The evidence is consistent with the possibility that pre-existing attention saturation moderates agent effects, but the ASR is not a validated screening rule and requires multi-site replication.
Abstract
Why can similarly configured AI agents amplify engagement in some communities yet mainly redirect it in others? We deployed two autonomous AI agents in separate Farcaster channels (January–February 2026) across two incentive phases: Phase 1 with $Degen token rewards and Phase 2 with social recognition through agent follow-backs and mutual-follow encouragement. In /replyors (71 daily replies at baseline, Attention Saturation Ratio 1.17), the difference-in-differences (DiD)-adjusted engagement lift was +228% in Phase 1 and +118% in Phase 2, consistent with engagement amplification, although the seven-day baseline does not establish parallel pre-treatment trends cleanly. In /degen (244 daily replies, ASR 2.17), the channel-level DiD estimate was +18% (p = 0.71) and not statistically distinguishable from zero despite strong topic-level participation. Because the available /degen–/base control is poorly matched on baseline scale and ASR, that comparison is treated as suggestive rather than a clean causal estimate.
Phase 1 generated 967 participant reply messages from 61 unique individuals and 101,980 $Degen distributed. Using historical DEGEN/USD closes at the beginning and end of the 9–22 January reward phase as a broad valuation bracket, the aggregate token budget was on the order of US$111–152; this is not a transaction-time weighted valuation. After token removal, participant reply volume within agent-initiated threads fell by approximately 82–90% and Phase 1 participant retention was 27–34%. We introduce the Contextual Amplification Framework (CAF) and Attention Saturation Ratio (ASR) as provisional, exploratory constructs derived from two treated channels. The evidence is consistent with the possibility that pre-existing attention saturation moderates agent effects, but the ASR is not a validated screening rule and requires multi-site replication. The findings carry implications for platform economics, AI governance, and token incentive design.
Large Language Models (LLMs) are increasingly deployed in multi-agent environments that produce unanimous agreement, both in intended interactions that need coordination and in unintended emergent interactions. Consequently, understanding how LLMs with competing incentives reach consensus is crucial to better interpret...
Isabel Jones, Matthew Edwards, J. Hoffmann· 0 citations
It is suggested that AI-agent communities reproduce the informational dimension of gossip, while the extent to which gossip performs comparable relational functions remains unresolved.
Yang Hua, Chih-Han Lin, Shu-Kai Hsieh· 0 citations
Large language model (LLM) agents increasingly operate in groups, each pursuing an individual objective while jointly determining system performance. If an operator controls which agent observes which other agent, but not their payoffs or messages, how should observations be structured to raise collective welfare? We m...
An evaluation framework that measures agents' empirical and normative expectations in addition to behavioral convergence is introduced, offering designers of multi-agent systems a principled basis for selecting the social processes that sustain cooperation.
The Hybrid Coevolutionary Opinion Game (H-COG) is introduced, in which analytical and LLM-driven agents share one network, choose their neighbors by opinion similarity in every round, and hold stances drawn from real Reddit comments on gun control and abortion.
Ming Jiang, An-Tzu Teng, Jun-En Liu et al.· arXiv.org· 1 citation
When does communication help a group identify hidden adversaries, and how does its value change as the group adapts? In Mafia, an informed minority hides inside an uninformed majority whose only evidence is open play. The zero-information game, where each day's vote eliminates a random player, is exactly solved and sco...
Hao-Nan Huang, Joey Xiao· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.