AgenticFlict: A Large-Scale Dataset of Merge Conflicts in AI Coding Agent Pull Requests on GitHub

Researchers Daniel Ogenrwot and John Businge have published a groundbreaking study analyzing how AI coding agents like GitHub Copilot and ChatGPT create integration headaches for development teams. Their new AgenticFlict dataset examines 142,000+ AI-generated pull requests (Agentic PRs) from 59,000+ GitHub repositories, processed through deterministic merge simulation. The findings reveal that 27.67% of these AI contributions cause merge conflicts—situations where automated code changes clash with existing code, requiring manual developer intervention to resolve.

This 8-page paper identifies 336,000+ specific conflict regions across 29,000+ problematic pull requests, showing that AI-generated code integration challenges are both frequent and substantial. The research marks a shift from studying AI's productivity benefits to examining its real-world collaboration costs in Software Engineering 3.0, where AI agents transition from assistive tools to active contributors. The dataset, available on Zenodo, provides the first large-scale empirical evidence that AI coding assistance creates new workflow bottlenecks that tools and processes must address.

The study's preliminary analysis indicates conflict rates vary significantly across different AI agents, suggesting some models handle code integration better than others. This research provides crucial data for developers evaluating AI coding tools and for tool creators needing to improve how their agents understand and adapt to existing codebases. As teams increasingly rely on AI for routine coding tasks, understanding and mitigating these integration challenges becomes essential for realizing the promised productivity gains of AI-assisted development.