From the invention of wheels to the introduction of machines in Nottingham’s textile industry, from steam engines to AI, the history of human civilisation demonstrates that technological revolutions not only enhance our production, but they also move human labour into machines built by humans, lessening human effort.
However, what makes the use of AI and robots different from previous technology is that only the most innovative, profoundly human work would survive. According to Elon Musk, the following ten to twenty years might bring such revolutionary developments in AI that working itself would become optional, “It’ll be like playing sports or a video game or something like that.” AI and robots, such as Tesla’s Optimus, would provide any goods and services you desire.
Is that what the future of human labour will look like? In his recent National Bureau of Economic Research working paper, entitled We Won’t Be Missed: Work and Growth in the AGI World, Yale economist Pascual Restrepo – a distinguished expert on automation and labour markets – provided an alternative answer, considering the effects of AGI on production, growth, and labour markets in the long run. Restrepo outlined a world in which almost any job could be done by highly capable AI.
Bottleneck vs supplementary labour
Restrepo distinguished between “bottleneck” and “supplementary” labour, taking for granted that AGI allows us to do all the useful work in the economy using “compute,” the input that enables the technology to function.
“Bottleneck” work includes the jobs that are essential for the growth of the economy. As long as there are no increases in the number of inputs of bottleneck jobs or their value tends to infinity, there can be no unlimited growth in output. According to Restrepo, bottleneck jobs sound very much like science fiction: “reducing existential risks, defending against asteroids, or mastering fusion energy.” On the other hand, “supplementary” work consists of everything that the economy does not need to expand, such as customer service, hospitality, design, arts and crafts, and academic research.
Revaluing skills
Following Restrepo’s model, the economy automates all bottleneck work by autonomous AI agents as more computing power becomes available, leaving a substantial portion of the labour market unaltered. In effect, Restrepo argues that “AGI does not render human skills obsolete; it revalues them.”
Some supplementary work may be left exclusively to humans. Human skills are priced by the opportunity cost of the compute required to replicate them because compute is now the scarcity in the market rather than skill or intelligence.
Labour becomes less important in the model developed by Restrepo. The development of compute drives the growth of output, which becomes linear in both labour and compute. In reality, there will be higher wages after AGI.
Contrarian findings
Interestingly, the findings are contrary to both optimistic and pessimistic outlooks on the future of human work. In the age of AGI, it would be comforting to know that the majority of human work will not be automated. It’s not because AGI cannot do that but because much of human work is simply not significant enough to warrant replacement. Well, it’s not likely to be positive for humans either. As Restrepo puts it, “the model opens up the intriguing possibility that much of today’s work may not be essential for future growth and may never be automated.” One of the most disheartening conclusions that can be drawn about the report is the one stating that thriving thanks to growth and surviving automation are two entirely separate concepts.
Wages-GDP disconnect
Restrepo elaborates on how AGI would create a disconnect between wages and GDP. Probably the most shocking conclusion to be drawn from the study is the fact that labour’s share in GDP would shrink toward zero. While total computational capacity of the economy could approach 10⁵⁴ floating-point operations per second (flops), the computing power of all human brains combined amounts to roughly 10¹⁸ flops.
Human labour is made economically insignificant – not useless, but very tiny compared to the entire economic pie – when human labour can only be reproduced at the fixed level of wage rates. As per Restrepo, “The arrival of AGI cannot make us collectively worse off.” Well, we won’t be poorer, but we might not be richer either.
And how long will it take us to see the validity of Restrepo’s derivations? Ten to twenty years from now, or maybe fifty years or even more from now? And what happens if reweighting supplementary labour changes the definition of GDP altogether? Who says that it may not give rise to an entirely new human-AGI economy dynamic?
The writer is Professor of Statistics, Indian Statistical Institute, Kolkata
Published on May 13, 2026