UW-Madison PhD Student Gabriel Orlanski Presents SlopCodeBench at Frontier Data Summit: A Benchmark for AI Coding Agents' Iterative Code Quality Degradation

Gabriel Orlanski, a PhD student at the University of Wisconsin–Madison and Project Lead for SlopCodeBench, presented his work at the 2026 Frontier Data Summit. He introduced SlopCodeBench, a new benchmark designed to test AI coding agents across iterative software development tasks, specifically examining how agents adapt a codebase as requirements evolve.

According to the publisher's video description, Orlanski's talk explains how SlopCodeBench measures "redundant code" and "structural erosion," comparing cumulative versus isolated task performance. The project received support through Snorkel AI’s Open Benchmarks Grants. For developers and researchers concerned with the long-term maintainability and quality degradation of AI-generated code, this resource offers a specific evaluation framework reference.