Researchers introduce MILO (Meta-evolutionary Island Orchestration), a framework that co-evolves agent harnesses alongside the search strategy used to discover them, combining hierarchical lineage memory over island-based trees, per-island mutator agents, and an adaptive orchestrator. Across Terminal-Bench 2.1, PaperBench, and DeepSWE, MILO-discovered harnesses outperformed eight state-of-the-art harnesses and six prior search methods, improving resolution by up to 28.3% when paired with the Opus 4.8 model. On Terminal-Bench 2.1, MILO reached 86.1% accuracy, surpassing the official leaderboard’s top entry while using 26% fewer tokens than its initial harness.