The Jobs AI Won't Take-And Why That's Not Good News
A leading economist of automation has upended the standard narrative about AI and employment. Pascual Restrepo, an associate professor at Yale University, argues in a new paper that most human work won't be automated in an era of artificial general intelligence. Not because AI can't do it, but because most jobs simply aren't worth automating.
The distinction matters for how you think about workforce planning. In Restrepo's model, AI will focus computational resources on what he calls "bottleneck" work-tasks essential for economic growth like energy production, infrastructure maintenance, and scientific advancement. Everything else gets left to humans by default, not by design.
Two Categories of Work
Restrepo divides the economy into two types of work. Bottleneck tasks are critical for growth. Supplementary work-hospitality, customer support, arts, design, even academic research-is everything the economy can function without.
AI will automate the bottleneck work. Supplementary work survives because fully replicating it with compute would cost more than it's worth when AI has bigger problems to solve. A barista or novelist keeps their job not because of human magic, but because replacing them makes no economic sense.
Growth and Wages Decouple
Here's the unsettling part: surviving automation is not the same as sharing in economic growth.
Today, when GDP rises, workers typically benefit through wage increases and improved living standards. In the post-AGI economy Restrepo models, that link breaks. Once AI handles all growth-critical work, economic expansion runs on computational resources alone. Human wages become anchored to what it would cost to replace that work with compute-a ceiling that is, long-term, very low.
The paper's starkest finding: labor's share of GDP converges to zero. The computing power of all human brains combined is roughly 10^18 floating-point operations per second. Total computational resources in the economy could eventually reach 10^54 flops. In an economy where wages track the cost of computation, human labor becomes economically marginal.
"Most income will accrue to owners of computing resources," the paper concludes.
The Distribution Question Becomes Central
This shifts the defining challenge of the AGI era from automation to ownership. Who owns the compute determines who captures the gains.
BlackRock CEO Larry Fink warned in his annual letter that AI risks concentrating wealth further. The top 1% of U.S. households already holds more wealth than the bottom 90%. Restrepo notes that addressing this requires structural choices: universal income, treating compute as public infrastructure, or other redistribution mechanisms.
The Transition Matters Now
How we get to that future affects workers today. Restrepo identifies two paths. In a "compute-binding" transition, hardware constraints slow AI adoption. Adjustment is gradual. Wages follow predictable paths. Workers have time to move between roles.
An "algorithm-binding" transition-which resembles the current moment-creates jagged disruption. Capabilities advance in sudden leaps. Workers whose tasks can't yet be automated command large temporary premiums. Others face sudden wage declines.
This is already visible in construction. Electricians, plumbers, and HVAC technicians working on data center projects earn roughly 32% more than those on standard builds, according to hiring platform data. Some electricians pull in $260,000 annually. Electrical work accounts for 45% to 70% of data center construction costs. The U.S. needs roughly 300,000 new electricians over the next decade, plus replacements for 200,000 expected retirements.
Collective Gain, Individual Loss
Restrepo offers one reassurance: workers as a group don't become worse off. AGI expands what the economy can produce. Total labor income across all workers in the post-AGI world exceeds the pre-AGI baseline. We cannot collectively become poorer because we could always retreat and produce as before.
But that collective gain provides cold comfort if concentrated at the top. Forty percent of Americans lack meaningful exposure to capital markets, Fink notes. Without structural intervention, the AI-driven boom leaves them further behind.
Work and Recognition
The paper's title captures something beyond economics. "Historically, work provided not only income but also recognition that one's efforts improved society's well-being," Restrepo writes. "Work gave people the sense that they would be missed."
In the AGI world, that connection severs. If half the workforce stopped showing up today, the economy would collapse. In the AGI economy, we would not be missed.
For HR leaders, the implications are immediate. The transition happening now-with sudden wage premiums for scarce skills and wage declines for automatable roles-requires active workforce planning. Understanding AI's impact on recruitment, talent management, and labor market dynamics is no longer optional. For those in senior HR roles, strategic preparation for the AGI era means rethinking how organizations attract, develop, and retain talent in a world where compute ownership, not skill scarcity, determines economic value.
Your membership also unlocks:
