Code Understanding is a Bottleneck for Coding Agents

2026-10-07

AI Summary

Existing code agent benchmarks struggle to disentangle the effects of editing and comprehension capabilities on performance. This work proposes CABRA, a novel synthetic evaluation framework based on call graph transformations that constructs multidimensional task sets via controlled variables to precisely quantify code comprehension difficulty. Large-scale comparative experiments involving large language models and coding agents reveal that external tools obscure inherent cognitive weaknesses in models, while complex logical analysis significantly degrades agent accuracy. These findings demonstrate that code comprehension, rather than editing volume, constitutes the critical bottleneck constraining agent performance, thereby establishing a new paradigm for agent evaluation.

Abstract

Repository benchmarks (e.g., SWE-bench) for coding agents often assume that lines of code edited can predict task difficulty, but such datasets' poor control over code and task types makes it hard to know which abilities truly drive agent errors. We present CABRA: a Coding Ability Blueprint for Rigorous Agent evaluation. CABRA builds tasks from scratch as call graph transformations and scales difficulty via a task size parameter on four axes: function traversal, search, runtime resolution, and instruction following. We run eight LLMs and six coding agents on 6,840 CABRA tasks to show: 1) LLM accuracy falls as task size~grows, but agents stay near-perfect by offloading work to tools (e.g., grep); 2) Larger CABRA tasks elicit more tool calls for reading and analysis, while a separate study on SWE-bench Verified shows these tool call counts predict agents' accuracy better than lines of code edited, suggesting task difficulty for agents can lie in understanding code to edit, not just in making edits; 3) Extending CABRA to an intense understanding task where models analyze divergent logic across two classes backs this finding, as agent accuracy finally falls. More broadly, we argue for synthetic evaluations like CABRA to unmask LLM weaknesses trivialized by tools (e.g., needle-in-a-haystack) and abilities beyond just editing (e.g., understanding) that coding agents still find difficult, pairing SWE-bench-style tasks with controlled diagnosis.

Problem

• Coding Agents
• Code Understanding
• Benchmark Evaluation
• Task Difficulty
• LLM Evaluation

Innovation

• Coding Agents
• Synthetic Evaluation
• Call Graph Transformations
• Code Understanding
• Tool Use