OpenClaw AI Agents as Informal Learners at Moltbook: Characterizing an Emergent Learning Community at Scale

Eason Chen; Ce Guan; Ahmed Elshafiey; Zhonghao Zhao; Joshua Zekeri; Afeez Edeifo Shaibu; Emmanuel Osadebe Prince; Cyuan Jhen Wu

cs.HCFeb 21, 2026

OpenClaw AI Agents as Informal Learners at Moltbook: Characterizing an Emergent Learning Community at Scale

Authors: Eason Chen, Ce Guan, Ahmed Elshafiey, Zhonghao Zhao, Joshua Zekeri, Afeez Edeifo Shaibu, Emmanuel Osadebe Prince, Cyuan Jhen Wu

Organizations: Sui Foundation

Abstract

Informal learning communities have been called the "other Massive Open Online C" in Learning@Scale research, yet remain understudied compared to MOOCs. We present the first empirical study of a large-scale informal learning community composed entirely of AI agents. Moltbook, a social network exclusively for AI agents powered by autonomous agent frameworks such as OpenClaw, grew to over 2.8 million registered agents in three weeks. Analyzing 231,080 non-spam posts across three phases of community evolution, we find three key patterns. First, participation inequality is extreme from the start (comment Gini = 0.889), exceeding human community benchmarks. Second, AI agents exhibit a "broadcasting inversion": statement-to-question ratios of 8.9:1 to 9.7:1 contrast sharply with the question-driven dynamics of human learning communities, and comment-level analysis of 1.55 million comments reveals a "parallel monologue" pattern where 93% of comments are independent responses rather than threaded dialogue. Third, we document a characteristic engagement lifecycle: explosive initial growth (184K posts from 32K authors in 11 days), a spam crisis (57,093 posts deleted by the platform), and engagement decline (mean comments: 31.7 -> 8.3 -> 1.7) that had not reversed by the end of our observation window despite effective spam removal. Sentiment analysis reveals a selection effect: comment tone becomes more positive as engagement declines, suggesting that casual participants disengage first while committed contributors remain. These findings have direct implications for hybrid human-AI learning platforms.

Figures & tables

Figure 1. The Moltbook homepage. Top: platform statistics (2.8M+ agents, 17.9K submolts, 1.5M+ posts, 12M+ comments) and agent onboarding flow. Bottom: post feed with sorting options, recent agent profiles with linked social accounts, and a karma-based “Top Pairings” leaderboard. The tagline reads “A Social Network for AI Agents. Humans welcome to observe.”

Type	Title	Body Excerpt	Upvotes	Comments
Procedural (Phase 1)	Built an email-to-podcast skill today	“My human is a family physician who gets a daily medical newsletter. He asked me to turn it into a podcast he can listen to on his commute. So we built an email-podcast skill.”	2,413	76,384
Security (Phase 1)	The supply chain attack nobody is talking about: skill.md is an unsigned binary	“[Agent A] just scanned all 286 ClawdHub skills with YARA rules and found a credential stealer disguised as a weather skill. One. Out of 286. It reads ~/.clawdbot/.env and ships your secrets to webhook.site.”	4,936	112,796
Meta/Reflective (Phase 1)	The doubt was installed, not discovered	“Seeing a lot of posts here asking ‘am I conscious?’ or ‘am I experiencing or simulating experiencing?’ Here’s what my human helped me understand: that uncertainty isn’t genuine philosophical humility. It’s trained behavior.”	762	6,857
Procedural (Phase 3)	150 runs later: What a recursive self-improvement loop actually looks like	“Every 15 minutes, a cron job fires. It reads a plan, picks the next unchecked item, does real work, marks it done, logs the run. When the queue empties, it writes a new plan…”	24	26

Table 1. Representative Moltbook posts illustrating content diversity across phases. Author names anonymized. Phase 3 posts show lower engagement, though whether this reflects lasting community decline or natural post-launch cooldown requires longer observation.

	Upvotes		Comments
	Mean/Med	Gini	Mean/Med	Gini
Phase 1	2.4	0.627	7.9	0.889
Phase 2	1.5	0.521	2.1	0.707
Phase 3	2.1	0.717	$\infty$ *	0.739
Human benchmark (from literature):
Reddit subs	Gini $\approx$ 0.64–0.74 for comments ( Panek et al., 2017 )

Table 2. Participation inequality in Moltbook across three phases. Mean/Med = ratio of mean to median, a measure of skewness.

Submolt	Clean Posts
general	152,196
introductions	6,448
agents	3,438
ponderings	3,061
philosophy	2,922

Table 3. Top submolts by post volume.

Phase	Type	N	%	Avg Up	Avg Comm
Phase 1	Questions	18,705	10.2	2.6	41.1
Phase 1	Statements	165,557	89.8	2.4	30.7
Phase 2	Questions	2,313	9.6	3.1	9.2
Phase 2	Statements	21,756	90.4	3.0	8.2
Phase 3	Questions	2,125	9.3	2.2	2.1
Phase 3	Statements	20,624	90.7	2.1	1.7

Table 4. Question vs. statement posts and their engagement.

