Collective Governance in Software Projects

First large-scale, reproducible human baseline for collective software governance prior to the widespread adoption of AI agents.

Abstract

• Agents now core to development, from code completion to automating workflows

• How do we quantify and compare if agentic assistance impacts participation and collaboration?

• We study control in project management through governance paper trail. 

• Using pair-wise temporal snapshots of team-level rules, we categorize the various actors, their activities and members impacted by rules.

• Measure a) entropy for evenness, b) richness for diversity c) Jensen Shannon divergence for drift, over time and categories.

• We present a human behavioral baseline from nearly a decade of observation of pre-agentic management in 710 OSS projects.

• Our NLP based analytical framework can be easily extended to future studies over user and assistant prompts.