Cambridge, Mass., June 11, 2026 (GLOBE NEWSWIRE) -- As artificial intelligence capabilities advance, economic and technological progress is being decoupled from human biology for the first time in history. In a new working paper, “Some Simple Economics of AGI,” MIT Sloan School of Management researcher Christian Catalini found that the rise of agentic AI will force individuals, companies, and policymakers alike to grapple with the challenge of steering this new force toward verifiable, high-quality outputs that preserve an economy based on human oversight.
What does the collision of two curves, automation and human verification, look like?
Historically, automation was limited to repetitive, manual tasks. Today, AI can generate hypotheses, design experiments, and navigate complex problems once considered the exclusive domain of highly educated, high-wage workers. Catalini models this transition as a race between two cost curves: the nosediving cost of automation and the stubborn plateau of human verification, limited by biology and time. This asymmetry creates a "Measurability Gap" in which AI agents execute tasks faster than humans can audit them.
“When getting an answer to a question costs almost nothing, the value of human work shifts to knowing what to ask and being certain the result is correct,” Catalini explained. “We are moving from a world where we were valued for what we could create, to one where we are valued for our ability to steer and stand behind what is created.”
What is the Risk of a “Hollow Economy”?
The mismatch between the speed of AI and humans’ ability to direct and verify its outputs creates serious systemic risks — one of which Catalini refers to as the “Trojan Horse” externality. As deployment of AI becomes cheaper, firms will be incentivized to use AI agents that satisfy visible metrics (like speed or volume) without proper (human) verification. Over time, these practices will lead to accumulating, unverified data in the systems, degrading the quality of outputs while consuming real resources and creating what Catalini calls a “hollow economy.”
Here is a playbook for the augmented economy
That hollow economy, however, is not inevitable. Catalini argues that the proper approach to the coming economic shift can instead deliver an “augmented economy,” in which close human oversight and guidance drives an era of social, scientific, and economic advancement.
To get there, he said, individuals, companies, investors and policymakers must stop treating AI as a faster version of a human and start treating it as a volatile agent that requires careful direction.
For Individuals: From "Doing" to "Underwriting"
The Augmented Economy inverts the traditional bargain between talent and resources. Since intelligence is now an abundant commodity, the nature of human work must shift. Key takeaways for individuals include:
- Accelerate learning and development: Entry-level roles are most at risk in the AI shift — but savvy individuals can use AI as a “sandbox” to learn and practice new skills, potentially compressing years of work experience into just months.
- Move up the new value chain: In many industries, professional success will depend on the capacity to successfully steer agentic AI and verify its outputs, rather than the ability to manually verify those outputs.
- Focus on Connection: In a world of automated output, increased value will also migrate to roles anchored in status, human connection, and social coordination — making “human-made” a value signal.
For Companies: Verification as a Competitive Advantage
In an economy where raw output is commoditized, competitive advantage shifts to the talent and data capable of reliably certifying agentic systems. Verification is no longer a mere compliance function; it is a primary product.
- Adopt the "sandwich topology": Organizations can thrive with a three-layer model: human intent (defining goals) → machine execution (high-volume production) → human underwriting (expert certification).
- Invest in observability: Companies should prioritize creating and securing "verification-grade ground truth"— especially data on edge cases and failures — to increase the quality, rather than the quantity, of AI outputs.
For Investors: Capitalizing the Unmeasurable
The paradigm shift for capital is to move away from funding commoditized execution and toward the “unmeasurable” areas of economic growth where benchmarks and short-term feedback loops do not yet exist.
- Target the frontier: Capital should flow toward deep tech, long-horizon R&D, and the infrastructure of the augmented economy.
- Re-ground valuations: As revenue models shift from selling software to selling outcomes, firms should be valued on their ability to absorb risk and reliably warrant autonomous outcomes. Software transitions from software-as-a-service to liability-as-a-service, with companies taking on both automation and risk underwriting.
For Policymakers: Protecting Against Systemic Risk
Policymakers must act quickly to protect against the “Trojan Horse” externality: firms capturing private gains from unverified agents, while increasing risk across the broader system and society.
- Truth as a public good: Policymakers must treat verification infrastructure and ground truth as foundational public goods. This includes implementing liability regimes that internalize tail risks to ensure safe scaling isn't outcompeted by reckless deployment.
- Universal access: If the right safeguards and reward structures are put in place, this leap forward in technology could lead to the expansion of public services such as individualized healthcare and education, delivered at marginal costs that make universal access economically feasible.
“Ultimately, our goal is to ensure that humanity remains the architect of our economy and society,” Catalini said. “We have built a massive new capacity for execution; now, we must scale our capacity for oversight to match it. Only by staying responsible for what we create can we ensure this technology continues to improve lives.”
Attachment
Casey Bayer MIT Sloan School of Management 914.584.9095 bayerc@mit.edu
