Claude Kills COBOL: FALSE

The day Anthropic announced it was training on COBOL, IBM's share price dropped. Analysts called it the beginning of the end for legacy infrastructure, but they were wrong. https://claude.com/blog/how-ai-helps-... And the reason is called Decision Velocity — the rate at which an organisation can actually act on information. And it's the concept that explains why AI capability keeps hitting an invisible wall inside the enterprises that need it most. COBOL systems in major banks don't just contain old code. They contain 40–50 years of accumulated business logic, regulatory exceptions, edge cases, and workarounds for systems that haven't existed since the 1980s. AI can translate the syntax in an afternoon. Reconstructing the intent correctly? That's a different problem entirely. And before a single line changes in a regulated environment, you need architecture review, security sign-off, risk committee approval, UAT across every downstream system, parallel running, rollback plans, and compliance documentation. The organisations that win the AI era won't have the best models. They'll be the ones who can decide and act fast, repeatedly, and confidently.