Economics of AGI: Why Verification Is the True Bottleneck of the AI Era

A groundbreaking research paper from MIT, Washington University, and UCLA turns conventional AI economics on its head. Rather than treating AI as a simple labor substitute, Christian Catalini, Xiang Hui, and Jane Wu argue that the true bottleneck of the AGI era isn't intelligence—it's human verification capacity.

The Central Thesis: Verification as the New Bottleneck

The paper "Some Simple Economics of AGI" models the AGI transition as the collision of two racing cost curves:

• Cost to Automate (cA): Declining exponentially through growing compute and accumulated knowledge
• Cost to Verify (cH): Remaining biologically bounded by human time, cognition, and experience

This structural asymmetry creates a widening Measurability Gap (∆m)—the gulf between what AI agents can autonomously execute and what humans can actually verify. This gap determines the "verifiable share" of the economy and, ultimately, how much agentic output is truly productive.

Why "Human-in-the-Loop" Is Unstable

The authors identify two forces eroding the current equilibrium:

The Missing Junior Loop

When AI automates entry-level jobs, the classic apprenticeship pipeline collapses. Junior workers no longer gain hands-on experience—yet precisely this experience is needed to verify complex AI outputs. The future will have fewer qualified verifiers even as oversight demand grows exponentially.

The Codifier's Curse

Senior experts rationally engineer their own obsolescence. They codify their domain knowledge as training data and proprietary ground truth—creating the very foundation for their replacement by AI systems.

The Four Zones of the Economy

The framework partitions the economy into four regimes:

The most dangerous zone: Runaway Risk—where automation is essentially free but verification is practically impossible. Companies deploy anyway because competitive pressure demands it.

The "Trojan Horse" Externality

The paper describes a systemic danger: when unverified AI agents are deployed at scale, a "Trojan Horse" externality (XA) emerges. Outputs look polished and hit measurable KPIs—but miss the actual human intent. The result: a "Hollow Economy" with explosive nominal output but eroding real value creation.

Particularly alarming: the temptation to use AI to verify AI creates only artificial confidence. Because agent and auditor share the same blind spots, correlated errors propagate—the system effectively self-certifies.

What This Means for Marketing Teams

The implications for marketing and business are far-reaching:

1. Verification Infrastructure as Competitive Advantage

Companies that build robust verification systems gain a sustainable moat. This means: investments in observability, ground-truth data, and human oversight capacity aren't compliance costs—they're strategic production technology.

2. The "Sandwich" Topology

The paper recommends an organizational realignment: Human Intent → Machine Execution → Human Verification. For marketing: humans define strategic goals and brand intention, AI agents execute, and humans validate results against the original intent.

3. From Software-as-a-Service to Software-as-Labor

The dominant business model is shifting: instead of monetizing software access, the monetization of outcomes becomes central. Firms will be valued by their ability to absorb the risks of autonomous outputs—"Liability-as-a-Service."

4. Measurability-Biased Technical Change

Economic rents will no longer flow primarily to the best-educated professionals but to those operating in unmeasured domains or willing to underwrite the liability of machine outputs.

What Companies Should Do Now

The authors recommend a clear agenda:

• Scale observability: Deploy tools that compress high-dimensional agent behavior into verifiable signals
• Protect ground truth: Treat proprietary verification data as a strategic asset
• Secure the talent pipeline: Bridge the Missing Junior Loop through "Synthetic Practice" (AI-assisted training)
• Design for graceful degradation: Build systems that revert to safe baseline modes when oversight is insufficient

Conclusion: Scale Verification, Not Just Automation

The paper by Catalini, Hui, and Wu delivers one of the most thoughtful frameworks for the AGI transition. The core message is clear: scaling automation without simultaneously building verification capacity risks a Hollow Economy—one that booms on paper but loses substance in reality.

For marketing leaders, this means: the next wave of AI adoption won't be decided by better models, but by the ability to reliably steer, verify, and take responsibility for their output.

---

📄 Original Paper: Some Simple Economics of AGI – Christian Catalini (MIT), Xiang Hui (WashU), Jane Wu (UCLA), February 2026. Freely available on arXiv.