AI · 8 min read · April 17, 2026
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.
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.