AI Won’t Kill COBOL. It Might Finally Save It

Play Audio Version

Last year I wrote here on this blog that many companies are struggling to transform their AI tests and pilot projects into enterprise-wide initiatives. The problem is mainly around data. There is a need to modernize how data is organized before the greatest benefits can be achieved. Last year, MIT had reported that 95% of AI pilots were not scaling up to become full enterprise systems.

But I saw an article from February on the Anthropic blog that explored how AI could in fact be an important part of the solution – not just for modern systems, but also for legacy ones. AI can help with mapping and understanding data flows before any attempt at improvement.

The Claude Code blog was specifically looking at COBOL. Legacy systems using COBOL have always been notoriously hard to modernize for several reasons. First, it’s harder to find people with the skills needed to figure out what is happening inside a large COBOL system and secondly, a lot of these old systems comprise critical infrastructure that can’t just be switched off for improvement.

The Anthropic article suggests that it is often more expensive to try understanding the code than to redevelop a system – more time is spent just figuring out what is going on than time spent improving the system.

Over 95% of ATM transactions in the US still depend on COBOL systems so it really is everywhere and it is in systems that people really depend on. However, COBOL is almost seventy years old now. This twin issue of being embedded in essential services and the vanishing skills base are both very real problems.

Tools like Claude Code have been in the media frequently in recent months because there is a general fear about skilled tasks like coding computer systems being replaced by AI. But the Anthropic blog demonstrates that there is a very powerful way that human coders and AI can work together. Trying to understand code that may have been written decades ago.

Tools like Claude Code can take an entire COBOL system and reverse-engineer the code to create documentation that describes what is going on. This allows the human coders greater freedom to design how the system might be modernized safely.

This can also be an important process for the creation of a digital twin or support documentation. If the AI can systematically document every decision point in the system, it can dramatically improve the quality of a digital twin and give greater clarity to any team that needs to support the mainframe system.

COBOL modernization is a complex process. It’s not just a question of allowing coders to take a look at the software source code and then allowing them to make a few changes to make it run more efficiently.

The Anthropic blog describes the issue here: “COBOL modernization differs fundamentally from typical legacy code refactoring. You aren’t just updating familiar code to use better patterns, you’re reverse engineering business logic from systems built when Nixon was president. You’re untangling dependencies that evolved over decades, and translating institutional knowledge that now exists only in the code itself.”

Just to frame this precisely, they are talking about code that was probably created in the 1970s and then edited and changed ever since. Your business may depend on code that was written before most of your software engineers were ever born.

This is why the use of AI in COBOL and mainframe modernization looks like a game changer. It allows the software developers to find dependencies automatically. It allows them to build workflow maps from code that nobody understands or has documentation for. It can highlight risks that require more investigation before they cause a system crash and it can provide your development team with enough insight to make more informed decisions about where and how to modernize.

Many COBOL modernization projects have been avoided because the costs were prohibitive. AI is now moving the dial on this and making some projects possible. This is because all this planning, mapping, and insight that AI can provide dramatically reduces the risk of a mainframe modernization project.

It can also reduce the time to market for these projects. Modernization on some complex COBOL systems has previously been planned in years. Now it is feasible to plan for similar projects to be concluded in quarters or months. What was impossible is becoming possible – thanks to AI.

What Anthropic is suggesting aligns closely with some of the mainframe support and modernization work that IBA Group has been performing for years. Follow the links below for more information on how IBA has been using AI for mainframe modernization – and even for decommissioning systems that need their processes transferred to other applications.

For more information on mainframe support and modernization with IBA Group please click here.