AI Pair Programming and Knowledge Sharing in Developer Communities

Abstract

With the increasing availability of artificial intelligence-supported pair programming tools in professional software development, past research has largely concentrated on their impact on individual developers. Much less is known about their potential impact on developer communities. Specifically, we still lack empirical insight into whether the proliferation of artificial intelligence coding assistants is associated with observable changes in community participation, collaborative communication, and externalized knowledge creation across large-scale developer platforms. In this paper, we introduce the Loneliness Framework, a socio-technical process model comprising six mechanisms that link AI pair programming to community-level behavioral change, and examine whether statistically significant longitudinal shifts in platform-trace measures of developer community behavior are temporally correlated with the adoption of artificial intelligence-supported pair programming. We employ an observational, non-causal longitudinal design and analyze two real-world data sets: a Stack Overflow data set from the Stack Exchange Data Dump and a public GitHub events data set from GH Archive, spanning January 2018 to December 2024. We analyze nine monthly behavioral metrics through trend, period-based comparison, correlation, and effect-size analyses, using false discovery rate correction for the primary hypothesis tests. We find that all nine metrics exhibit statistically significant longitudinal changes. The largest changes include decreases in question volume, pull request discussion density, and documentation-related behaviors, together with increases in median time to first answer. Many of these changes became most pronounced after late 2022, when large language model-based coding assistants became widely available. We also find positive relationships among selected documentation-related metrics, indicating coordinated shifts in externalized knowledge behaviors. Overall, the results indicate statistically detectable changes in developer community behavior during the era of artificial intelligence-supported programming. However, because this is an observational study, the findings should be interpreted in a descriptive, non-causal manner.