The code holding the global financial system together was written before most of its current users were born — and almost nobody knows how to touch it anymore.
That reality has made COBOL modernisation one of the most expensive, slow-moving problems in enterprise technology. It has also made it enormously profitable for the consulting firms willing to wade into the mess. Now, AI startup Anthropic says it has found a shortcut — and the market responded by wiping out more than a decade of investor goodwill in a single session.
IBM shares fell 13% earlier this week, marking the company’s worst single-day drop in over 25 years. The catalyst was an Anthropic blog post claiming that its Claude Code tool can dramatically compress the timeline for COBOL modernisation. Where the process once demanded armies of consultants spending years mapping workflows, Anthropic argues that Claude Code can automate the exploration and analysis phases that consume the bulk of the effort, potentially shrinking multi-year engagements into quarters.
To appreciate why that claim hit so hard, consider the scale of what it’s aimed at. Hundreds of billions of lines of COBOL code run in production daily across finance and government sectors. The language handles an estimated 95% of ATM transactions in the United States alone. The core problem isn’t the code itself — it’s who understands it. The generation of developers who built these systems has largely retired, and that talent scarcity is precisely what made large consulting engagements from firms like IBM, Accenture, and Cognizant essentially unavoidable. Both Accenture and Cognizant also declined following the Anthropic announcement, a sign that investors are reassessing the entire consulting model around legacy modernisation, not merely IBM’s mainframe hardware.
What the selloff may have underweighted is that IBM has been making this exact argument itself for years. The company launched its own product — watsonx Code Assistant for Z — roughly three years ago with the same pitch: use AI to help translate and modernise COBOL. IBM CEO Arvind Krishna said as recently as July 2025 that the tool has seen strong adoption, with most customers using it to better understand their codebases and determine what to modernise and what to leave intact.
IBM’s Rob Thomas, Senior Vice President and Chief Commercial Officer, drew the sharpest line in response to the Anthropic post: translating code is one thing, modernising a platform is something else entirely, and the gap between the two is where most enterprises run into serious trouble. His argument is that IBM’s mainframe value has never lived in COBOL as a language. It lives in the vertically integrated stack beneath it — z/OS, transaction processing architecture, quantum-safe encryption, and decades of hardware-software co-optimization that no code translation tool is designed to touch.
Evercore ISI analyst Amit Daryanani reinforced that point, noting that clients already had the option to migrate away from the mainframe and have consistently chosen not to. The fear of displacement, in other words, may be running well ahead of the operational reality.
The pattern itself is becoming instructive. Just the week prior, cybersecurity stocks dropped sharply after Anthropic announced Claude Code Security, a vulnerability scanning tool. Each new AI capability announcement now functions as a trigger for market reassessment — which revenue streams could be compressed, and by how much? The pricing of fear is immediate, even when the underlying disruption remains contested.
IBM’s mainframe business and COBOL’s grip on critical infrastructure didn’t build themselves in a quarter. Whether they can be displaced in one remains a different question entirely.
Photo by Nick Karvounis on Unsplash
Source: Original reporting
