Salesforce: Co-evolving Harnesses and Models Shapes Fine-tuning Outcomes

Salesforce's new paper shows that the fit between an agent harness and the model determines fine-tuning outcomes: co-evolving harnesses with models, combined with local expert corrections, helps weak models overcome imitation learning shortcomings.

2026-09-10 ~ 2026-09-10 · 2 related posts

• Salesforce: Co-Evolving Harnesses and On-Policy Correction Help Weak Models Catch Up — Saleforce · 2026-09-10
• Salesforce paper: fine-tuning on evolved harnesses makes weak models worse on all 7 tasks — omarsar0 · 2026-09-10