IBM shares cratered, falling 13.2 per cent, after Anthropic posted its Claude Code tooling, which it claims can help modernise COBOL systems far faster than the traditional approach.
Investors treated that as a direct hit to IBM’s mainframe ecosystem, where COBOL applications still run core processes across banking, insurance and government.
The worry is not that COBOL vanishes, but that the costly, labour-intensive work of understanding and refactoring it becomes cheaper and quicker.
Anthropic argued that AI can automate large chunks of the exploration and analysis work that normally soaks up teams of consultants for years.
In its framing, timelines that used to run for years could be compressed into quarters, a claim that makes markets reach for the fire alarm.
IBM’s stock had already been sliding, and the drop left shares down about 27 per cent in February.
The move dragged on other parts of the market as investors tried to price in a wider “AI eats the middle” risk for legacy modernisation work.
COBOL’s early versions arrived in 1960 and its longevity comes from the fact that the code is tied to decades of business rules, data formats, integrations and audit trails that organisations cannot casually rewrite.
Modernisation is less about generating lines of code and more about proving equivalence, validating outputs, surviving audits, and migrating with near-zero downtime.
Anthropic’s announcement does not break IBM’s grip overnight, but it does challenge the assumption that modernisation must always be slow, bespoke and painfully expensive.
IBM now has to convince customers and investors that its mainframe value is in reliability, scale and integration, not in the scarcity of people who can read the old stuff.
