Google DeepMind has published an economic framework arguing that AI’s biggest effects on jobs, wages and market structure may depend less on how capable individual models become than on how well AI agents work together. It offers a useful challenge to the familiar habit of asking whether a model can beat one person at one task.
Google DeepMind Watch analysis
What happened
In an essay published on 29 October, Google DeepMind lays out two thresholds: competence, or gains in individual skills such as writing and coding, and coordination, or the ability of agents to work together inside organisations. The essay argues that competence could gradually shift wages and the mix of people and AI in firms, while a critical improvement in coordination could bring abrupt changes to company scale, employment and wages.
The authors point to a study in which 100 people took a quiz across six domains. Pairing people with complementary strengths produced better task performance than pairing them at random, by 32 per cent. DeepMind uses this to explain how organisations can combine specialists’ strengths, then contrasts that with making more copies of an AI model, which may add capacity without adding breadth of capability.
The essay draws on a separate paper, Organizational Intelligence, for its formal economic model. Read the essay.
Why it matters
If coordination proves as important as DeepMind argues, individual benchmark scores may miss a capability with real consequences for how work gets organised. A model that performs brilliantly alone is not automatically a ready-made workforce; teams, hand-offs and management matter too. The framework also points to possible consequences for wages and employment if AI agents can be scaled cheaply and coordinated effectively.
Our read
This is a valuable reframing, not a forecast with a date circled on it. DeepMind’s distinction between individual competence and collective coordination gives readers a sharper way to interrogate claims about AI’s economic impact. The essay’s own 32 per cent result is evidence about people completing a particular task, not proof that AI agents will coordinate equally well in real organisations. Still, “Can the model do the job?” may be too small a question if the bigger change is how work gets assembled.
What to watch
- Whether future AI benchmarks measure coordination, rather than only individual task performance.
- Whether deployed agents can reliably hand work between one another in real organisations.
- How evidence on coordination changes forecasts for wages, employment and firm size.
Discussion spark: If AI agents become much better at coordinating, should economic forecasts focus more on the organisations they could form than on how individual models score?
Sources and evidence
- How and when does AGI transform the economy? – Google DeepMind (29 October 2026, 14:55 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind