AI coding agents find the right file but miss the exact lines that matter, study shows

A study evaluating AI coding agents such as Claude Code and Codex using the new SWE-Explore benchmark reveals that while these models can accurately locate the correct files, they often fail to identify the specific lines of code that require attention. This highlights the importance of providing sufficient context for effective code repair, indicating that improvements in context-awareness are necessary for enhancing the performance of AI coding tools in practical applications.