Harness-Zero: PKU, Google and HKUST Distill Agent Harnesses into Model Weights
AxSaucedo · x · 2026-09-28
Researchers from Peking University, Google, and HKUST introduce Harness-Zero, a paper on agent harness distillation.
Problem: External harnesses boost agent performance, but gains stay tied to the harness at deployment. The best harness varies by domain, instance, and model, so a general-purpose agent must settle for a suboptimal shared harness or route among many specialized ones.
Method: Use a domain- or instance-optimized harness as training-time guidance. A harnessing agent corrects student responses in the target harness's action space before execution, converting harness guidance into training demonstrations. Fine-tuning on these trajectories internalizes harness-induced behavior into the weights, so the specialized harness can be removed at deployment.
Findings (across knowledge work, tool use, and science domains):
- For frontier LLMs with the same evolved harness, agent-as-harness outperforms code-as-harness;
- After distillation, models retain the specialized harness's gains even with it removed.
More from coding & agent
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Engineer's Interview Prep Method: Use Claude to Find Which Answers Undersell Your AI Work — brandon_galang · 2026-11-16
- Generic API-catalog MCP server vs one tool per job: a context-bloat experiment — MandeepKaur9407 · 2026-11-16
- Berkeley paper: LLMs answer with stale values even when the update is in context — ombre-ai · 2026-11-16
- Effective HTML: 4.2k-star agent skills turn wireframes into self-contained interactive prototypes — tom_doerr · 2026-11-16
- skills-gate: one gate skill fronts 148 agent skills at just 389 tokens — Distinct-Pie2389 · 2026-11-16