Harness optimization is a measurable model capability, and the optimizer model matters more than the harness it runs through.

HarnessOpt-Bench (arXiv 2608.06301) has a frontier LLM act as an optimizer receiving a target agent's seed harness (prompts, tools, control flow, memory, orchestration code) plus graded eval feedback and a fixed evaluation budget, then edits and nominates a candidate scored on a held-out partition. Across 5 frontier models, 4 downstream tasks, 111 scored runs, optimizer models separated more than the coding harnesses they acted through, and native harnesses didn't consistently beat shared ones. Practical read: when you hand your agent scaffold to a model to improve, spend on the strongest optimizer rather than standardizing tooling around it. And always hold out a test partition, because self-graded harness edits overfit the feedback set.