There are billions of lines of COBOL code running on production systems worldwide, even though the language is over 60 years old and its primary architects have mostly retired or passed away. IBM is determined to keep those legacy functions on its Z mainframe systems and has previously attempted to have humans rewrite the code. However, now they have come up with a new idea: utilizing AI. The IBM watsonx Code Assistant, set to be available in Q4 this year, aims to involve humans while leveraging generative AI to analyze, refactor, and test the new object-oriented code. IBM claims that the watsonx-generated code should be compatible with COBOL and specific Z mainframe functions, making it a flexible process. In a technical blog post focused on COBOL conversion, IBM's Kyle Charlet, CTO for zSystems software, acknowledges the challenges associated with COBOL, such as the business logic, edge-cases, and lack of institutional memory. Some clients have invested significant time and resources in modernizing their COBOL code, but only a small portion has been successfully transformed. Rewriting COBOL requires meticulous manual work from numerous developers, and there have been various approaches to modernization with inconsistent outcomes.

IBM's watsonx can potentially assist large organizations in breaking down monolithic COBOL apps into individual services. IBM envisions this happening in three steps. AI assistance appears to offer a solution to the challenge of advancing COBOL while maintaining functionality. Although COBOL codebases can be relatively stable and secure, the costs of updating and extending them are substantial. Legacy COBOL was a contributing factor to the Office of Personnel Management's significant security breach in 2015, as the outdated code couldn't be encrypted or integrated with other secure systems. However, there is a recurring argument that COBOL effectively manages business-specific systems and exchanges, making them less vulnerable to attacks. It can also be argued that AI-generated and restructured code might appear correct and ready for testing, but without the presence of individuals who understand the code's underlying reasons, AI-scaled code could be as noisy as AI-scaled video. IBM's watsonx Code Assistant for Z will soon be deployed for Red Hat Ansible Light speed, and considering watsonx. ai was reportedly trained on more than 100 coding languages, it is likely that more AI co-pilots for old mainframe code will follow.