3.52 Million Production Code Changes Analyzed: AI-Generated C++ Costs 5-8% More Compute, and Targeted Feedback Cuts Warnings 11.1%

arXiv 2608.06640·high signalXBlueskyLinkedInCopy link

A large enterprise with per-line production observability tracked 3.52M code changes across its brownfield C++ codebase from April 2025 to April 2026, comparing AI-generated against human-written code. AI code showed higher interface and coupling burdens, copy and allocation overhead, and a preference for explicit loops over optimized standard APIs — translating to more review effort and a measured 5-8% increase in compute resource consumption. Giving models targeted, taxonomy-informed feedback produced an 11.1% reduction in targeted static analysis warnings, making this one of the few studies with a concrete mitigation rather than just a complaint.