Reviving Old Code: IBM's AI Solution to Modernize COBOL
No attachments for this post
COBOL, a programming language with roots from 1959, remains astonishingly resilient. A 2022 survey disclosed a surge in COBOL's usage, from 220 billion lines in 2017 to over 800 billion in production systems. Despite its reputation for inefficiency and navigational challenges, moving away from COBOL is pricey and intricate. The Commonwealth Bank of Australia's shift from its core COBOL platform in 2012, for example, cost a staggering $700 million and took five years.
IBM has unveiled an innovative remedy to this legacy challenge. Enter "Code Assistant for IBM Z", an AI-driven tool poised to convert COBOL code into Java. Set to debut for general use by late 2023, attendees at IBM's TechXchange conference this September will get an early glimpse.
This new tool aims to help enterprises restructure mainframe applications while upholding performance and security, according to IBM Research's chief scientist, Ruchir Puri. The powerhouse behind Code Assistant is CodeNet, an AI model adept at grasping COBOL, Java, and approximately 78 other programming languages.
Puri highlighted the unique prowess of Code Assistant, stating it doesn't compromise COBOL's strengths and produces cost-effective, maintainable code, a trait lacking in many competitors. The system even recommends when certain segments of an application should remain in COBOL, ensuring a tailored transformation process.
However, Code Assistant is not without potential pitfalls. A Stanford study indicated that developers using such AI code generators might inadvertently introduce vulnerabilities. Thus, Puri stresses the importance of expert human review before deploying Code Assistant-produced code.
Despite potential risks, IBM views tools like Code Assistant as key to its forward trajectory, especially given that 84% of its mainframe clientele use COBOL, largely within financial and governmental sectors. IBM also envisions a broader landscape for AI-driven code generators, vying with platforms like GitHub Copilot and Amazon CodeWhisperer.
Comments on this post
No comments have been added for this post.
You must be logged in to make a comment.