Reproduce “Beneath the Surface: What We Overlook When We Talk About Agent Evolution” 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

Xiaobo Wang, Zimu Wang, Siyu Ye, Jiaqi Li, Min Tang, Jun Bai, Yuling Shi, Ruobing Wang and 8 more

The authors examine agent-harness evolution, changing prompts, tools, or code while keeping model weights fixed. Across 16 evolved agents, each gains coverage on some tasks and loses it on others. One HyperAgents/MiniSWE case raises eight-pass reliability by 5.2 percentage points while losing ten covered tasks. EvoAudit revises A-Evolve into AGE, which improves both measures in four tested settings against that reference.

Summary by alphaXiv.

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