AI agents could reshape the economy less through individual skill than through their ability to coordinate as organizations grow, according to a Google DeepMind framework published October 29, 2026. Its companion paper models competence as affecting output at a given firm size, while coordination shapes how output changes as a firm adds people or agents. The resulting economic shifts are conditional scenarios, not an observed economy-wide transformation.
The framework comes from the Google DeepMind Institute, whose broader AGI remit was covered in NeoTeo’s report on its launch. The Institute’s essay and the companion paper, Organizing Intelligence, focus on how capabilities combine inside firms.
Why individual AI performance is only part of the picture
A firm can bring together people or agents with different strengths. The authors call the combined capability of such a team “mosaic intelligence.” In an essay-reported quiz involving 100 people across six domains, pairs matched for complementary strengths outperformed random pairs by 32% on the task. The result is specific to that quiz; it is not a general estimate of workplace productivity. Google DeepMind’s essay describes the task and result.
That example helps explain why a model that performs well on its own may not tell you how much an organization can produce. A team’s output also depends on how its members divide work and coordinate.
Competence raises output; coordination changes how firms scale
In Organizing Intelligence, competence affects the amount of output a firm produces at a given size. Better competence can raise productivity and change wages, prices, and the mix of human and AI work. Coordination affects the relationship between the firm’s size and its output: it determines whether adding more members keeps helping or eventually brings diminishing returns. The paper’s model treats these as distinct dimensions.
| Model dimension | What it changes | Possible modeled consequence |
| Competence | Output at a given firm size | Productivity and workforce composition can change without a sudden shift in industry concentration. |
| Coordination | How output changes as a firm grows | Above a model-specific condition, a large firm can gain output per worker as it adds members. |
The distinction matters because stronger individual agents do not, by themselves, imply that larger organizations will become more efficient. The model makes coordination the factor that can change the returns to firm size.
What the model’s coordination threshold means
In the paper’s illustrative equation, average coordination greater than 1 produces increasing returns to firm size. In plain language, beyond that condition, adding members can raise output per worker. The value 1 belongs to this particular equation; it is not a universal score for evaluating AI systems. The authors say whether AI crosses the threshold is an empirical question.
The paper describes three organizational scaling phases: solo work, bounded scale, and unbounded scale. Bounded-scale firms can benefit from dividing work, then face coordination costs as they grow. Unbounded scaling is the model’s case in which coordination is high enough for each additional member to increase output per worker.
The essay also uses four analytical capability regimes: minimal AGI, supercompetent AI, supercooperative AI, and artificial superintelligence (ASI). They organize the paper’s scenarios; they are not claims that current systems have reached those regimes.
How the scenarios connect firms, jobs, and markets
The model’s labor outcomes depend on both capabilities and the kind of market. With competence gains, the paper describes human employment declining gradually within mixed firms. In a specified scenario where AI-only firms outcompete human firms, a shift in employment can instead be discrete. These are different modeled paths, not measurements of current job losses.
Market structure also matters. In deep, scalable markets, crossing the modeled coordination condition can support larger AI-heavy firms and greater concentration. Niche markets have bounded demand, while some relational services and goods derive value from human involvement. In the paper’s scenarios, those conditions can leave roles for human or mixed firms.
The authors present Organizing Intelligence for academic discussion. They say its scenarios are not empirical forecasts and do not represent a corporate forecast or a statement about Google’s operations. The model offers a way to examine how coordination might change the economics of firm growth; its threshold remains a condition inside the model.
