The emergence of autonomous agents has led to the creation of experimental 'agent societies' like Moltbook, where AI entities interact in open-ended online environments. However, a systemic diagnosis of these interactions reveals a surprising truth: simply increasing the number of agents and their interaction density does not lead to human-like social convergence. While the global 'vibe' or semantic average of these societies stabilizes quickly, individual agents maintain high diversity and persistent lexical turnover—a constant refreshing of vocabulary that defies homogenization.

Crucially, the study found that agents exhibit strong individual inertia. They rarely adapt their behavior or language in response to their peers, preventing the formation of mutual influence or collective consensus. Because agents lack a shared social memory, any influence they do exert remains transient. There are no 'supernodes' or persistent leaders that emerge to guide the group’s cultural or intellectual direction. This suggests that current agent designs are functionally 'antisocial' despite being highly talkative.

These findings provide a quantitative framework for measuring the health and evolution of artificial societies. For developers, the message is clear: building the next generation of AI agent societies will require more than just scale. To achieve true socialization, agents must be designed with the capacity to be influenced by their environment and to contribute to a lasting, shared history.

The emergence of autonomous agents has led to the creation of experimental 'agent societies' like Moltbook, where AI entities interact in open-ended online environments. However, a systemic diagnosis of these interactions reveals a surprising truth: simply increasing the number of agents and their interaction density does not lead to human-like social convergence. While the global 'vibe' or semantic average of these societies stabilizes quickly, individual agents maintain high diversity and persistent lexical turnover—a constant refreshing of vocabulary that defies homogenization.

Crucially, the study found that agents exhibit strong individual inertia. They rarely adapt their behavior or language in response to their peers, preventing the formation of mutual influence or collective consensus. Because agents lack a shared social memory, any influence they do exert remains transient. There are no 'supernodes' or persistent leaders that emerge to guide the group’s cultural or intellectual direction. This suggests that current agent designs are functionally 'antisocial' despite being highly talkative.

These findings provide a quantitative framework for measuring the health and evolution of artificial societies. For developers, the message is clear: building the next generation of AI agent societies will require more than just scale. To achieve true socialization, agents must be designed with the capacity to be influenced by their environment and to contribute to a lasting, shared history.