What's the tea on harnesses?
Harness engineering alone can dramatically improve agent performance without changing the underlying model.
“we moved from 30th to 5th on Terminal Bench just by doing some harness engineering, without even changing the underlying model”
LangChain explains that a harness—the tools, execution environment, system prompt, and file system around a model—is now central to building capable agents, driven by stronger models and labs fine-tuning for these setups. They argue coding-agent patterns generalize to data analysis and research, and that tuning harness components like the system prompt and context yields large gains, citing a jump from 30th to 5th on Terminal Bench with no model change. It matters because it shows competitive agent performance is achievable through harness engineering rather than only proprietary model training.