Claude tackles ancient COBOL code, sends IBM tumbling 10%

If you withdraw cash from an ATM in the U.S., there is about a 95% chance that your transaction runs on COBOL code written decades ago.

Hundreds of billions of lines of COBOL still power the world's most critical infrastructure across banking, airlines, and government agencies. The catch? The engineers who wrote that code have largely retired, turning legacy system maintenance into a notorious black box.

For decades, that talent shortage has been an absolute goldmine for IT consulting giants like IBM, Accenture, and Cognizant. Whenever a company wanted to modernize, it had to hire armies of consultants for multi-year, multi-million-dollar discovery engagements just to figure out what the old code actually did.

Anthropic went straight after that business model.

Automating the consulting playbook

Anthropic announced Claude Code, a new AI tool purpose-built to streamline COBOL modernization, alongside a detailed Code Modernization Playbook.

According to Anthropic, Claude Code automates the most labor-intensive phases of modernization: mapping dependencies across thousands of lines, documenting workflows, and identifying hidden risks that usually take human analysts months to dig up.

The tool can find program entry points, trace execution paths, and map complex data flows across hundreds of legacy files. Anthropic claims this can compress modernization timelines from years down to quarters.

The logic is simple: legacy modernization historically stalled because understanding old code cost more than rewriting it. Anthropic argues AI flips that equation on its head.

Wall Street immediately felt the heat. IBM shares sank roughly 10% on Monday afternoon, with Accenture and Cognizant tumbling in tandem.

Why it matters

We think Anthropic just pulled off one of the smartest enterprise targeting moves in recent memory. Instead of fighting over generic web apps or trendy Python scripts, it aimed directly at the highest-margin, billable-hour profit engine of legacy enterprise consulting.

For enterprise tech leaders, this is massive. If your core systems are trapped in COBOL, your biggest fear has never been building the new stack—it has been breaking the old one without knowing why. Claude Code gives engineering teams a realistic shot at turning cryptic legacy codebases into documented logic without mortgaging their budget to external advisory firms.

For the legacy consulting industry, the threat is existential. Companies like IBM and Accenture rely on high headcount and long timelines to generate enterprise revenue. If AI can do the tedious forensic discovery work in minutes, the entire premise of multi-year legacy audit contracts starts to unravel.

That said, the real test will be trust. These systems handle trillion-dollar financial rails and flight routing. If Claude misinterprets a single legacy edge case, the fallout could be catastrophic. Enterprises will happily use Claude Code to read and map their COBOL today, but fully trusting an AI to decommission those mainframes will take far more convincing.