Legacy code drives the modern world. IBM wants to change that with generative AI for code, targeting legacy COBOL code.
IBM has announced a new AI-based product called Watsonx Code Assistant for Z that aims to accelerate the translation of COBOL code to Java on IBM mainframes. The product uses an AI model with 20 billion parameters trained on more than 100 programming languages to refactor, transform and validate legacy COBOL code to reduce the costs and risks associated with mainframe modernization projects.
Watsonx Code Assistant for Z will be available in Q4 2023 and will be demonstrated at TechXchange in September.
COBOL still drives many organizations
COBOL (Common Business Oriented Language) is a programming language for business applications developed in the 1950s, and it is estimated that more than 800 billion lines of COBOL are used in production systems every day. The language is still widely used in financial institutions, insurance companies, airlines, retailers, and many other businesses.
Watsonx Code Assistant for Z is designed to help developers selectively convert COBOL business logic to Java while preserving mainframe functionality. The product is based on the watsonx.ai code model, one of the largest generative AI models for code, with knowledge of 115 languages and trained on 1.5 trillion tokens.
The generated Java code is optimized to work with other mainframe components such as CICS, IMS, and DB2 to preserve key functionality, and IBM says it is more useful than completely rewriting applications or converting COBOL directly to Java syntax.
Code AI could speed up development by 30 percent
According to a study by Gartner, AI assistants such as the Watsonx Code Assistant could reduce programming time by 30 percent in the future and be used by more than 80 percent of all developers.
IBM itself has been supporting this process for some time and, in May 2021, published the CodeNet large code database. Even then, the company stated code migration as an explicit goal.