OO-Harness: Google paper boosts agent ~16 points by evolving harnesses, not models
alex_verem · x · 2026-10-24
A Google Cloud AI Research / UIUC paper shows you can gain 16 points on agent benchmarks without touching the model — just optimize the harness code around it.
Method
- An evolution agent uses execution records to grow a family of specialized harnesses via object-oriented inheritance
- A test-time router reads each task and picks the fitting harness
Results
- Business workflows: 61.62% → 77.41%; fully solved 25.44% → 37.22%
- Office file tasks: 25.45% → 39.48%
- On unseen apps the router falls back to base (59.18%) while the compared method drops to 44.90%
- Search finished in 29h vs 109h for the older method
Surprise: best-of-5 sampling (60.24%) scored below a single run on business workflows.
More from coding & agent
- Microsoft launches Agentic DevOps Engineer Expert certification via Exam AZ-401 — adnan_hashmi · 2026-10-25
- PL researcher admits Opus 5.5 is already a better programmer than him — tweetsatpreet · 2026-10-25
- Free LLM Dojo launches: 50+ interactive labs and 490 AI-engineering interview questions — sharpeye_wnl · 2026-10-25
- mlx-decision runs decision models locally on Apple Silicon, up to 6x faster than PyTorch — armbues · 2026-10-25
- How do teams share memory across multiple AI agents? — belhelha · 2026-10-25
- Vibe Monitor ships after 41 beta iterations: local stdio MCP adds architecture checkpoints to coding agents — Either-Alternative32 · 2026-10-25