The next time you use an ATM, transfer money between accounts, or swipe a credit card at a checkout, there is a strong chance that somewhere behind the scenes — invisible, silent, and almost certainly running on a mainframe older than most of the building it sits in — a programming language called COBOL is doing the actual work.
COBOL is 67 years old this year. It was designed in 1959, when the average car cost $2,200 and Eisenhower was president. Its co-designer was Rear Admiral Grace Hopper — the same Grace Hopper from the famous moth-in-the-relay story, who spent her career advocating for programming languages that ordinary people could read.
It was supposed to be a stopgap.
It has been running the world’s financial system, almost continuously, ever since.
The numbers
The standard figure for how much money COBOL handles every day is one of those statistics that sounds wrong the first time you hear it, and remains true anyway.
According to the Communications of the ACM, roughly $3 trillion in daily commerce flows through COBOL systems. Around 95% of ATM transactions touch COBOL code. Around 80% of in-person banking transactions go through it. More than 40% of U.S. banking systems are still built on it. Roughly 220 billion lines of COBOL code remain in active use across financial institutions worldwide.
The U.S. Treasury runs on COBOL. The IRS runs on COBOL. So does the Department of Veterans Affairs. So do major payroll systems, retirement systems, and large parts of the insurance industry. The Social Security Administration runs on it. If COBOL stopped working tomorrow, large parts of the American economy would go dark within hours.
To put the scale differently: this is a language most computer science students aren’t taught anymore, written for hardware that has been technically obsolete for decades, and it is currently the load-bearing software of the developed world’s financial infrastructure.
Why it never went away
The obvious question is why nobody has just replaced it.
The answer is partly that COBOL is genuinely good at what it does. It was designed for a specific job — moving money around reliably, in large volumes, with absolute precision — and it does that job exceptionally well. COBOL systems are, by reputation, more stable than most modern software. They were designed in an era when computer time was expensive and errors were career-ending, and they reflect that discipline.
The deeper reason is that replacing them is much harder than it sounds. A modern bank doesn’t run one COBOL program. It runs an enormous, interconnected web of programs, some written in the 1960s and modified continuously since, encoding decades of business rules, regulatory requirements, and edge-case handling that exists nowhere else. Untangling those rules and rewriting them in a modern language isn’t a software project. It’s an archaeology project, performed on the still-beating heart of the bank’s operations.
Several major banks have tried. The Commonwealth Bank of Australia attempted to migrate off COBOL and spent five years and roughly $750 million on the effort. Various U.S. banks have similar stories — multi-year, nine-figure migrations that either failed or finished years late. The cheaper option, almost always, has been to keep COBOL running.
But keeping COBOL running requires people who can read it.
The retirement cliff
Here is where the story gets genuinely worrying.
The engineers who built and have maintained the world’s COBOL systems were trained in the 1960s, 1970s, and 1980s. Most are now in their 60s and 70s themselves. They are retiring at a rate of roughly 10% per year, and they are not being replaced.
University computer science programs largely stopped teaching COBOL in the 1980s. The result is a generational gap: there are people in their 50s today who have never seen a line of COBOL code in their professional lives. The pipeline of new specialists has been essentially closed for forty years.
This was manageable as long as the existing workforce stayed on the job. It is now a problem that is becoming acute. Industry surveys cited by Metaintro’s coverage suggest roughly 60% of organisations running COBOL systems are struggling to find qualified developers. A company called COBOL Cowboys — yes, that’s the real name — has built a business specifically out of contracting retired COBOL programmers back into emergency consulting work, often at significant hourly rates.
When New Jersey’s unemployment system collapsed under pandemic-era load in 2020, the state had to publicly appeal for retired COBOL programmers to volunteer their help. That image — a government agency begging retirees to come back and fix critical infrastructure — is the future the banking sector is now staring down at much larger scale.
What banks are doing about it
The current response is a mixture of three strategies.
The first is paying the existing COBOL experts whatever it costs to stay. Senior COBOL contractors today can command rates that would have looked absurd a decade ago. Some retired programmers are quietly earning more in consulting work than they did in their full-time careers.
The second is training younger developers specifically in COBOL — an unusual reverse migration. A small but growing number of programmers in their 20s and 30s have noticed that COBOL skills are now genuinely scarce and increasingly valuable, and have invested in learning a language most of their peers consider professionally embarrassing. They are, slowly, becoming highly paid.
The third strategy is to bring in AI. Large language models, including modern ones, can read COBOL reasonably well, and several major banks are now using AI tools to help maintain and modernise their COBOL codebases. It’s an unusual moment in computing history: the newest and most fashionable technology in the field is being deployed, in significant part, to keep its oldest still working.
Why it matters
It is tempting to file all this under technological curiosity. It’s more serious than that.
For most of computing history, the assumption has been that software gets replaced. New languages emerge, old ones fade, and the underlying infrastructure of the world refreshes every couple of decades. COBOL is the clearest evidence that this assumption is wrong. Some software, once embedded deeply enough in critical systems, simply never goes away. It just acquires layers, and the people who understand it get older.
Sixty-five years on, COBOL is still running the money. The engineers who built it are nearly gone. What replaces them — younger developers, retired veterans, AI tools, or some combination — will quietly determine whether the financial system stays standing or not.
Most people will never know any of this is happening. That, in many ways, is the point.
