DemoEvolve: Overcoming Sparse Feedback in Agentic Harness Evolution with Demonstrations
The paper introduces DemoEvolve, a method for enhancing agent harness evolution using demonstrations. This approach addresses challenges posed by sparse feedback in long-horizon stochastic environments. The findings suggest that leveraging human trajectories can improve the stability and effectiveness of harness edits.
- ▪DemoEvolve aims to improve the adaptation of frozen language-model agents by modifying their external harnesses instead of updating model weights.
- ▪The method utilizes demonstrations from competent human trajectories to guide the evolution process when rewards are sparse and outcomes are high-variance.
- ▪Experiments indicate that DemoEvolve outperforms traditional self-rollout evolution methods in challenging environments.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24539 (cs) [Submitted on 23 May 2026] Title:DemoEvolve: Overcoming Sparse Feedback in Agentic Harness Evolution with Demonstrations Authors:Lirong Che, Yuzhe yang, Peiwen lin, Chuang wang, Xueqian wang, Jian su View a PDF of the paper titled DemoEvolve: Overcoming Sparse Feedback in Agentic Harness Evolution with Demonstrations, by Lirong Che and 5 other authors View PDF HTML (experimental) Abstract:Agent harness evolution improves frozen language-model agents by modifying the executable structures around them.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
