Negotiating AI in Open Source Software Communities: A Case Study of the LLVM Project

Sammanfattning

Open source software (OSS) communities sustain critical infrastructure through volunteer contribution and review, yet AI coding tools are placing new pressure on the social mechanisms that reproduce these practices. This study examines how one such community collectively deliberates on the creation of an AI tool policy, examining how apprenticeship-based learning communities adapt to AI technologies. Prior research on AI in OSS has concentrated on technical and legal issues, with less attention given to how developers themselves articulate and respond to AI's integration. This study draws on Communities of Practice and Legitimate Peripheral Participation, frameworks to explore how situated learning can continue when newcomers depend their learning on AI tools rather than engaging directly with the community's core members. Employing a qualitative case study design, this research conducted a thematic analysis of 195 replies across four Request for Comments (RFC) threads from the LLVM Discourse forum, posted between May 2024 and February 2026. Participants were classified into groups based on forum trust levels and self-referential indicators within their replies. The findings reveal that four concerns persisted across the deliberation, Code Understanding, Accountability, Reviewer Protection, and AI Use Disclosure. Positioning varied by participation level, with Leaders acting as moderators, Core Members defending review as a relational mechanism, and Peripheral Participants negotiating their entry on contested terms. The tension of AI integration falls at the core-peripheral interface, where situated learning takes place. The principle of “human in the loop” is emerging as a shared repertoire articulating what these communities need to remain reproducible under AI pressure.