Harness-Zero: Harness Distillation via Agent-as-Harness
Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized
Reading in Plain English mode — toggle below to switch
Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized
This research introduces a novel approach to transfer the benefits of optimized agent harnesses into model weights, allowing for improved agent performance without being tied to a specific deployment harness. For ML engineers, this means developing more robust and adaptable agents that can maintain high performance across varying deployment environments, reducing the need for specialized, domain-specific harnesses at inference time.