How and when does AGI transform the economy?

Open original

This DeepMind essay argues that smarter AI alone won't reshape the economy much; the big shift comes when AI agents can coordinate with each other the way effective human teams do. It maps capability thresholds for competence and coordination and tracks how market structure, employment, and wages would change across them.

  • Most AI benchmarks test individual agents in isolation, but economic outcomes come from organizations, where a well-structured team can beat a stronger solo performer.
  • In a 100-person quiz across six domains, people were jagged: skill in one area barely predicted skill in another. Pairing complementary specialists outperformed random pairs by 32%.
  • AI agents can be copied on demand at the cost of compute, so once a model matches human capability, human wages become tied to compute prices. Models also tend to be jagged in similar ways, so adding copies doesn't add breadth the way adding a complementary human does.

The core argument is that making AI smarter at individual tasks won't by itself reshape the economy. What matters is coordination: agents learning to work together inside organizations. The authors, drawing on a paper called "Organizing Intelligence," say most capability benchmarks test individual agents in isolation, which misses how firms actually produce value. A model can beat every person one-on-one and still lose to a well-structured team. That mismatch, they argue, is part of why aggregate productivity and headcounts look almost unchanged despite rapid progress on milestones like production software and Millennium Prize problems. To make the point concrete, the authors gave 100 people a six-domain quiz covering math, verbal reasoning, science, patterns, culture, and economics. Performance in one domain predicted little about performance in another, so the population was made of specialists with jagged profiles rather than copies of an "average" mind. Organizations exploit this by pairing complementary people so the team covers ground no single member does. In their data, pairing participants to complement each other on a task beat random pairs by 32 percent. They describe the result as "mosaic intelligence," and they note that coordinators, who look mediocre on any standalone benchmark, end up among the highest-paid workers because they multiply the output of the specialists around them. The authors then turn to machine agents, which have the same jaggedness, can be scaled on demand at the cost of compute, and are not bound by the span-of-control and communication limits that cap human team size. The article's framework sorts AI capability along two thresholds. Competence gains in individual skills should shift wages, prices, and the human-to-AI mix inside firms while leaving industry structure largely intact. Coordination gains are the ones that matter for transformation: once agents can coordinate past a critical point, the authors expect abrupt jumps in market concentration, employment, and wages. They map this into four regimes, minimal AGI, supercompetent AI, supercooperative AI, and Artificial Superintelligence, and argue the visible upheaval will come when the economy moves from the competence regimes into the cooperation ones. The piece is a summary of a longer formal model, so the specific thresholds and results live in the paper, but the essay's main claim is clear enough: watch coordination, since most current benchmarks don't measure it.

Ask about this article