How models and their surrounding harnesses factor into AI agent quality

A user describes a paper by 17 researchers, led by the former head of Huawei’s Noah’s Ark AI lab. Its central claim, as the user presents it, is that AI-agent quality depends on both the model and the harness around it, not the model alone. The post points to a six-part harness, task-specific setups and SWE-bench results broken out by model and harness.