CMU Blog: Automated Harness Search vs Manual Tuning for Science Agents

CMU's Looni Lab compares automated harness search with manual tuning using Claude on drug design tasks, finding no universal winner but confirming that harness engineering alone—without retraining—can substantially improve science agents.

2026-10-03 ~ 2026-10-03 · 2 related posts

• Automated agent harness search vs human taste: no winner across drug design tasks — niloofar_mire · 2026-10-03
• CMU blog: harness engineering boosts scientific agents without retraining models — niloofar_mire · 2026-10-03