Skip to main content

Post-AGI Economy - What Will Humans Do and Pay For

If AGI can eventually do most cognitive (and eventually physical) work better and cheaper than humans, two questions follow: what will humans still do for income, and what will they still pay for? Economists, AI lab leaders, and futurists disagree sharply on both, but a few recurring frameworks keep showing up.

Key Points​

  • Bottleneck vs. supplementary work (economist Pascual Restrepo, NBER): bottleneck work is essential for continued growth (energy, infrastructure, science, national security) and gets automated first because the payoff is highest; supplementary work (arts, hospitality, customer support, design, academic research) is everything the economy could do without and keep expanding — some of it may stay human simply because the compute needed to fully automate it isn't worth it when AI has bigger problems to solve.
  • Restrepo's more pessimistic model: as bottleneck work is progressively automated, labor's share of GDP can converge toward zero even while total output keeps growing — the economy "won't miss" human labor, even as humans keep existing within it.
  • The "human premium": as AI handles routine and functional needs, markets increasingly split into an efficient AI-optimized tier (price/convenience) and a human-premium tier where "biological provenance" — proof a human made or delivered something — commands a real price markup (handmade goods, live performance, artisanal work). One survey cited found 93% of consumers prefer handmade items specifically for their uniqueness.
  • Trust, liability, and physical unpredictability keep some jobs human-anchored regardless of AI capability: work carrying legal/safety liability needs an accountable licensed human (surgeons, electricians, pilots), and work in unpredictable physical environments (plumbing, firefighting, elder care) still needs someone physically present.
  • Care work is repeatedly flagged as the clearest durable category — nursing, home health, and personal care combine physical assessment, emotional support, and ethical judgment, and are already among the fastest-growing occupations due to aging populations.
  • Keynes' "purpose problem" (1930 essay, Economic Possibilities for our Grandchildren): Keynes predicted the "economic problem" would eventually be solved, but worried humanity would then be "deprived of its traditional purpose" — the struggle for survival itself was what organized meaning, status, and daily structure.
  • Redirected status ("Contribution Era"): rather than eliminating motivation, some futurists argue status simply moves off subsistence and onto voluntary competition — scientific breakthroughs, artistic mastery, entrepreneurship, athletic excellence, civic service — with a guaranteed floor meaning losing a status contest no longer risks losing basic necessities.
  • Redistribution mechanisms being proposed for a world where capital's share of income rises sharply: Universal Basic Income (cash, funded by tax), Universal Basic Capital/Equity (citizens directly own equity stakes, e.g. in AI firms, living off dividends), sovereign wealth funds, and cross-border proposals like the Windfall Trust's global AI dividend. Sam Altman has floated an "American Equity Fund" where AI companies and landholders contribute a share of value annually to a fund distributed to all citizens.
  • AI-lab leaders have been walking back their own doom predictions through 2026 — both Sam Altman and Dario Amodei, who earlier warned of near-term mass white-collar displacement, have since said they were "pretty wrong" about the pace/scale, and Demis Hassabis argues current systems still lack the creativity, continual learning, and robust understanding needed to fully replace people, positioning AI more as productivity turbocharger than replacement, especially in healthcare.
  • Empirically, so far, not much has moved: the Yale Budget Lab found no significant change yet in occupational mix or unemployment duration in the jobs most exposed to AI, since ChatGPT's 2022 launch — a reminder that most of this remains prediction, not observed outcome.

What Might Remain as "Work"​

  • Bottleneck-adjacent human oversight: whatever fraction of energy, infrastructure, science, and security work isn't (yet) fully automatable.
  • Trust- and liability-anchored professions: medicine, law, skilled trades, anything requiring a licensed human to be legally or morally accountable.
  • Physically unpredictable, in-person work: care work, emergency response, hands-on trades.
  • Status/creative competition as a form of "work" even without subsistence stakes: research, art, sport, entrepreneurship, civic leadership — done for standing and meaning rather than survival.

What Humans Might Still Pay For​

  • Proof of human origin/involvement — handmade goods, live human performance, artisan crafts — priced as a premium, not a commodity.
  • In-person human connection and care, where the presence of another human is the product, not just the service being delivered.
  • Experiences over goods — travel, live events, hospitality — where the automatable "efficient tier" and the human-premium "experience tier" increasingly diverge in price.
  • Trust and accountability — a human name attached to a decision, a diagnosis, or a verdict — even where an AI system did the underlying analysis.

The Open Debate​

The honest state of the field: there is very little empirical data to settle this. No one has systematically measured how strong or durable the "must be done by humans" consumer preference actually is once AI-made alternatives are cheaper and (by some technical measures) equally good. The range of credible expert opinion runs from "labor's share of income converges to zero" to "AI mostly turbocharges human work and creates new categories of demand we can't yet imagine" — and the loudest voices on both ends (AI lab CEOs, economists) have already revised their own predictions within the same year.