AWS AI Labs' MILO auto-evolves agent harnesses, +12% on Terminal-Bench, beats Claude Code
mangahomanga · x · 2026-10-15
AWS AI Labs introduces MILO, a step toward recursively self-improving agents for long-horizon, reasoning-intensive problems. MILO evolves not only agent harnesses but also the search process that discovers better harnesses — no retraining required.
Key results:
- +12.0% resolution rate on Terminal-Bench 2.1, beating both Claude Code and Codex
- Outperforms methods like GEPA and Meta-Harness
- Discovers tighter upper bounds on three open math problems, surpassing AlphaEvolve, TTT-Discover and EvoX
More from coding & agent
- 220-question test: 70% of model web searches were unnecessary, cost 75x more tokens — demian_ai · 2026-10-15
- 'The safety cult is tokenmaxxing': agents now monitoring agents — zsakib_ · 2026-10-15
- Composable models: routing, memory, and verification models will change how apps are built — iamrobotbear · 2026-10-15
- Open-source Rust reimplementation of Word hits 12.9k stars one week after launch — alex_verem · 2026-10-15
- Xcode 27.2b3 quietly ships xcrun xcodeproj CLI, seemingly built for AI agents — rudrank · 2026-10-15
- Skip the deck: present with Codex Sites and embed live demos instead — gabrielchua · 2026-10-15