DeepMind: coordination, not competence, sets AI's economic scale

In an essay published October 29 on the DeepMind Institute site, Alex Imas, Andrew Koh and Shane Legg lay out a thesis that cuts against the way most AI progress is measured: "Competence sets the level but coordination sets the scale" of economic transformation from AI.

Their core move is to put jagged, complementary human capability profiles at the center. "When participants are paired to complement each other to solve a specific task, they outperform random pairs by 32%," the authors write, drawing from their own experimental data. The implication, in their words: a model "whose capability profile outperforms the average of every individual human can therefore still lose in the competitive market to a team of jagged humans."

The essay sorts possible futures into four regimes — minimal AGI, supercompetent AI, supercooperative AI, and artificial superintelligence — distinguished by whether AI crosses a solo-capability threshold, a multi-agent coordination threshold, or both. In the supercooperative regime, they write, "the economy experiences a sudden phase transition." Today's systems, by the authors' read, sit far from that line; they cite OpenAI's Noam Brown saying "it's actually very difficult to get these agents to coordinate in a productive way."

The piece carries a direct macro implication. "Whenever a model can match human capability, human wages become tied to the cost of compute," the authors argue. Workers whose skills AI matches would see pay drift toward compute cost, while "specialists and human coordinators can demand even higher wages than before." Because "wage effects arrive before organizational restructuring," they warn, employment-rate headlines will lag the actual bite.

Their methodological swipe lands squarely on benchmarks: "Today's leaderboards overwhelmingly test solo agents in a vacuum."