DataFlow-Harness closes AI data pipeline accuracy gap

Researchers from Peking University, Zhongguancun Academy, and Shanghai’s Institute for Advanced Algorithms Research launched DataFlow-Harness, an open-source framework for AI agents.

It guides agents to build governed, visual data-processing workflows from natural language, focusing on production-ready pipelines rather than disposable scripts.

In benchmarking, it achieved a 93.3% end-to-end pass rate and reduced cost and latency versus standard Claude Code.

The resulting artifacts are easier to audit and integrate, helping close the gap between code generation and enterprise-grade pipeline building.