Google's OO-Harness paper: inheritance-based agent harnesses, up to 15.79% absolute gain
rohanpaul_ai · x · 2026-10-22
This is the paper link for the OO-Harness thread above, with the arXiv abstract (arXiv:2610.28380).
- Authors: Gaotang Li et al., 12 authors from UIUC and Google Cloud AI Research; 20 pages
- Agent harnesses are executable code specifying prompts, tools, and control flow around a model; a single global harness struggles with heterogeneous tasks needing different execution strategies
- During training, an evolution agent uses execution records to build a family of specialized harness classes via inheritance; at test time a harness router uses accumulated routing lessons to pick one per task
- Best performance across three long-horizon benchmarks (business workflows, workspace tasks, tool use) with up to 15.79% absolute improvement over the base, plus extensive ablations
Related event: Google's OO-Harness Paper Boosts Self-Evolving Agent Performance to 77.4%(3 posts)→
More from coding & agent
- OpenClaw opens its team environment to community contributors via Focus Weeks — RichardSocher · 2026-10-22
- ESA's cat keyboard incident is a warning for leaving AI agents unattended — vincent_koc · 2026-10-22
- Defense must run at machine speed: agents patch, engineers judge — saranormous · 2026-10-22
- TraceForge synthesizes reasoning traces from answers, beating Claude's own CoT by +3.7/+7.1 — IndependentResearcher121 · 2026-10-22
- Internet Archive fends off PHP attacks overnight, using Claude to patch old code — kingslyj · 2026-10-22
- Combining Ghidra, AI Reverse Engineering and WebGPU Sparks an "Absolute Renaissance" — thursdai_pod · 2026-10-22