AI agents write Ruby but can't navigate it: a 5-model, 13-codebase benchmark

A benchmark evaluates the ability of AI agents to navigate and find dependencies in Ruby codebases, comparing the performance of agents with and without a structural map of the codebase.