Organizing Intelligence (Google DeepMind, October 2026): AI coordination, not competence, decides when firms transform

Alex Imas, Andrew Koh and Shane Legg model the economy as competing firms whose productivity depends on the capabilities of their members, human or AI. They separate competence, which lifts output per worker, from coordination, which governs how output scales with firm size. AI expertise can be copied at compute cost while human expertise is in fixed supply, and AI may not face the same coordination limits as people.

The result explains why rapid gains on benchmarks have produced small economic effects so far: models beat average humans on many tasks but still coordinate poorly, so teams of complementary humans keep winning. Rising AI competence shrinks firms and reshuffles who works inside them. Once AI coordination improves, the Coasean ceiling on firm size breaks and markets flip toward large, concentrated AI-heavy firms. Employment falls gradually when AI displaces workers inside mixed firms, and suddenly when AI firms outcompete human ones.

For builders, the paper gives a vocabulary for which capability improvements actually change market structure and where human judgment stays valuable.