When Bots Join the Team: Bot Adoption and the Institutional Fabric of Open-Source Software Projects
Citation. Shi, Y., & Gong, W. (2026). When Bots Join the Team: Bot Adoption and the Institutional Fabric of Open-Source Software Projects [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.13679
Summary. Shi and Gong study 2,991 open-source software projects around their first adoption of a rule-based bot, treating bots as visible participants in a shared interaction record rather than as tools used privately by individuals. Adoption coincides with more repeated human collaboration and bot-directed social memory, fewer conflict cascades, and more distinctive project outputs. The authors interpret these as behavioral signatures consistent with reinforced institutional coordination, while explicitly limiting the event-study evidence to precisely timed associations because it has no untreated comparison group.
Key claims
- Across 12-month windows before and after adoption, repeated engagement increased from 0.224 to 0.400 (Δ = +0.176, Cohen’s d = 0.69) and bot-directed social memory increased from 0.016 to 0.062 (Δ = +0.045, d = 0.54; both p < .001).
- Conflict-cascade rates fell from 0.642 to 0.229 (Δ = −0.413, d = −0.28) and output distinctiveness rose from 0.283 to 0.499 (Δ = +0.215, d = +0.49; both p < .001); adoption-month estimates indicate an 18% cascade-rate reduction and a +0.332 SD distinctiveness increase.
- In joint models, repeated engagement, bot-to-human addressing, and role differentiation were associated with fewer conflict cascades, whereas role differentiation, bot-directed social memory, and inter-bot interaction were associated with greater output distinctiveness.
- The bot-intensity association with conflict cascades attenuated from −0.354 to −0.015 after controlling for repeated engagement and role differentiation, while its association with output distinctiveness remained nearly unchanged (+0.162 to +0.156).
Evidence & limitations
- The preprint uses archival GitHub interaction data from 2,991 repositories observed for two years before and after first bot adoption, with paired pre/post comparisons, project fixed-effect event studies, and cross-sectional regressions.
- The study lacks an untreated comparison group and adoption may be endogenous; its authors therefore do not make causal or mediation claims, and the results concern predictable, domain-specific, non-conversational bots in established open-source projects rather than more capable agents or other workplace settings.
- Public provenance is verified through arXiv metadata and an exact SHA-256 match between the Drive artifact and the public arXiv PDF.