How IT pros should handle the great COBOL transition

There’s a phrase that comes to Paul Holland’s mind whenever he thinks of IT pros using AI to modernize internal systems: May you live in interesting times.

Periods of transition can be difficult, the Astadia CTO continued, and present challenges for teams working to shift organizational posture with the help of AI. This is especially true when it comes to COBOL, the outdated, mainframe-centric computer programming language that still maintains a presence throughout organizations from financial institutions to governments.

AI agents have been assigned to convert COBOL syntax into more accessible code languages like Java. It’s not exactly converting code, Holland explained.

“I’m going to look at this program, this code base, and I’m going to ask AI to build me out a specification for what it does, and then I’m going to give a coded assistant an IDE like Cursor or Copilot or Kiro,” Holland said. “I’m going to give it that specification and say, ‘Write me a Java program that fulfills that functionality’; reverse engineer to a specification, forward engineer to a new program.”

Status stable. COBOL’s continueding role underpinning systems makes it an important part of many tech stacks. But its age and relative capabilities require change; AI’s role in retranslating and replacing it with more agile languages like Java is both the key to the future and the difficulty. Anthropic’s Claude Code is already assisting in the effort, as IT Brew reported in February.

Further complicating things is the role of mainframes running COBOL. Physical infrastructure that needs to be updated is its own can of worms, necessitating deep changes to key aspects of the organizational budget. In addition, the mainframes running COBOL are often a vital part of mission-critical systems, such as state unemployment claims, and can’t be taken offline for very long.

As Agustin Huerta, SVP of digital innovation at Globant, told IT Brew different generations of mainframes make migration especially challenging as organizations move from hosted COBOL code to more modern solutions. What infrastructure is better than the existing mainframes, which are old but stress-tested by decades of operation, is an open question.

“I believe the bigger challenge with COBOL is the underlying infrastructure, and I’m not just saying this because mainframes are old; you have different generations of mainframes,” Huerta said. “The biggest challenge right now is who can, with a strong degree of confidence, say that their infrastructure is as reliable as a mainframe is.”

Get new. Modernization is front of mind—and has been for some time; engineers were telling IT Brew about the importance of the shift last year. In order to manage the multiple potential hiccups and complications, any IT team negotiating this transition needs a clear plan.

“If you can leverage AI to help you understand those complexities—extracting the specs, what is that code doing, how it is doing it, how the modules interconnect with each other, and how the data structures connect with each other—you have a cleaner roadmap of what you need to do,” Huerta told IT Brew. “You have a clear guidance of what needs to be performed, and so the risk of having an issue within the migration or of missing something while doing the migration is lower than what it was before.”

Ultimately, Holland said, the challenge is one which AI will solve—and may have already done so. The potential and capability is there, he believes; it’s just a matter of solving for complex implementation challenges.

“It’s not autonomous—that’s the big distinction at the moment, this is not autonomous agentic AI delivery,” Holland said. “This is agents making a massive difference in these projects, but these projects are still fundamentally large-scale human-led transformation projects; that hasn’t gone away today. Give it four years, it’ll be a much more autonomous event.”