Google-affiliated researchers Benjamin Bratton, Blaise Agüera y Arcas and James Manyika argue that advanced AI is more likely to emerge as a network of people, agents and institutions than as one isolated superintelligence. They call the alternative Artificial Symbiotic Intelligence.
The essay treats intelligence as a property of coordinated systems. Current orchestration tools already divide tasks among models, while future interfaces may show groups of agents as a network rather than forcing users to hold separate one-to-one chats. People would direct roles, dependencies and decisions across the whole system.
That framing changes the safety problem. If capability comes from an institution of many participants, governance cannot focus only on making an individual model larger or aligning it once. Rules, procedures, permissions, feedback loops and the division of responsibility shape what the combined system can do.
The proposal is a conceptual argument, not evidence that this architecture will produce general intelligence. It also warns against interpreting agent language as proof of human-like experience. Its useful implication is narrower: organizations deploying groups of agents should evaluate the coordination layer and institutional incentives, not only benchmark each model separately. A well-defined process may matter more than the nominal intelligence of any one component.