A recent study on arXiv (2607.28271) examined AI-powered migration of legacy COBOL programs to Java. The researchers found that the translated code often preserved existing bugs from the original COBOL. This means the AI replicated logic flaws, not just functionality. The study highlights both the potential and the limitations of automated code translation.
So the AI didn't just copy the code. It copied the bugs. On the surface, that sounds like a failure. But think deeper. Legacy systems are running banks, airlines, governments. They work, but they're held together with duct tape and prayers. Rewriting them from scratch is risky and expensive. AI translation offers a bridge.
Yes, bugs tag along. But now we have a modern codebase we can actually test, refactor, and improve. The alternative is leaving those systems untouched for another decade. This isn't a bug, it's a feature. It's a starting point, not an ending. We can finally see what's under the hood and fix it properly.