Harness-of-Harness (HoH) lets LLM-based coding agents autonomously develop software across iterative planning-coding-testing loops, balancing bug fixes against new capabilities and validating progress through independent evaluation rather than prescriptive workflows. The authors test HoH across three benchmarks and model pairings, reporting average performance gains of 52.25%. A case study has HoH develop a playable first-person-shooter game over more than 70 iterations.
