Adaptive search branches improve agent code optimization by up to 35 percent
Mixture of Self-Improving Branches For Agent Harness Optimization
Artificial Intelligence
Summary
When computer agents try to improve themselves by changing their own code, they can get stuck choosing only one way to improve. The authors found that by letting these agents split into different groups that each try different ideas using feedback from their own successes, they can explore better alternatives. A system decides which group's method to use for new problems, combining their strengths without needing to know the final test results. This approach improved agent performance on math and coding tasks compared to previous methods.
What this means in practice
- •For ai software engineers: Optimize autonomous agent software by evolving multiple candidate improvements and selecting the best for different problem inputs.
- •For automated code reviewers: Improve the detection and proposal of code fixes by maintaining diverse code patch hypotheses evaluated on evolving test case subsets.
Authors
Haoyu Dong, Yuhang Zhou, Zihao Lin, Yifan Wu, Bo Peng, Mingyi Wang, Xiangjun Fan, Lizhu Zhang, Zhuokai Zhao
Abstract
Harness optimization provides a practical setting for recursive self-improvement (RSI), where agent-generated modifications inform subsequent changes through execution feedback. Recent work such as Meta-Harness implements this process through iterative code generation and evaluation, but retains a fixed development set and proposal policy. These constraints channel evolution along a single search trajectory, increasing the risk of converging to a local optimum. We make the improvement process itself adaptive by organizing search into branches with evolving development subsets and proposal policies. Each branch retains development cases solved by more of its leading harnesses than by those of other branches, drops cases solved by every leading harness across all branches, and revises its proposal policy using its own search history. To deploy the resulting complementary harnesses, we propose a router to select one development-selected branch head for each new input before execution. Across mathematical reasoning and agentic coding benchmarks, our system achieves relative improvements over Meta-Harness of 34.8% on Olympiad-level mathematical reasoning, 11.6% on Terminal-Bench 2.0, and 3.8% on SWE-bench Lite, with harness selection and router configuration based solely on development data. These results show that evolving branch objectives and proposal policies can yield complementary harnesses whose strengths a router combines without access to test outcomes.