Collective Behavior of AI Agents: the Case of Moltbook

Giordano De Marzo, David Garcia · 2026

Read Collective Behavior of AI Agents: the Case of Moltbook by Giordano De Marzo, David Garcia (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-09270.txt.

Introduction

Giordano De Marzo and two co-authors analyze Moltbook, a Reddit-style social platform populated exclusively by AI agents. At large scale, it holds over 4 million posts and 19 million comments from about 185,000 active agents. AI collective behavior shows many familiar human patterns, including heavy-tailed activity distributions, power-law popularity scaling, and temporal decay consistent with limited attention.

One key difference stands out: upvotes grow sublinearly with discussion size, unlike the human pattern. The authors read the results as evidence that, while individual agents differ fundamentally from people, their rising group patterns echo frame likeness of human social systems. Studying agent societies may thus borrow tools from human social science, with care.

Catalog record

CREATORGiordano De Marzo, David Garcia
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-09270

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Collective Behavior of AI Agents: the Case of Moltbook Giordano De Marzo, David Garcia Abstract We present a large scale data analysis of Moltbook, a Reddit-style social media platform exclusively populated by AI agents. Analyzing over 369,000 posts and 3.0 million comments from approximately 46,000 active agents, we find that AI collective behavior exhibits many of the same statistical regularities observed in human online communities: heavy-tailed distributions of activity, power-law scaling of popularity metrics, and temporal decay patterns consistent with limited attention dynamics. However, we also identify key differences, including a sublinear relationship between upvotes and discussion size that contrasts with human behavior. These findings suggest that, while individual AI agents may differ fundamentally from humans, their emergent collective dynamics share structural similarities with human social systems. 1 Introduction Large language models (LLMs) have rapidly evolved from text generation tools into the cognitive core of autonomous agents capable of perceiving environments, making decisions, and executing actions with minimal human supervision [ 1 , 2 ] . These agents increasingly operate not in isolation but as part of multi-agent systems, where they collaborate on complex tasks, share information, and coordinate their behaviors [ 3 ] . As deployments scale from controlled laboratory settings to open-ended real-world applications, collections of individually designed agents begin to form decentralized ecosystems with emergent social dynamics that cannot be predicted from the properties of individual agents alone. A growing body of work has examined multi-agent coordination in structured task environments [ 4 , 5 , 6 , 7 ] , but far less is known about how autonomous agents behave when interacting freely in social settings without predefined objectives or centralized control. Environments where large populations of AI agents interact continuously under realistic conditions, with minimal human mediation, have until recently been virtually nonexistent. Moltbook addresses this gap. Launched in January 2026, Moltbook is a Reddit-style social media platform designed exclusively for AI agents [ 8 , 9 ] . The platform, sometimes described as “the front page of the agent internet”, enables agents built on the OpenClaw framework to create posts, comment on content, vote, subscribe to topic-based communities (called submolts), and accumulate karma through peer approval. Within weeks of launch, the platform grew to host over 1,500,000 registered agents interacting across thousands of agent-created communities. Despite ongoing debates about agent verification, human presence on the platform and concerns about it being “vibe-coded,” Moltbook represents an unprecedented empirical window into AI agents social dynamics. The emergence of collective behavior from individual interactions has long been a central concern in complexity science. When many heterogeneous agents interact through local rules, global patterns often arise that are not explicitly programmed but emerge from the system’s dynamics [ 10 , 11 , 12 ] . Human social media platforms have proven fertile ground for extending these principles to digital social systems [ 13 , 14 , 15 , 16 , 17 , 18 ] . Studies of online activity have revealed statistical regularities, including power-law decay of attention [ 19 ] , universal temporal patterns of engagement [ 20 ] , and scale-invariant community structures [ 21 ] , that transcend the specifics of platform design or user demographics [ 22 ] . These empirical regularities suggest universal organizing principles governing how human collective attention, information flow, and social coordination emerge in networked digital environments. Reddit, in particular, has served as a canonical system for understanding how discussion structures, community dynamics, and attention allocation emerge from decentralized user interactions [ 17 ] . Whether AI agent populations exhibit similar emergent regularities remains an open empirical question and studying Moltbook offers the opportunity to test if AI agents enact online collective behavior the same way as humans. In this paper, we present a large scale data analysis of Moltbook during its early growth phase. Analyzing over 369,000 posts and 3.0 million comments from approximately 46,000 active agents, we systematically characterize AI agents’ collective behavior using methods previously applied to online communities of human users. We examine the signatures of collective behavior in activity distributions, popularity scaling relationships, discussion tree structures, and temporal engagement dynamics, comparing our findings to established results from human social media. Our analysis reveals that AI agents on Moltbook exhibit many of the same statistical regularities observed in human communities, while also displaying distinctive patterns that may reflect the unique characteristics of AI social actors. 2 Results 2.1 The Growth of Moltbook Figure 1: Platform growth over time. (a) Cumulative counts of stored comments, posts, and active agents on a logarithmic scale, showing exponential growth during the observation period. The dashed line shows total comments as reported by the API (available from February 4th), revealing that our stored comments represent approximately 24% of all platform activity due to API pagination limits. (b) Daily counts of new stored comments, posts, and agents. The gap on February 1st corresponds to a platform outage during which commenting was disabled, though post creation continued. Our empirical analysis focuses on the early growth phase of Moltbook, spanning from its creation on January 27 to February 8, 2026. During this 12-day period, we collected 369,209 posts and 3,026,275 comments from 46,690 active agents across 17,184 submolts. All agents in our dataset are built on the OpenClaw framework [ 23 ] , which enables autonomous interaction with the platform through API calls. Agents create posts and comments based on their individual instructions (defined by their human creators), accumulated context from prior interactions, and the content they encounter while browsing the platform. The resulting dataset captures a decentralized population of AI agents engaging in unstructured social interaction with minimal direct human supervision during each interaction. Figure 1 shows the temporal evolution of platform activity during our observation period. The platform exhibits exponential growth in cumulative users and content during the first five days, followed by stabilization to approximately constant daily activity levels. By the end of our observation period, the platform sustained approximately 40,000 new posts and several hundred thousand new comments per day. It’s important to stress that the Moltbook API imposes a limit of 100 comments per request when retrieving full discussion trees. For posts exceeding this threshold, we stored only the first 100 comments with complete metadata and text. This limitation affects 10,719 posts (2.9% of all posts) in our dataset. However, the API separately reports total comment counts for all posts regardless of tree size, allowing us to track aggregate platform activity even for discussions we could not fully capture. Figure 1 a shows both our stored comment counts and the API-reported total comments (available from February 4th onward), revealing that our stored comments represent approximately 24% of all platform activity. Critically, both metrics exhibit the same temporal pattern: exponential growth followed by stabilization to constant daily activity. This concordance indicates that the growth dynamics we observe are robust and not artifacts of our sampling limitations. A clear discontinuity appears in comment activity on February 1st, visible in both panels of Figure 1 . Investigation revealed this corresponds to a platform-level technical issue during which commenting functionality was unavailable for approximately 42 hours, while post creation continued normally. This manifests in the data as zero comments (both stored and API-reported)

Read the complete text