GenAI and Mainframe COBOL Modernization

Generative and agentic AI applied to mainframe COBOL modernization is one of the most underrated shifts in the software industry. It's been about a decade since I last worked with IBM mainframes, so I was excited to join IBM zDay 2025 and see what's new.

COBOL Continues to Evolve

COBOL, like C and C++, continues to evolve. The latest COBOL-2023 standard added major features, and IBM's Enterprise COBOL compiler is finally catching up:

• Pass-by-value (2016)
• 64-bit addressing (2022)
• Free functions (2023)
• Major optimization and vectorization updates

These compiler improvements enable in-place refactors that can yield 30–50% performance gains and real mainframe cost savings.

How GenAI Changes the Game

Large-scale refactoring has long been hard for enterprise shops, but GenAI changes the game:

• LLMs accelerate onboarding into legacy code, helping developers understand unfamiliar systems faster
• They excel at rote tasks like test generation, reducing the manual burden of improving code coverage
• They can produce deterministic scripts for scalable, human-verified refactors

I can envision COBOL shops managing source the way Google uses clang-tidy for C++. With more fine-tuning, AI agents may soon propose "pull requests" for human review—a promising space to watch!