Reproduce “XAgen: An Explainability Tool for Identifying and Correcting Failures in Multi-Agent Workflows” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xinru Wang, Ming Yin, Eunyee Koh, Mustafa Doga Dogan

An explainability tool named XAgen was developed to help users, regardless of AI expertise, understand, detect, and correct failures in multi-agent workflows. It integrates interactive log visualization, an LLM-as-a-Judge for automatic error identification, and human-in-the-loop feedback, demonstrating improved user understanding and debugging efficiency compared to baseline methods.

Key insights

  • Effective explainability for multi-agent systems requires shifting focus from individual model prediction explanations to dynamic, system-level behavior tracing and distributed failure attribution.
  • Formative user research with practitioners across varying expertise levels is critical for establishing empirically grounded design guidelines that address real-world debugging challenges.
  • Integrating visual log representations and LLM-driven error identification can significantly improve user understanding and localized failure detection in complex multi-agent workflows.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.