Context management and planning improve coding agent efficiency and accuracy
An Empirical Study of Harness Design for Coding Agents
Artificial IntelligenceComputation and LanguageMachine LearningSoftware Engineering
Summary
Writing long computer programs is hard for automated coding helpers because they have to remember lots of information and plan their steps carefully. The authors studied how different ways of managing this information and planning steps affect how well these helpers work. They found that managing what information is remembered becomes more helpful when there is less space to keep it, and using simple rules before summarizing helps the most. Also, helping weaker models with predefined tools improves performance, while stronger models work well just using the bash command line. These findings help make better tools for coding agents depending on their strengths and limits.
What this means in practice
- •For software engineering teams: Optimize automated coding assistants to balance context memory and planning based on model strength and task budget.
- •For cloud platform operators: Reduce computing costs by deploying coding agents with minimal external tools when models have strong command-line skills.
Authors
Run-Ze Fan, Zihao Zhang, Simin Ma, Yebowen Hu, Shouju Wang, Kaiqiang Song, Fei Liu, Hamed Zamani, Xiaoyang Wang
Abstract
Coding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, yet existing work typically evaluates harnesses as monolithic systems, leaving the effectiveness of individual components unclear. To enable component-level comparisons, we study this question with a lightweight coding harness whose execution loop is fixed while three components are varied: planning, action space, and context management. Across four models evaluated on SWE-Bench Verified and Terminal-Bench 2.1, we evaluate 176 matched settings spanning five context-management strategies, four context-window budgets, and targeted ablations of planning and action space. We find that: (1) Context management becomes increasingly valuable as the context-window budget tightens, with most of its benefit coming from preventing context-overflow failures. (2) Staging rule-based elision before LLM-based summarization provides the strongest overall efficiency among the context-management strategies, whereas making elided content recoverable adds machinery that models rarely use and yields no accuracy gain. (3) Planning shifts from an accuracy scaffold for weaker models to a cost saver for stronger models, with little change in accuracy. (4) Predefined tools improve performance for models with weaker bash proficiency, whereas bash-capable models can operate effectively with a bash-only interface and achieve substantially lower cost, especially on command-line-centric tasks. Trajectory-level analysis explains these effects: context management extends execution trajectories without substantially altering agent behavior, planning changes where trajectories stop, and the action space changes the granularity at which code is written. These findings inform model- and budget-aware harness design and provide a modular framework for evaluating future harness components.