A Hacker News query asked COBOL developers how AI coding tools affect their work. The original poster noted that the massive COBOL codebase running critical economic infrastructure remains largely untouched by AI coding agents. This sparked discussion about whether LLMs pose a threat to these specialized roles or offer new efficiencies.

Commenters shared mixed experiences. Some found AI models poorly trained for COBOL compared to modern languages, though they remain useful for tedious tasks like parsing documentation. Others reported success using fine-tuned models for specific mainframe codebases. The consensus leans toward AI as an assistive tool rather than a replacement, requiring significant human oversight.

The conversation highlights a broader industry tension: legacy systems are notoriously difficult to automate due to their complexity and lack of public training data. While AI can handle routine tasks, the intricate business logic embedded in decades-old mainframe applications likely demands human expertise for the foreseeable future. The key question becomes how developers integrate these tools into their workflow.