IBM stock was down 10% on Monday afternoon after Anthropic published a blog post about how its Claude Code tool can be used to modernize software written in the COBOL language, which handles large-scale batch transactions. Many of the software systems used by the federal government, banks, and airlines are written in COBOL (“Common Business-Oriented Language”), and most of those systems run on IBM mainframes.

IBM also generates revenue from servicing, modernizing, and consulting on those mainframes. If COBOL code were converted to a more modern language, the systems would likely migrate to newer cloud servers.

But modernizing COBOL—which was developed 67 years ago—is a slow and expensive process, largely because the code can be difficult to understand and easy to break. It often reflects decades of institutional knowledge and workflows, and is frequently poorly documented—meaning its true intent can only be uncovered through close analysis. These challenges are compounded by the shrinking pool of programmers who know COBOL. Most university computer science programs no longer teach it.

Anthropic says this analysis phase is the most time-consuming and costly. That’s where Claude Code comes in. The tool can uncover and document workflows hidden within the code, identify dependencies across different parts of a code base, and give engineers insights into how to redesign systems.

“With AI, teams can modernize their COBOL code base in quarters instead of years,” the company writes in the blog post. 

IBM says the analysis phase is not the hardest part. “Translating COBOL is the easy part—the real work is data architecture redesign, runtime replacement, transaction processing integrity, and hardware-accelerated performance built over decades of tight software and hardware coupling,” an IBM spokesperson said in an email. “That is the problem IBM has spent decades learning to solve, and AI is the most powerful tool we have ever had to do it.”

COBOL was developed in 1959 via a public-private partnership that included the Pentagon and IBM, with the goal of creating a universal, English-like programming language for business applications. But private-sector companies have largely moved away from it. The code is difficult and costly to maintain and was designed for batch processing, making it poorly suited for modern cloud-based and real-time applications. (Anthropic and IBM did not immediately respond to requests for comment.)

© Fast Company