AutoSaddler Automates Agent Harness Repair, Boosting SWE-Bench by 9.6 Points
omarsar0 · x · 2026-08-25
Microsoft introduces AutoSaddler, a method to automatically optimize agent harnesses. It treats the harness as code, learning to patch it offline from failure traces by generating structured updates to prompts, tool configs, and control logic.
The system runs mini-batches of tasks, diagnoses failures, and validates updates. It achieved gains of 9.0 points on GAIA2, 9.6 on SWE-Bench Pro, and 10.0 on Terminal-Bench 2.0. The paper finds that deep debugging beats shallow reflection and targeted edits outperform unconstrained ones.
More from coding & agent
- Educator now uses Claude Code for all video graphics and end-to-end editing — HarperSCarroll · 2026-08-25
- 16 Types of RAG Architectures Shaping AI Innovation — goyalshaliniuk · 2026-08-25
- SpaceXAI explores using agents to transform design workflows — soleio · 2026-08-25
- fx v0.0.6: Remote HTTP MCP, 23% Faster Workflows — evilrabbit_ · 2026-08-25
- Production agents with write access: strict policy layers still needed beneath MCP — Flat-Eggplant-9890 · 2026-08-25
- Amp introduces Orbs: remote coding agents controllable from your phone — heyneighbor · 2026-08-25