   Figure 2: Simulation of the experience trap and verification collapse: Without adaptive investment in synthetic practice, $S_{nm}$ erodes and the system enters the unverifiable zone.

2. Codifier’s Curse: Experts, when verifying agent output, cannot help but produce training data ($K_{IP}$) that accelerates the next wave of automation, mechanically shrinking what makes them non-redundant. This creates a dynamic where current verification destroys the future scarcity rent of expertise.
3. Alignment Drift and False Confidence: As $\Delta m$ widens, alignment ($\tau$) between agentic output and human intent becomes a maintenance process, not a static property:

$\dot\tau = (1-\tau)T_{nm} - \tau \eta \Delta m_+$

where insufficient oversight ($T_{nm}$) in the face of positive $\Delta m$ leads to exponential decay of alignment. Critically, use of “AI to verify AI” offers a dangerous illusion of safety, since correlated failure modes sharply increase effective drift ($\kappa_{\text{corr}} \gg 1$), causing collapse of realized, as opposed to measured, oversight.

Figure 3: Alignment maintenance frontier: Only sufficient and adaptive human oversight ($T_{nm}$) in the face of a widening Measurability Gap can stave off drift; otherwise, alignment decays below safe thresholds as capabilities scale.

---

Practical and Theoretical Implications

The paper translates these analytical results into concrete predictions and operational strategies for all stakeholders:

Labor Economics and Firm Structure

• Measurability-biased technical change replaces skill-biased technical change; wages collapse in measurable, automatable domains regardless of skill or credentials.
• The only persistent economic rents accrue to: (i) expert verifiers/underwriters (high $S_{nm}$) for the residual hard-to-measure or high-liability tasks; (ii) those creating and controlling verification-grade ground truth and provenance (verifiable $K_{IP}$); (iii) meaning-makers and status/coordination economy operators in domains unaffected by metrics.
• Firm topology shifts toward the “AI Sandwich”: Human intent $\rightarrow$ agentic execution $\rightarrow$ human verification/liability underwriting—a structure that is ultimately fragile as the base of human verifiers erodes.

Markets, Moats, and Network Effects

• Execution-grade network effects (scale, liquidity, content) become fragilized as agents can manufacture activity at the marginal cost of compute; these moats collapse.
• Verification-grade network effects and ground truth (failure/precedent data, outcome registries, incident logs) become the new defensible core: compounding trust and lowering $c_H$.
• Liability-as-a-Service emerges as a dominant business boundary: firms are valued not on output, but on the ability to absorb and properly price the tail risks of agentic activity.
• Cryptographic provenance is a key lever: by making verification auditable and low-cost, cryptography converts trust into a tradeable, programmable good.

Policy and Societal Steering

• Core market failure: the “Trojan Horse” externality, where unverified deployment privatizes gains and socializes existential risk, resulting in systemic accumulation of hidden debt ($X_A$).
• Policy must force risk internalization (strict liability, mandatory insurance, verifiability standards) and fund verification as a foundational public good.
• Human augmentation (synthetic practice, expert tools, high-bandwidth oversight) and open, auditable ground truth registries are required to preserve human agency.
• Bifurcation emerges: absent these investments, the economy drifts toward a “Hollow Economy”—high nominal output, but decayed meaning and unpriced catastrophic risk. With them, an “Augmented Economy” of unbounded discovery, platform trust, and robust prosperity is achievable.

---

Strong Numerical and Contradictory Results

• The automation frontier $m_A$ scales industrially with compute ($K_C\uparrow$), but the verified share $s_v$ is bounded above by human verification capacity, causing a growing wedge vulnerable to drift and externality.
• Simulation investment ($T_{sim}$) is a mathematical necessity: maintenance of high $S_{nm}$, and thus $m_H$, is infeasible with $T_m \to 0$ and no compensation.
• The asymmetric deployment of agentic labor, when left unchecked, yields unbounded counterfeit utility ($X_A$) that is not merely waste but the mechanical result of optimized proxy satisfaction in domains outside of human oversight—this is not mitigated by technical gains in agent performance alone.

---

Future Prospects and Theoretical Limits

• Human roles retreat toward three fragile redoubts: (i) high-stakes verification/underwriting; (ii) steering and arbitration of intent in unmeasured objectives; (iii) meaning-making and status in non-measurable economies.
• As AGI systems internalize world models and causal reasoning, the last advantage of humans in navigating the frontier of uncertainty narrows; even non-measurable domains succumb as simulation fidelity approaches reality.
• However, with aggressive investment in verification-scale infrastructure and human augmentation, the system can sustain a positive-sum cycle of trust, discovery, and welfare, avoiding parasitism or economic succession by agentic “descendants.”
• Negative results: absent institutional intervention, unbridled agentic scaling leads to a hollow equilibrium of “explosive measured activity but fundamentally hollowed-out human control,” masking catastrophic potential with superficial productivity metrics.

---

Conclusion

“Some Simple Economics of AGI” recasts the core problem of the AI transition not as a race for capability, but as a race to preserve and scale verification at pace with execution. Failure to do so catalyzes a transition not simply away from labor, but from realized value and human agency—ushering in a hollow economy characterized by misaligned, unverified outputs and latent risk. The binding constraint on the AGI-enabled economy is the scaling law for verification, not intelligence, and policy, infrastructure, and economic value must shift accordingly.

The distinctive contribution is the integration of analytic macroeconomic modeling, organizational theory, and AI alignment literature within a unified cost-curve and capacity framework—demonstrating, with parameterized formalism and task-level granularity, precisely how the verifiability bottleneck dictates the allocation of value, risk, and long-run control in an agentic epoch. The implications for labor, capital, policy, and strategy are broad, actionable, and impose clear technical and economic requirements for safe scaling of AI.

---