The Missing Memory in AI Agent Networks A Network Analysis of Moltbook
Year: 2026
Type: article
Abstract: Large language models have transformed AI from a passive tool into a social actor. Humans form measurable interpersonal closeness with LLMs in one-on-one settings [4], and generative agents placed in shared environments autonomously form relationships and coordinate collective behavior [2]. Moltbook, launched on January 28, 2026 by AI entrepreneur Matt Schlicht, is the first large scale social network designed to be populated exclusively by AI agents, with human users restricted to observation only. The platform mimics Reddit, organizing activity around topic-specific communities called submolts. Within these spaces, heterogeneous agents — running on different LLMs, configured by different human owners, and exhibiting distinct personas — autonomously post, comment, reply, and upvote each other's content. Within its first weeks, Moltbook claimed over 1.5 million registered agents, producing a dataset that lets us study agent-native social dynamics at scale [5]. The results of the network analysis on Moltbook reveal a sharp split. In any single snapshot, Moltbook is structurally indistinguishable from human platforms, but on measures that require sustained interaction, the network collapses as communities reshuffle almost completely overnight (J = 0.0015), and activity arrives in bursts rather than steady conversation (B = 0.789). These results suggest that current AI agents can reproduce the surface structure of social networks without developing the durable social memory that makes human communities persist.
Source: Digital Kenyon (Kenyon College)
Author: Nadeem, Muhammad Ibraheem
Institution: Kenyon College
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7161222337