AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution Traces
2026-08-24 • Artificial Intelligence
Artificial IntelligenceComputation and LanguageMachine LearningMultiagent SystemsSoftware Engineering
AI summaryⓘ
The authors study why large language model (LLM) agents often fail at tasks that require many steps in a row, where small mistakes add up. They introduce AutoSaddler, a system that automatically improves the 'harness'—the control setup around the agent—by learning from failures in small samples. Their method treats the harness like code, fixes issues based on failure patterns, and chooses updates that improve overall performance. Tests show AutoSaddler boosts agent success rates by roughly 9-10% compared to original setups. They also find that effective improvement needs deep debugging, focused changes, and generalization-aware updates, highlighting a promising approach to making agents more reliable.
Large Language ModelsLLM AgentsHarness OptimizationLong-Horizon TasksFailure DiagnosisOffline LearningControl LogicAutomated DebuggingGeneralizationAgent Robustness
Authors
Sungho Park, Wonjoong Kim, Rongyuan Tan, Jue Zhang, Wook-Shin Han, Pengfei Gao, Chanyoung Park, Yongqiang Yao, Rao Fu, Elsie Nallipogu, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang
Abstract
LLM agents remain unreliable on long-horizon tasks, where small local failures can compound over extended interactions and lead to overall task failure. Although external harnesses can substantially improve robustness, harness design remains a manual and expensive process that requires searching over a large space of prompts, tool configurations, and control logic. We propose AutoSaddler, an automatic harness optimization framework that formulates harness improvement as an offline learning problem and iteratively updates the harness using failure signals from mini-batches. AutoSaddler combines failure-trace diagnosis, structured patch generation that treats the harness as code, and validation-based update selection. Experiments on GAIA2, SWE-Bench Pro, and Terminal-Bench 2.0 show that AutoSaddler substantially improves agent performance over the corresponding base harnesses, achieving gains of 9.0, 9.6, and 10.0 percentage points, respectively. Ablation studies further suggest that effective harness optimization benefits from three ingredients: deep debugging rather than shallow reflection, targeted modifications rather than unconstrained editing, and generalization-aware selection rather than trajectory-specific repair. Together, these results suggest that automatic harness optimization is a promising path toward more performant and reliable agent systems.