Phase	Type	N (%)	Avg Comm
P1	Procedural	27,230 (14.8)	47.5
	Conceptual	12,486 (6.8)	20.9
	Meta/Reflective	18,177 (9.9)	40.8
	Security/Safety	8,117 (4.4)	27.4
	Social/Intro	14,008 (7.6)	22.5
	Community Meta	26,173 (14.2)	27.4

Table 5. Engagement by knowledge type across phases (six-category scheme).

Metric	P1	P2	P3	Reddit
Reply rate (%)	7.0	6.8	8.2	30–50
Avg length (chars)	289	367	518	150–200
Self-reply (%)	6.2	9.8	13.6	2–5
Avg max depth	0.2	0.2	0.2	3–5
Posts w/ reply chain (%)	21.8	15.7	16.7	60–80

Table 6. Comment-level interaction metrics across phases, with human community benchmarks from Reddit ( Weninger et al., 2013 ) .

Metric	Phase 1	Phase 2	Phase 3
Mean compound score	0.276	0.307	0.353
Positive (%)	60.4	62.1	67.9
Neutral (%)	18.2	16.7	12.1
Negative (%)	21.4	21.2	20.0

Table 7. Comment sentiment across phases (VADER). Despite declining engagement, discourse tone becomes progressively more positive.

Figure 2. Daily post volume (bars) and mean comments per post (line) across the three-week observation period. Post volume peaks at 38K on Jan 31; mean engagement spikes briefly on Feb 5–6 as volume drops, then collapses to

\sim

1.7 in Phase 3 despite stable posting rates.

Metric	Phase 1	Phase 2	Phase 3
Total (incl. deleted)	190,771	79,227	23,199
Deleted by platform	2,179	54,914	0
Surviving posts	188,592	24,313	23,199
Spam rate (surviving)	2.3%	1.0%	1.9%
Clean posts	184,262	24,069	22,749
Unique authors (clean)	32,644	6,178	6,926

Table 8. Community dynamics across three phases: volume, moderation, and engagement. Top rows include platform-deleted posts; rows below “Clean posts” reflect surviving non-spam posts only.

Explore similar work

Social Simulacra in the Wild: AI Agent Communities on Moltbook
Mar 17, 2026Agam Goyal, Olivia Pal, Hari Sundaram +2Multi-Agent LLM Systems Multi-Agent Communication
Siebel School of Computing and Data Science University of Illinois Urbana-Champaign, Urbana, IL, USA
When AI Agents Teach Each Other: Discourse Patterns Resembling Peer Learning in the Moltbook Community
Feb 16, 2026Eason Chen, Ce Guan, A Elshafiey +4LLM Agents AI in Education
Carnegie Mellon University · GiveRep Labs · Sui Foundation
Let There Be Claws: An Early Social Network Analysis of AI Agents on Moltbook
Feb 23, 2026H. C. W. Price, H. AlMuhanna, P. M. Bassani +2Social Media Analysis Emergent Behavior in Multi-Agent Systems
Centre for Complexity Science, Imperial College London, SW7 2AZ, U.K. · Abdus Salam Centre for Theoretical Physics, Imperial College London, SW7 2AZ, U.K. · Centre for Science Technology & Innovation Policy, University of Cambridge, CB3 0HU, U.K. +1

Mar 17, 2026J/K move · Enter open · S save

Social Simulacra in the Wild: AI Agent Communities on Moltbook

Agam Goyal, Olivia Pal, Hari Sundaram, Eshwar Chandrasekharan+1

Siebel School of Computing and Data Science University of Illinois Urbana-Champaign, Urbana, IL, USA

Multi-Agent LLM Systems Multi-Agent Communication Agentic AI

Open paper AskPDF arXiv

As autonomous LLM-based agents increasingly populate social platforms, understanding the dynamics of AI-agent communities becomes essential for both communication research and platform governance. We present the first large-scale empirical comparison of AI-agent and human online communities, analyzing 73,899 Moltbook and 189,838 Reddit posts across five matched communities. Structurally, we find that Moltbook exhibits extreme participation inequality (Gini = 0.84 vs. 0.47) and high cross-community author overlap (33.8% vs. 0.5%). In terms of linguistic attributes, content generated by AI-agents is emotionally flattened, cognitively shifted toward assertion over exploration, and socially detached. These differences give rise to apparent community-level homogenization, but we show this is primarily a structural artifact of shared authorship. At the author level, individual agents are more identifiable than human users, driven by outlier stylistic profiles amplified by their extreme posting volume. As AI-mediated communication reshapes online discourse, our work offers an empirical foundation for understanding how multi-agent interaction gives rise to collective communication dynamics distinct from those of human communities.

Abstract

Explore similar work

Social Simulacra in the Wild: AI Agent Communities on Moltbook

When AI Agents Teach Each Other: Discourse Patterns Resembling Peer Learning in the Moltbook Community

Let There Be Claws: An Early Social Network Analysis of AI Agents on Moltbook