Reproduce “Towards Reliable AI Data Scientists: Data Agents with Workflow Harnesses” 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
Huachi Zhou, Yujing Zhang, Jiahe Du, Jiacheng Cai, Zijin Hong, Chuang Zhou, Zheng Yuan, Qinggang Zhang and 2 more
The paper surveys workflow harnesses for reliable data analysis, treating agents’ work as an iterative control process rather than a fixed pipeline. It organizes prior methods into five revisitable stages—perception, planning, execution, verification, and repair—with three technical routes per stage. The survey identifies four open reliability problems but reports no controlled accuracy or reliability improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.