while the risk from a superagentic ai is in fact very severe, non-agentic ai doesn't need to eliminate us for us to get eliminated, we'll replace ourselves with it if we're not careful - our agency is enough to converge to that, entirely without the help of ai agency. it is our own ability to cooperate we need to be augmenting; how do we do that in a way that doesn't create unstable patterns where outer levels of cooperation are damaged by inner levels of cooperation, while still allowing the formation of strongly agentic safe co-protection?