AI Has Learned to Rewrite COBOL — and the Market Has Noticed

Two hundred twenty billion lines. That is how much COBOL code, by various estimates, still processes transactions at banks, insurers, government agencies, and payment systems worldwide. Created in 1959, a decade before the Moon landing, the language remains the invisible foundation of global financial infrastructure. And that foundation is cracking. Every year there are fewer specialists who can maintain it, and a botched replacement can cost billions. Now AI offers the first realistic way out of this dead end, and financial markets have already started recalculating their bets.

The COBOL problem is not just a technical curiosity for retrocomputing enthusiasts. It is a systemic risk that people prefer not to talk about out loud. According to Reuters, about 95 percent of ATM transactions in the US and roughly 80 percent of in-person financial transactions pass through COBOL systems in one way or another. The average COBOL programmer is well past sixty. Every year the industry loses people who hold critical knowledge: people who understand not just the syntax but the business logic that has piled up over decades across millions of lines of code, written without modern documentation, without version control, and often without any comments at all.

Modernization has been attempted many times. The largest banks have spent hundreds of millions of dollars on projects to rewrite legacy systems, and failed again and again. Commonwealth Bank of Australia spent more than a billion dollars replacing its core system, and its infamous failure became a textbook example. The hard part is not writing new code. It is faithfully reproducing the behavior of the old code, with all its non-obvious quirks, workarounds, and decades of accumulated business rules that nobody remembers anymore.

This is where large language models come in. Modern LLMs have shown surprisingly strong abilities to analyze and translate legacy code. They can read COBOL programs, reconstruct the business logic embedded in them, generate equivalent code in Java or Python, and, critically, explain exactly what each fragment does. In effect, AI acts as a translator between eras, compensating for lost institutional memory. Several companies have already brought specialized tools to market: IBM has built AI migration capabilities into its watsonx platform, and startups such as Phase Change and Modern Systems are raising significant funding on the specific promise of automated COBOL modernization.

The market responded with its usual bluntness. Shares of several consulting firms whose business was built on manual legacy modernization — multi-year projects with huge teams and predictably ballooning budgets — came under pressure. Investors are rightly asking: why pay thousands of consultants for a five-year project if an AI tool can do a significant share of the work in months? That does not mean human expertise is becoming unnecessary. Validation, testing, and integration still demand deep understanding. But the balance between manual and automated work on these projects is shifting radically.

Still, it is important not to get carried away. AI-driven COBOL modernization is not a magic button. Language models can hallucinate, producing code that looks correct but behaves differently in edge cases. In financial systems, where a one-cent error across billions of transactions can have catastrophic consequences, every line of translated code has to go through multi-layered verification. Regulators, from the Federal Reserve to the European Central Bank, have not yet developed clear standards for AI-assisted migration of critical infrastructure. That creates legal uncertainty that could slow adoption even where the technology is ready.

Even so, the direction is clear. COBOL is a trillion-dollar problem, and for the first time in decades there is a tool that can make solving it economically and technically feasible. For the financial industry, this marks the start of the largest technology migration in its history. For AI companies, it opens a huge and extremely well-funded market. And for the rest of us, it offers hope that the system carrying our salaries, payments, and savings will finally stop depending on code written before the internet was invented, and on a handful of specialists who still remember how it works.