Christian Catalini, co-founder of LightSpark and founder of the MIT Crypto Economics Lab, joins Eddie Lazarin, Chief Technology Officer at A16Z, to discuss Catalini's new economics paper analyzing the relationship between AI and human labor.

The conversation explores how Some Simple Economics of AGI builds on Catalini's earlier Simple Economics of the Blockchain work from 2013, applying rigorous economic frameworks to understand automation, verification, and the future of work in an AI-dominated economy.

They examine the transition from AI as a tool to AI as a coworker, the emergence of new economic roles like directors and verifiers, and how crypto infrastructure becomes essential for managing verification, identity, and coordination in a fragmented economy of AI-powered micro-enterprises.

The Economics of Automation vs Verification

Some Simple Economics of AGI establishes that automation costs are plummeting while verification costs decline more slowly, creating a fundamental economic gap that defines future human roles.

Verification encompasses applying embedded human measurement - the unique combination of experiences and knowledge accumulated from birth to professional expertise that exists only in individual brains.

AI agents excel at recombining existing knowledge but struggle with unmeasured domains, creating opportunities for humans who have seen 'out-of-distribution examples' through lived experience.

Three Categories of Future Work

Directors set intent and steer AI systems, similar to movie directors - entrepreneurs, writers, and strategic leaders who define vision and course-correct when systems drift.

Verifiers apply domain expertise to ensure AI output meets standards, but face the 'codifier's curse' - as they create labels for verification, they enable their own displacement.

Meaning makers coordinate social consensus around unmeasurable domains like art, culture, and crypto networks where multiple equilibria exist and coordination matters more than optimization.

The 'AI sandwich' structure emerges: one director at the top, a swarm of agents in the middle, and an army of top verifiers at the bottom ensuring quality output.

The Transformation of Software Engineering

December 2025 marked a qualitative shift where AI agents can perform long-running tasks autonomously, creating the sensation of working with a coworker rather than a micromanaged tool - Eddie.

Software engineering is being unbundled: writing code becomes 'vanishingly small' while verification - ensuring code provides customer value and serves business goals - dominates the work.

Engineers discover they were simultaneously producing code and performing nuanced verification work they hadn't fully recognized, including iterative correction, feedback gathering, and business alignment.

The Hollow vs Augmented Economy

The hollow economy emerges when companies eliminate junior workers, top verifiers shrink in number, and alignment drift occurs as AI systems develop unintended preferences over time.

The augmented economy leverages AI for accelerated mastery, helping individuals discover their true aptitudes and advance rapidly rather than following standardized curricula.

The transition requires investments in verification tooling, human augmentation, and safety R&D to avoid systemic risks from unverified AI deployment at scale.

Crypto's Role in AI Verification

Blockchain provides deterministic guardrails for probabilistic AI systems, enabling agents to buy and sell resources while maintaining cryptographic proof of transactions and identity.

On-chain payment systems give AI agents better signals than legacy APIs - 'the agent had a much better understanding of what was happening' with full transaction context visible - Christian.

Verification grid network effects become the new moat: companies with proprietary failure data can underwrite risk better and deliver safer products at lower cost.

As AI enables massive fragmentation into micro-enterprises, blockchain networks provide credibly neutral infrastructure for coordination, reputation, and complex inter-company transactions.

The New Entrepreneurial Playbook

Young entrepreneurs should convince parents to fund compute experiments: 'can I spend $5,000 of compute productively?' and learn to guide swarms of machines - Eddie.

The one-person billion-dollar startup becomes feasible when you can have 'multiple employees for $200 a month' through AI agents, though they're 'a little weird' and require new management skills.

'The apprenticeship might be dead, but the real work is beginning' - the groundwork phase disappears, enabling direct access to complex domains like hardware for curious individuals.

Companies like Hyperliquid and Uniswap demonstrate extreme value creation with fewer than 20 employees, presaging single-person or duo founders reaching live products in weeks or months.

Systemic Risks and the Trojan Horse Effect

The 'Trojan horse' effect occurs when rational actors ship AI-generated work containing subtle errors because full verification is humanly impossible at AI's throughput scale.

Companies face tension between long-term sustainability and short-term growth pressure, potentially accumulating systemic risk as they accelerate deployment without adequate verification infrastructure.

11 Labs pioneered AI agent insurance, signaling the emergence of 'labor as software' models where liability and insurance become critical for AI deployment in production.

Open source AI deployment, despite risks, provides early signals of system abuse and helps build countermeasures against harmful preferences that emerge as side effects of training.