Google DeepMind report: AGI's economic impact depends on multi-agent coordination capability

Current frontier large models have approached or surpassed human experts on many vertical tasks (high expertise), but their multi-agent coordination capability remains weak (low coordination). In the Google DeepMind report "The Economics of Intelligence: How and When Will AGI Trigger Economic Change?", models were used to accurately predict a true picture of the current labor market—macro employment and industry concentration have not changed dramatically, but startups have become leaner, organizational hierarchies flatter, and demand continues to rise for senior managers capable of coordinating complex work across teams.

Most existing AI benchmarks evaluate individual agents in isolation, but the basic unit of economic competition is the organization (the firm), not the isolated individual. Based on the different stages of evolution along the two dimensions of "expertise" and "coordination capability," the report divides the economy into four regimes:

1. Minimal AGI

A single AI reaches average human level, but coordination capability is weak; humans still prevail through complementary "mosaic intelligence" teams; existing corporate structures remain basically unchanged. The impact on overall employment, industry structure, and wage levels is weak.

2. Supercompetent AI

Individual professional skills surpass top humans, but coordination capability remains limited; no dramatic monopoly upheaval occurs in industries; because human managers can use powerful AI to independently complete more work, firms instead tend toward smaller scale and leaner, flatter structures. Pay for specialized execution roles aligns with compute costs; the value and pay of human coordinators (managers) soar;

3. Supercooperative AI

Professional skills remain at an ordinary level, but coordination capability breaks through the critical threshold, breaking the "Coasean ceiling" and triggering a market phase transition; increasing returns to scale enable a few ultra-large pure-AI firms to rapidly monopolize the market. Traditional human firms are pushed out on a large scale, the advantage of human coordinators collapses, and employment and wages fall off a cliff;

4. ASI

Both professional capability and coordination capability comprehensively surpass humans; ultra-large or highly efficient AI organizations dominate standardized mainstream markets; as single-agent capability becomes extremely powerful, the number of agents needed to meet market demand may instead fall again. Humans fully withdraw from mainstream production and retain roles only in niche markets that depend on specific human emotions or preferences;

To measure when AI triggers disruptive economic change, one cannot look only at isolated single-agent benchmark scores; one must focus on when the coordination capability of multiple agents collaborating with low friction in long chains and large-scale organizations approaches the critical threshold.