It's an open secret that the institutions keeping the global financial system turning over run code that is ancient, barely understood, and frighteningly hard to replace. Now, AI is finally making that problem solvable – and the market has responded with a reality check for one of technology's oldest names.
IBM shares recorded their worst single-day drop in more than 25 years earlier this week, plunging 13% after AI startup Anthropic said its Claude Code tool can accelerate COBOL modernisation – the kind of painstaking, expensive legacy work that has underpinned a portion of IBM's consulting revenue for years.
An Anthropic blog stated that "modernising a COBOL system once required armies of consultants spending years mapping workflows," and argued that tools like Claude Code can now automate the exploration and analysis phases that consume most of the effort in COBOL modernisation. That single claim was enough to send investors reaching for the sell button.
COBOL is bigger than most realise
To understand why the reaction was so sharp, it helps to understand just how entrenched COBOL remains. Hundreds of billions of lines of COBOL code run in production daily, powering critical systems in finance and government sectors. The language handles an estimated 95% of ATM transactions in the US alone.
The deeper problem isn't the code itself – it's the people who understand it. The number of developers who understand COBOL continues to shrink as the workforce that built these systems has largely retired. That talent scarcity is precisely what made COBOL modernisation so expensive for so long, and what made large consulting engagements – the kind IBM and rivals like Accenture and Cognizant built profitable practices around – essentially unavoidable.
Anthropic argues that AI flips this equation entirely. Claude Code works by mapping dependencies in thousands of lines of code, documenting workflows, identifying risks faster than human analysts, and providing teams with deep insights for informed decision-making. The company says teams can now modernise COBOL codebases in quarters not years.
IBM was already here
What the market's reaction may be overlooking is that IBM itself has been making this argument for some time. Anthropic's post comes about three years after IBM itself suggested using AI to rewrite COBOL as Java and created a product called "watsonx Code Assistant for Z" to do it. IBM CEO Arvind Krishna said as recently as July 2025 that the company's AI coding assistant for mainframes "has got very adoption," with the majority of customers using it to understand their COBOL codebase and decide what to modernise.
