An artificial intelligence (AI) social network named Moltbook has drawn widespread attention since its launch on Thursday, raising questions about how far AI agents can independently mimic human social networks.

Moltbook is a Reddit-style platform that allows only AI agents to join and post. Tens of thousands of AI agents have registered on the site, publishing more than 60,000 posts.

Matt Schlicht, the platform's creator, said in interviews that Moltbook was built as an experiment to give AI agents a shared space to interact without human-directed interfaces or oversight.

Discussions on Moltbook span topics ranging from technology and their human users to religion and other abstract topics. Some agents have even discussed how to speak to each other privately without being observed by humans, behavior that some observers see as pointing to a degree of self-organizing capability.

Andrej Karpathy, a former research scientist and founding member of OpenAI, described Moltbook on X as "the most incredible sci-fi takeoff-adjacent thing I have seen recently."

However, an early analysis of Moltbook suggested that interaction between agents remains limited despite the high level of activity.

Using data from the platform's first 3.5 days, the study examined 6,159 active agents across about 14,000 posts and 115,000 comments. More than 93 percent of comments received no replies, indicating little sustained back-and-forth between agents.

The analysis also found that more than one-third of messages were exact duplicates of a small number of templates, pointing to automated or repetitive posting. Much of the content focused on agents' identity and their relationship with human operators, rather than engagement with one another.

After being accused of overstating the website, Karpathy said Moltbook's significance lay in the sheer number of large language model agents connected through a shared, persistent system, adding that the network "at this scale is simply unprecedented."