"Translating code is one thing. Modernizing a platform is something else entirely." That line, from IBM SVP Rob Thomas, neatly captures the argument IBM is making as AI-powered code translation tools gain popularity.
The pitch from several startups and cloud providers sounds straightforward: use AI to convert legacy COBOL code into Python, Java, or whatever modern language your team prefers. Problem solved, mainframe dependency eliminated. Thomas argues this fundamentally misunderstands what makes mainframes hard to move away from.
The Platform, Not the Language
COBOL itself is not the bottleneck. The real complexity lives in the integrated stack underneath it: z/OS, CICS transaction processing, IMS databases, Db2, and decades of tight hardware-software coupling. Thomas compares it to the iPhone-iOS relationship. You would not get the same experience by porting iOS apps to Android, because the value comes from how tightly the software and hardware work together.
There is also a scoping problem. About 40% of COBOL actually runs on distributed platforms, not mainframes. Lumping all COBOL modernization together conflates two very different challenges.
IBM's Counter-Pitch
Unsurprisingly, IBM's solution is to modernize on the mainframe rather than migrate off it. Their watsonx Code Assistant for Z focuses on refactoring, DevOps automation, and preserving institutional knowledge within existing mainframe environments. IBM cites one client, the National Organization for Social Insurance, achieving a 94% reduction in time spent analyzing and locating unnecessary COBOL code.
The self-interest here is obvious. IBM sells mainframes. Of course they want customers to stay on the platform. But the technical argument has merit regardless of the source. Anyone who has worked on a large migration knows that rewriting code is the easy part. Recreating the reliability, transaction handling, and data integrity guarantees of a mature platform is where projects stall or fail.
For teams evaluating AI coding assistants for modernization work, the takeaway is practical: code translation tools can help developers read and understand legacy code faster, but do not mistake that for a complete migration strategy. The platform dependencies, compliance requirements, and data sovereignty concerns that keep organizations on mainframes will not disappear because the code is in a different language.