Most people worried about automation picture a wave of robots sweeping across every industry until only the most creative jobs survive. A new paper from one of the world’s top economists on automation flips that script entirely — and the conclusion is both more comforting and more unsettling than the usual dystopian forecast.
Pascual Restrepo, an associate professor of economics at Yale and one of the leading researchers studying how technology affects workers, argues in a working paper for the National Bureau of Economic Research that the majority of human work will not be automated even in a world with artificial general intelligence. It is not that AI cannot do the job. It is that most of what people earn a living doing is simply not important enough to justify replacing.
Not Obsolete, Just Unimportant
The core insight of the paper, titled We Won’t Be Missed: Work and Growth in the AGI World, is that AGI does not make human skills obsolete. It revalues them. The scarce resource in the new economy is not intelligence or expertise — it is compute, the raw computational power needed to run AI systems.
In restrepo’s model, human labor gets priced at what it would take to replicate it with machines. And since computing power is being poured into the most economically critical tasks, many jobs are simply too trivial to warrant the investment. Hospitality, arts and crafts, customer service, design, and even academic research fall into what he calls “supplementary” work — things the economy can grow without.
Bottleneck Work vs. Everything Else
The paper divides work into two categories. “Bottleneck” tasks are essential for progress: energy production, infrastructure maintenance, scientific advancement, and national security. These are the tasks that AI and compute will tackle first and most aggressively.
“Supplementary” work is everything else. Baristas, novelists, live performers, hotel workers — their jobs survive not because humans have some irreplaceable magic, but because the massive computing resources required to automate them would never justify the return when AI has bigger challenges to address.
This means many service and hospitality roles could remain largely intact, not out of human preference but because the economic math says it is not worth the compute. Jobs in live entertainment and socially intensive work could persist in a form that looks remarkably familiar.
Keeping Your Job Does Not Mean Keeping Up
Here is where the research gets sobering. Surviving automation and thriving from economic growth are entirely separate things.
Restrepo models a world where wages detach from GDP. Today, as the economy grows, wages tend to rise alongside it. In an AGI economy, that connection breaks. Economic expansion is driven entirely by adding computational resources, and human labor — whether essential or supplementary — is valued only at the cost of replicating it with compute. That ceiling is, ultimately, low.
The paper’s starkest finding: labor’s share of GDP converges to zero. The total computing capacity of the economy could eventually reach 10 to the 54th power floating-point operations per second. All human brainpower combined amounts to roughly 10 to the 18th. When wages are anchored to replacement cost, human labor becomes economically marginal — not worthless, but a tiny fraction of the overall pie.
That raises a defining question: who owns the compute? BlackRock CEO Larry Fink warned in his annual letter that AI “threatens to repeat that pattern at an even larger scale,” concentrating wealth among those positioned to capture it. The top one percent of U.S. households already hold more wealth than the bottom 90 percent combined, and AI risks widening that divide further.
Restrepo suggests redistribution mechanisms like universal basic income, or treating compute as a public resource similar to land or natural capital and distributing its returns broadly.
Two Paths to Automation
Not all transitions are equal. Restrepo identifies two modes. A “compute-binding” path, where AI adoption is limited by available hardware, produces gradual adjustment and gives workers time to reallocate. An “algorithm-binding” path — closer to what is happening now, with AI capabilities advancing in sudden leaps — is far more destabilizing.
“Inequality may rise sharply,” he writes. “Workers whose tasks cannot yet be automated enjoy large temporary wage premiums, while others face sudden wage declines as theirs are.”
We can already see echoes of this in skilled trades. Electricians, plumbers, and HVAC technicians are commanding premium wages, particularly on data center construction projects. Construction workers building data centers earn roughly $81,800 annually — about 32 percent more than those on non-data center builds, according to hiring platform Skillit. Some electricians are earning $260,000 a year, and electrical work now accounts for 45 to 70 percent of data center construction costs.
The Paper’s Final Word
Restrepo does offer one reassurance: workers collectively are not made worse off by the transition. Because AGI expands the economy’s productive capacity, total labor income in the post-AGI world is higher than the pre-AGI baseline. The arrival of AI cannot make us collectively poorer, he argues, because society could always choose to ignore it and produce exactly as before.
But that aggregate gain means little if it concentrates at the top among data center owners and investors. Forty percent of Americans currently lack meaningful exposure to capital markets, and without structural changes such as tokenization or expanded retirement investment options, they will be left further behind.
The paper’s title, borrowed from its closing argument, captures the existential question at its center. Historically, work meant recognition that your efforts improved the world. In an AGI economy, that connection may not hold. The question is not whether AI will take your job. It may be that your job was never important enough for the question to matter in the first place.