AI Agents Write Ruby but Can't Navigate It: A Benchmark Across 5 Models and 13 Codebases

This benchmark report evaluates five AI models across 13 Ruby or Ruby on Rails codebases. Its central claim is that code generation ability does not necessarily translate into effective repository navigation: agents may write Ruby while struggling to locate relevant implementations, understand project structure, or move across files. The supplied metadata does not include the model names, task definitions, scores, or detailed methodology, so the report itself should be consulted before drawing quantitative conclusions.