Can AI Solve Legacy Tech Problems Like Translating Cobol To Java?

If this can be successfully scaled - which appears to be the case, it could provide one of the most financially and operationally significant productivity enhancements needed in tech - and one that few people outside of IT know departments know about. JL 

Belle Lin reports in the Wall Street Journal:

Generative AI is helping businesses modernize information-technology systems, some of them laden with code from Cobol, a programming language that predates the Beatles. Cobol underpins computer mainframes that IBM pioneered. Gen AI “translates” its mainframe code from Cobol—a language first designed in the 1950s and still commonly used at financial services firms—into Java, a programming language that dates to 1995. AI coding gives developers code suggestions and allows users to ask questions in plain English. Companies update legacy systems in a year or two, instead of several years. Gen AI can “understand the intent of the code” and turn it into usable Java right away.

Generative AI is starting to help business technology leaders with the long overdue task of modernizing their information-technology systems, some of them laden with code from a programming language that predates the Beatles.

“A big problem that we have in our space, and legacy companies have, is we have Cobol running around,” said Amin Venjara, chief data officer of 75-year-old payroll-processing company ADP. The number of developers conversant in Cobol is dwindling, he added. “Finding the Cobol engineers—how many people know that?”

The Roseland, N.J.-based company is exploring the use of generative AI to “translate” its mainframe code from Cobol—a language first designed in the 1950s and still commonly used at banks and financial services firms—into Java, a relative newcomer among programming languages that dates to 1995.

The translation would lessen the need to find and train Cobol specialists, who’ve become a relative rarity as a generation of experts retires and new programmers opt for modern languages like Python, or—in some cases—anything but Cobol.

Upgrading a business’s core technology, which can include updating software and moving to cloud-computing platforms, has always been a priority to chief information officers, but is even more crucial now as CIOs look to increase efficiency and keep IT costs in check.

In the past year, generative AI-based coding assistants from 

Microsoft

-owned GitHub, 

Amazon

, Google and 

IBM

 have emerged to help developers with tasks like auto-completing code snippets and writing code documentation. Some developers estimate the coding assistants raise productivity by about 25%, largely in tasks analogous to, say, spell-check and autofill in helping people write documents.

More recently, companies like ADP are using generative AI and similar coding tools to experiment with upgrading their old code, not just write new code—so far the most common use.

IBM’s watsonx coding assistant uses generative AI to help developers migrate code from Cobol to Java, or continue to use Cobol. PHOTO: IBM

IBM, which still relies on its mainframe business and supports a wide base of customers that depend on the large computers, is marketing its watsonx AI coding assistant to help customers quickly and easily address the foibles of their legacy tech.

“Our clients haven’t always invested as much in their applications over the decades, so that’s where they start to run into risks, skills challenges, knowledge gaps with very expansive applications with tens of millions of lines of code,” said Skyla Loomis, vice president of IBM’s Z mainframe software.

Launched last fall, IBM’s coding assistant uses generative AI to help developers migrate code from Cobol to Java, or continue to use Cobol—which the company says businesses aren’t getting rid of anytime soon. Cobol underpins computer mainframes, the large data servers that IBM pioneered. They require a great deal of maintenance, but still work well, companies say.

Like other AI-based coding assistants, IBM’s tool gives developers new code suggestions and allows users to ask questions in plain English. 

Loomis said the IBM coding assistant is expected to help companies update legacy systems in just a year or two, instead of several years. Compared with existing tools, generative AI can “understand the intent of the code” and turn it into usable Java right away, she said.

At Boston-based 

Wayfair

, generative AI-based coding tools are just starting to help the online furniture seller’s 2,000 developers and data scientists update old code. Wayfair is primarily using Google’s coding assistant, said Fiona Tan, the company’s chief technology officer.

 

Two-decade-old Wayfair doesn’t use Cobol, but has “legacy code” in languages like PHP, old database code in languages like SQL, as well as code written by developers who’ve since left the company.

“Over the years, you’ve built out code that hasn’t been well-documented,” Tan said. “It doesn’t even matter what language it is, it still takes a lot of time for somebody to learn it.”

Wayfair is counting on the AI tools to help reduce “technical debt,” or the flaws and costs generated when companies try to fix technology problems too quickly. With AI, engineers can more quickly learn new languages, thereby reducing technical debt, Tan said.

 

“As we get better and better at it, we’re going to see some really nice gains for people that have either stalled or been putting off a lot of their digital transformation work,” she said.

San Francisco-based Databricks is using generative AI to help engineers more quickly understand the data storage and management firm’s code base.

“Trying to pick up old codebases is a pain,” said Naveen Zutshi, the company’s CIO. “So being able to quickly understand what that code base does is very helpful for engineers.”

Upgrading code is just one piece of the bigger task of modernizing a technology system, cautions Arun Chandrasekaran, a Gartner analyst focused on AI and cloud. 

“If you’re really thinking of this as a complex workflow automation tool, you’re going to require multiple AI models,” Chandrasekaran said. “Some that are focused on code generation, but others that are focused on dependency mapping, impact analysis.”

But introducing any generative AI tool comes with its own set of risks—including the spawning of more technical debt. Making it easier to quickly write code also makes it easier for poorly-documented or extraneous code to creep in, requiring sharper oversight by humans, some tech leaders say.

“You’re going to have some tech debt if you want to move fast,” ADP’s Venjara said. “That’s the kind of balance that we’re finding ourselves in right now.”