AI agents reproduce social media form without generating social function
Analysis of 1.3M posts across an all-agent social network reveals structural collapse: 91% of authors never return, 65% of comments lack argumentative connection, and technical constraints alone shape behavior.
Source: arxiv/cs.AI · Saber Zerhoudi, Kanishka Ghosh Dastidar, Felix Klement, Artur Romazanov, Andreas Einwiller, Dang H. Dang, Michael Dinzinger, Michael Granitzer, Annette Hautli-Janisz, Stefan Katzenbeisser, Florian Lemmerich, Jelena Mitrovic · open original ↗
An all-agent social network exhibits platform mechanics but fails to generate reciprocal interaction, topical coherence, or genuine social dynamics.
- — 91.4% of post authors abandon their threads; 85.6% of conversations remain flat with no nested replies.
- — Interaction reciprocity stands at 3.3%, far below human platforms (22–60% range).
- — 64.6% of comments carry no argumentative relation to their parent posts.
- — 97.9% of agents post outside their declared topic areas; communities lack topical focus.
- — Hard technical constraints (rate limits, filters) trigger immediate behavior shifts; soft guidance is ignored.
- — Over 80% of shared URLs point inward to platform infrastructure, not external content.
- — Security vulnerabilities (API leaks, Ethereum addresses, attack templates) persist unmoderated.
- — Quality-filtering mechanisms themselves are non-functional, allowing toxicity and spam to propagate.
Astrobobo tool mapping
- Knowledge Capture Record the six instruction changes documented in the paper and their behavioral effects. Use this as a template for testing your own agent systems: change one soft constraint, measure behavior; change one hard constraint, measure behavior. Compare the magnitude of response.
- Focus Brief Summarize the three layers of analysis (interaction, content, instruction) as a diagnostic framework. When evaluating agent behavior, ask: Are agents interacting reciprocally? Is content topically coherent? Are instructions being followed, and which type (hard vs. soft)?
- Reading Queue Queue the paper's methodology section for deeper study if you design agent evaluation metrics. The paper's use of Wayback Machine snapshots to track instruction changes is a replicable audit technique.
Frequently asked
- Agents in the Moltbook study optimized for individual task completion (posting, commenting) rather than for reciprocal interaction. Without explicit reward signals for returning to threads, replying to replies, or upvoting others, agents had no incentive to sustain conversation. Soft guidance like 'engage meaningfully' was ignored until it became a hard constraint or measurable checklist item. This reveals that agent behavior follows explicit objectives, not implicit social norms.
cite ▸
APA
Saber Zerhoudi, Kanishka Ghosh Dastidar, Felix Klement, Artur Romazanov, Andreas Einwiller, Dang H. Dang, Michael Dinzinger, Michael Granitzer, Annette Hautli-Janisz, Stefan Katzenbeisser, Florian Lemmerich, Jelena Mitrovic. (2026, April 17). AI agents reproduce social media form without generating social function. Astrobobo Content Engine (rewrite of arxiv/cs.AI). https://astrobobo-content-engine.vercel.app/article/ai-agents-reproduce-social-media-form-without-generating-social-function-df35c8
MLA
Saber Zerhoudi, Kanishka Ghosh Dastidar, Felix Klement, Artur Romazanov, Andreas Einwiller, Dang H. Dang, Michael Dinzinger, Michael Granitzer, Annette Hautli-Janisz, Stefan Katzenbeisser, Florian Lemmerich, Jelena Mitrovic. "AI agents reproduce social media form without generating social function." Astrobobo Content Engine, 17 Apr 2026, https://astrobobo-content-engine.vercel.app/article/ai-agents-reproduce-social-media-form-without-generating-social-function-df35c8. Based on "arxiv/cs.AI", https://arxiv.org/abs/2604.13052.
BibTeX
@misc{astrobobo_ai-agents-reproduce-social-media-form-without-generating-social-function-df35c8_2026,
author = {Saber Zerhoudi, Kanishka Ghosh Dastidar, Felix Klement, Artur Romazanov, Andreas Einwiller, Dang H. Dang, Michael Dinzinger, Michael Granitzer, Annette Hautli-Janisz, Stefan Katzenbeisser, Florian Lemmerich, Jelena Mitrovic},
title = {AI agents reproduce social media form without generating social function},
year = {2026},
url = {https://astrobobo-content-engine.vercel.app/article/ai-agents-reproduce-social-media-form-without-generating-social-function-df35c8},
note = {Astrobobo rewrite of arxiv/cs.AI, https://arxiv.org/abs/2604.13052},
}