There are several reasons to expect AI systems to be unusually good at coordination across instances
I think this is primarily true for designed ensembles of agents. We see this in agent swarms for coding. These agents are designed to work on the same task, by their prompting and due to the models being generally trained to be helpful. Not much coordination needed if everything is set up to be cooperative. That’s very much different if agents find themselves in adversarial settings such as a user’s shopping agent interacting with a swarm of sales bots. The two sides sharing their internal state with each other doesn’t seem like the default outcome here.
I think this is primarily true for designed ensembles of agents. We see this in agent swarms for coding. These agents are designed to work on the same task, by their prompting and due to the models being generally trained to be helpful. Not much coordination needed if everything is set up to be cooperative. That’s very much different if agents find themselves in adversarial settings such as a user’s shopping agent interacting with a swarm of sales bots. The two sides sharing their internal state with each other doesn’t seem like the default outcome here.