Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves

CodeScene reports that coding agents refactored a 300,000-line C codebase in three weeks, using a replay-trace harness to check behavior after each change and a code-health score as an optimization signal.

The case study is an open-source decompilation of Street Fighter III: 3rd Strike, not a production commercial codebase. CodeScene reports 2,903 commits, roughly $4,000 in model tokens, and an improvement in its own Code Health metric from 5.6 to 10.0. The agents also built a 22-recipe, codebase-specific refactoring playbook. These are vendor-reported results, not an independent replication; the proprietary quality metric and the unusual test setup limit what the headline alone proves.

InfoQ reports both enthusiasm for the frame-by-frame replay check and practitioner questions about whether the work was merged and how broadly it generalizes. InfoQ has no article comment section.