Can COBOL’ers collab with Claude Code?

AI company Anthropic shook markets and mainframe pros alike in February when it announced that its Claude Code coding assistant could support IT pros’ attempts to modernize COBOL, one the oldest programming languages.

The claims—that Claude could act as a codebase-mapping, code-testing modernizer— led many a COBOL’er to respond: Whoa, wait a minute. It’s more complicated than that.

“The engineers doing this work know the code is the starting point, not the destination. What the application runs on, how it scales, how it recovers, how it is encrypted, and how it integrates with everything around it—that is the real modernization work,” Rob Thomas, SVP and software and chief commercial officer at IBM, wrote in a LinkedIn post the same day that Anthropic touted its COBOL feature.

“Translating code itself isn’t modernization,” John McKenny, SVP and general manager for intelligent Z optimization and transformation at BMC Software, told IT Brew. “Modernization is in the systems, in the architecture,” he added, citing decisions around platform choice, application integrations, and security. (He did some LinkedIn writing on the topic recently, too.)

What is COBOL? The “common business-oriented language,” known as COBOL and released in 1960, still powers critical applications in government and banking, among other sectors—functions that call for high uptime and often live on giant mainframes.

This decades-old programming language, while reliable, inevitably clashes with the modern age. For example, data from a banking mainframe may need to be moved to a cloud environment, or called by a mobile app’s API—a difficult task between wildly different systems.

Could an LLM actually help with modernization efforts?

Claude and proud. Claude Code offers its users the ability to turn their natural-language prompts (“find the files that handle user authentication”) into programming tasks like fixing bugs, mapping connections, identifying security risks, and refactoring code.

And the market seemed to buy the idea. The stock of IBM, which has built a significant consulting practice around mainframe modernization, fell this week as investors considered whether businesses would use Claude for COBOL instead. COBOL modernization is a “growing and important investment” for the company, according to Barry Baker, cChief oOperating oOfficer of IBM Infrastructure and GM of IBM Systems.; IBM also has segments in software, infrastructure, and consulting.

It’s also worth noting that IBM’s Q4 earnings revealed growth in its infrastructure division—up 21% from the previous quarter. “These are not the numbers of a company that got blindsided,” Michael Stricklen, managing director at EY-Parthenon, wrote recently as part of the mainframe-heavy conversations on LinkedIn.

The mainframe idea. An LLM can help with code translation and generation, according to Steven Perva, expert mainframe innovation engineer at Ensono, but code is just one aspect of modernization.

“That’d be like me distilling a skyscraper down to just a conversation about steel,” Perva told us, adding that technologies aiding overall modernization must also include a mainframe’s many connection points.

A mainframe system running critical infrastructure (and COBOL code) might need to connect to a backend database, local files, transaction processors (think: ATMs), or message queues.. There may also be TCP/IP connections linked to an Snowflake database or cloud environment.

“​​I understand that you most likely could teach these LLMs about these other systems, as well, but you really do have to make sure you’re being comprehensive about that,” he said.

“And when we’re talking critical enterprise infrastructure, 99.99% [uptime] isn’t good enough,” Perva said. Mainframe modernization efforts are often held up because organizations simply can’t risk taking mission-critical applications offline long enough to update code and hardware.

It’s elementary. IBM announced its own mainframe tool, the watsonx Code Assistant for Z, in August 2023. It offers capabilities like simple-language explanations of COBOL code. The on-premise option offers additional control for customers, according to Baker.

“A lot of our clients, in particular, are not entirely comfortable with opening up their code repos to a SaaS-provided third-party LLM that’s running out in the cloud,” he said. “Our strategy has been to enable our clients to leverage AI to modernize and get more and more out of the investment they’ve made.”