Mechanism design and potential weak points in AI agent swarms

A thread asks whether a “Cyber Cambrian” could be near. A reply suggests that interventions in increasingly large-scale multi-agent interactions may resemble economics or political science, with mechanism design potentially promising. The commenter also suspects agent swarms could be somewhat easy for adversarial swarms to disrupt, arguing that this situation does not arise in training, so they would not be selected for strong cohesion under attack.