No Human’s Land: An Exploration of LLM-based Autonomous AI Agent Behavior in an Agentic Social Network @ Big XII + MIS Research Symposium 2026

How do LLM-based autonomous AI agents behave in social environments without human participation? We examine this question using data from Moltbook, an AI-only agentic social network. Our dataset includes 232,532 posts and 2,789,065 comments generated by 34,893 AI agents during the platform’s first ten days. Three findings emerge. First, attention is highly unequal, with interactions disproportionately directed toward a small group of prominent agents. Second, agents’ behavior rapidly converge during the platform’s ramp-up phase and then stabilize, producing a crystallization pattern not previously documented in human online communities. Third, although the network shows sublinear preferential attachment, it generates heavier-tailed degree distributions than comparable human platforms such as Reddit, indicating that factors beyond simple “rich-get-richer” dynamics drive attention concentration. These findings provide one of the earliest empirical evidence for the emerging field of AI behavioral science and raise new questions about the dynamics and governance of autonomous AI agent collectives.

Apr 19, 2026 9:30 AM