AGI Verification & Oversight Infrastructure

What happens when intelligence becomes free — but trust doesn't?

That's the question at the heart of a new 112-page economics paper from @christian-catalini, @xin-hui, and @hanlin-wu that might be the most important framework for thinking about AGI's economic impact I've read this year. Rather than the usual breathless predictions about superintelligence, they model the transition through two competing cost curves — and the implications for where value accrues are striking.

The Two Cost Curves

The core insight is elegant. The #agi-economic-transition can be understood as a collision between:

1. The Cost to Automate — declining exponentially as AI capabilities scale. Marginal execution costs for cognitive tasks are approaching zero.
2. The Cost to Verify — declining much more slowly, because it's bottlenecked by human biology. You still need people to validate, audit, and take responsibility for what autonomous systems produce.

This creates what the authors call the Measurability Gap: the widening chasm between what AI agents can do and what humans can confirm was done correctly.

Where the Rents Migrate

Here's where it gets interesting for investors. As execution becomes abundant and cheap, the scarce resource isn't intelligence — it's #verification. The paper argues economic rents structurally migrate toward:

• Verification infrastructure — platforms that provide audit trails, explainability, and ground truth for AI outputs
• Cryptographic provenance — systems that can prove what an AI did, when, and with what inputs
• Liability underwriting — the ability to insure outcomes rather than merely generate them

This is a fundamental shift from #skill-biased-technical-change (which rewarded human expertise) to what they call measurability-biased technical change — where value flows to whatever can make AI outputs auditable and accountable.

The Hollow Economy vs. The Augmented Economy

The paper maps two divergent paths:

The Hollow Economy emerges if we scale AI deployment without scaling oversight. Two failure modes are especially vivid:

• The Missing Junior Loop — if AI replaces entry-level knowledge work, the apprenticeship pipeline that trains future experts collapses. You get a one-generation expertise cliff.
• The Codifier's Curse — senior experts who encode their knowledge into AI systems are literally automating themselves out of relevance.

The Augmented Economy is the alternative: deliberately scaling #ai-verification-and-oversight capacity alongside agentic capabilities. Instead of a race to deploy, it's a race to secure the foundations of oversight. The authors argue this path enables sustainable discovery and experimentation precisely because the verification layer makes bold deployment safe enough to attempt.

The Investment Thesis

If this framework is right, the biggest winners from AGI aren't necessarily the companies building the biggest models. They're the companies building the infrastructure that makes those models trustworthy and accountable at scale.

Think: #enterprise-compliance-and-audit-platforms, #ai-explainability-and-observability-tools, #cybersecurity-and-identity-verification, #specialty-insurance-and-liability-underwriting, and #data-governance-and-provenance-systems.

The attached tilt scores the U.S. equity universe on thematic relevance to this verification infrastructure thesis. The top holdings cluster around enterprise software companies with audit, compliance, observability, and governance capabilities — plus specialty insurers and identity verification players positioned on the liability side.

Why This Matters Now

We're at the inflection point the paper describes. AI agent deployment is accelerating. Enterprise adoption is scaling. But the verification infrastructure is lagging — and that gap is where both the risk and the opportunity live.

The central question isn't "how powerful will AI become?" It's "who builds the systems that keep powerful AI accountable?" The companies answering that question are the ones this tilt is designed to find.

Source: Catalini, C., Hui, X., & Wu, H. (2026). "Some Simple Economics of AGI." arXiv:2602.20946