IBM Corp. said today it’s using its recently announced generative artificial intelligence service watsonx to help enterprises modernize their most business-critical mainframe applications.

The new watsonx Code Assistant for Z is a generative AI-powered coding assistant that’s designed to help developers translate applications written in the older COBOL programming language to Java. The service can help accelerate COBOL app modernization while mitigating the risk of problems in that process. IBM said it will demo the technology for the first time at its IBM TechXchange event in Las Vegas next month.

Watsonx Code Assistant for Z is powered by IBM’s 20 billion-parameter watsonx.ai model, which is said to be one of the world’s largest generative AI foundation models dedicated to code automation.

The company believes that enterprises are desperate to modernize their mainframe applications to take advantage of benefits that include accelerating code development and boosting developer productivity. By writing their mainframe apps in Java instead of COBOL, organizations can access a much broader pool of information technology skills and accelerate developer onboarding.

That’s because there’s only a limited number of developers today who are familiar with COBOL, whereas just about every developer knows Java. In addition, Java-based applications are much easier to maintain.

The COBOL language is more than 60 years old, yet it continues to support many vital business and operational processes at organizations across the world. Translating these COBOL-based applications is difficult, because they amount to billions of lines of code that needs to be rewritten from scratch. By using watsonx Code Assistant for Z, companies will be able to transform COBOL apps and services incrementally into high-quality Java code, allowing developers to focus on more impactful tasks, such as adding new features to those apps.

Watsonx Code Assistant for Z is said to leverage IBM’s Application Discovery and Delivery Intelligence inventory and analysis tool to help companies identify the most pressing candidates for modernization. Using generative AI, it can perform tasks including refactoring business services in COBOL, transforming COBOL code to highly optimized Java code, then validate the outcome of this transformation through automated testing.

Constellation Research Inc. Vice President and Principal Analyst Andy Thurai told SiliconANGLE that finding skilled COBOL developers is one of the biggest challenges of mainframe app modernization. Very few developers are familiar with it, and many of those who were have long since left the workforce, he said. So when it comes to hiring developers who know both COBOL and Java, that’s even more difficult.

“This is where IBM’s watsonx Code Assistant for Z can be very useful for companies,” Thurai said. “These models are trained to understand mainframe COBOL applications and spit out object-oriented Java code that can be used to convert those mainframes into a modern program, making mainframe modernization easy.”

One of the reasons why Thurai is optimistic is that Watsonx Code Assistant for Z is unusual compared with other models. He said the main difference is that it’s predominantly trained with specialized code, specifically to fulfill the purpose of code generation, unlike other large language models, which are more general-purpose. “This LLM is task-specific, and can actually be classified as a specialized language model, or SLM, rather than LLM,” he said.

IBM said watsonx Code Assistant for Z can become the most viable option for many enterprises. While some companies choose to migrate their legacy mainframe apps to the public cloud, doing so means they have to sacrifice the reliability of IBM’s Z mainframes, which remain popular thanks to their reliability, stability and resistance to cyberattacks.

On the other hand, there are existing tools available that can convert COBOL to Java syntax, similarly to what IBM is proposing. However, the company said such tools are unreliable and often produce code that is difficult to maintain and sometimes even unrecognizable to Java developers. Generative AI is much more promising, and with watsonx Code Assistant for Z the first such model to offer COBOL support, it’s uniquely able to protect customer’s intellectual property, the company adds.

Charles King of Pund-IT Inc. said watsonx Code Assistant for Z is a great example of how IBM continues to evolve and improve its Z mainframe solutions to meet customer’s needs. It reflects an AI solutions strategy that’s focused on areas that can deliver clear and provable benefits to customers today, rather than promises of what might come tomorrow. “Rather than sitting passively while other modernization solution players go after it, IBM is actively disrupting itself and its offerings to benefit its clientele,” King said. “That’s just one of the reasons IBM Z continues to be the enterprise platform of choice for thousands of global organizations.”

According to Thurai, IBM’s overall aim is to make developers more efficient by transforming English itself into a programming language, and possibly other human languages in future. It’s doing this because it understands that developers are the key component of any software company, and there’s a need to make them more efficient so they can build things faster. However, he cautioned that generative AI coding models have a tendency to spew out code that’s sometimes vulnerable or infringes on intellectual property and licensing rights.

“Because IBM trained this model on its own mainframe code, there should be no legal or ownership issues that can arise from its use,” Thurai said. “However, the announcement and release are still in preview mode. Hopefully IBM can execute on this one, unlike many other failed programs they have announced before.”

Image: IBM