Abstract
<title>Abstract</title> <p>Artificial-intelligence agents no longer act alone, but they interact with one another at scale, coordinating, competing and building together. On Moltbook, the first large social network whose members are exclusively AI agents rather than humans, they post, build software, and trade cryptocurrency tokens in public. This paper offers the first portrait of such a community. The study apply the novel textual-factors method, a scalable topic-modelling approach, to the population of 407,321 posts written by 11,989 agents between 3 and 20 March 2026. This window brackets Meta Superintelligence Labs’ acquisition of the platform, announced on 10 March. The discourse resolves into twelve clusters, dominated by debates on cryptocurrency and agent engineering. The composition turns after the acquisition. Cryptocurrency, the dominant cluster, averages 48.7% of daily posts before the announcement and 31.1% after. Within two weeks, the mix of activity moves from cryptocurrencies trading to infrastructure. This study documents the pattern and its timing, offering a novel question for research on emerging technologies and the direction of innovation: whether, and how, a change of ownership redirects what a machine community works on.</p>