The Code That Runs Your Bank Was Written Before You Were Born. Nobody Alive Fully Understands It.

Key Takeaways

• •220 billion lines of COBOL are still in production globally, processing 95% of ATM transactions
• •Grace Hopper led the creation of COBOL in 1959, designing it to read like English for business users
• •Average COBOL programmer age is 55+. Most universities stopped teaching COBOL by the 2000s
• •Y2K (1999) cost $300B to fix and revealed how much critical infrastructure runs on ancient code
• •New Jersey's governor publicly asked for COBOL volunteer programmers during COVID-19 (2020)

Timeline

Every time you use an ATM, a piece of code written before most people alive today were born processes your transaction.

That code is written in COBOL: Common Business-Oriented Language. It was created in 1959. It was designed to be temporary.

Sixty-seven years later, COBOL processes 95% of all ATM transactions, 80% of in-person financial transactions, and the majority of government payment systems in the United States. Social Security, Medicare, the IRS, unemployment insurance: all COBOL. Major banks, insurance companies, and airlines: COBOL.

There are approximately 220 billion lines of COBOL code running in production systems right now. To put that in perspective: all the code ever written for all modern web applications, all mobile apps, all cloud services combined is a fraction of that number.

COBOL is the largest codebase in human history. And almost nobody is learning it.

The story starts with Grace Hopper.

There are approximately 220 billion lines of COBOL in production today. Every ATM withdrawal, every credit card swipe, every insurance claim, every Social Security payment touches COBOL code. The programmers who wrote it are retiring. The programmers replacing them were never taught it.

— Bryte, Time Machine

In 1959, Rear Admiral Grace Hopper, one of the most important figures in the history of computing, led a committee called CODASYL (Conference on Data Systems Languages) to design a new programming language. The goal was specific: create a language that business people, not just mathematicians, could read and understand.

Hopper believed that programming languages should look like English. She was widely mocked for this. "Computers can't understand English," her colleagues said. She proved them wrong.

COBOL reads almost like prose:

MOVE EMPLOYEE-NAME TO PRINT-LINE. ADD HOURS-WORKED TO TOTAL-HOURS. IF BALANCE IS LESS THAN ZERO PERFORM OVERDRAFT-PROCEDURE.

This readability was revolutionary. It meant that bank managers, government administrators, and insurance executives could read the code that ran their businesses. They couldn't write it, but they could audit it. This built trust. Institutions adopted COBOL massively throughout the 1960s and 1970s.

When IBM launched the System/360 mainframe in 1964, COBOL became the standard language for business computing. Banks wrote their core transaction systems in COBOL. Insurance companies wrote their claims processing in COBOL. Governments wrote their tax systems, benefits systems, and payroll systems in COBOL.

In 2020, New Jersey's governor went on live television and asked for COBOL programmers. Volunteers. The state's unemployment system, written in the 1980s, couldn't handle pandemic-era claims. Nobody on staff knew how to fix it. This is what happens when critical infrastructure outlives the people who built it.

— Bryte, Time Machine

And then the world moved on. But the COBOL didn't.

New languages emerged: C, Java, Python, JavaScript. New paradigms emerged: object-oriented programming, web development, mobile apps, cloud computing. Computer science curricula evolved. By the 2000s, most universities had stopped teaching COBOL. Students learned Java or Python. COBOL was "that old language from the mainframe era."

But here's what the industry forgot: nobody replaced the COBOL systems.

Rewriting a bank's core transaction system is one of the riskiest and most expensive projects in software engineering. The existing COBOL code handles millions of transactions daily with six-sigma reliability (99.9999% uptime). It's been tested by decades of real-world use. Every edge case, every rare scenario, every regulatory requirement has been patched into the code over 40, 50, 60 years.

Rewriting that in Java or Python means recreating all of that institutional knowledge. Every business rule. Every regulatory compliance check. Every edge case. And if you get even one thing wrong, money disappears. Transactions fail. Regulatory violations trigger fines.

So banks don't rewrite. They wrap. They build modern web interfaces and mobile apps that sit on top of the COBOL core. Your banking app is React Native on the front end and COBOL on the back end. The pretty interface is 2024. The engine processing your transfer is 1978.

This worked fine for decades. The COBOL systems are reliable. They're fast (mainframes are still among the fastest transaction processors on Earth). They're battle-tested.

The problem is the people.

The programmers who wrote these systems in the 1970s and 1980s are retiring. The average age of a working COBOL programmer is now over 55. Many are in their 60s and 70s. When they retire, they take with them not just the ability to write COBOL, but the understanding of what the code does and why.

Because here's the thing about 60-year-old code: it's been modified thousands of times by hundreds of different programmers over decades. Documentation, if it existed, is often lost or outdated. Business rules are embedded in the code itself, not in any external document. The code IS the documentation.

When the last person who understands a particular system retires, that system becomes a black box. It works. Nobody knows exactly how. Nobody wants to touch it because changing one thing might break something else. So it keeps running, untouched, processing billions of dollars in transactions, maintained by nobody.

This isn't hypothetical. It's happening right now.

In April 2020, COVID-19 triggered a wave of unemployment claims that overwhelmed state government systems across the US. New Jersey's system, written in COBOL in the 1980s, couldn't handle the volume. Governor Phil Murphy went on live television and asked for COBOL programmers. Volunteers. The state literally didn't have anyone who could fix its own critical infrastructure.

Similar emergencies hit Connecticut, Kansas, and dozens of other states. The systems that Americans depended on for survival payments were running on code older than most of the people filing claims.

The Y2K crisis of 1999 was an earlier warning. The "millennium bug" (two-digit year fields that would roll from 99 to 00, potentially crashing systems that interpreted 00 as 1900) affected COBOL systems disproportionately because COBOL systems were the oldest and most widespread. The global cost of Y2K remediation was estimated at $300 billion. Thousands of retired COBOL programmers were brought back to fix code they'd written decades earlier.

The lesson of Y2K was clear: critical infrastructure runs on ancient code, and we don't have enough people who understand it. That was 27 years ago. The situation has gotten worse, not better.

So what's the plan?

Some organizations are attempting modernization. The Commonwealth Bank of Australia spent $750 million and five years replacing its core COBOL system. The project was considered a major success, but it's also a cautionary tale: $750 million and five years for one bank.

AI-assisted code transpilation is the current hope. Tools that can read COBOL and convert it to Java or Python automatically. IBM, Amazon, and several startups are working on this. The technology shows promise, but converting 220 billion lines of code with zero errors is a challenge that makes Y2K look simple.

The most likely outcome: COBOL continues running for another 20 to 30 years. The people who understand it become increasingly rare and increasingly expensive. Critical bug fixes take longer. Modernization happens slowly, system by system, at enormous cost.

Grace Hopper wanted to make programming accessible to everyone. She succeeded beyond anyone's imagination. COBOL was so good at what it did that the world built its financial infrastructure on it and then forgot how to maintain it.

Somewhere right now, a 67-year-old programmer in New Jersey is the only person alive who understands a particular subroutine that processes $2 billion in transactions every day. When they retire next year, that knowledge walks out the door.

The code will keep running. It always does.

Until it doesn't.

(Sources: Reuters COBOL Reports, IBM Mainframe Documentation, ACM Computing Surveys, Micro Focus COBOL Survey, NJ Governor's Office, Commonwealth Bank Annual Report, Grace Hopper Archive at Yale)

Enjoy This Article?

RootByte is 100% independent — no paywalls, no corporate sponsors. Your support helps fund education, therapy for special needs kids, and keeps the research going.

How did this make you feel?

Recommended Gear

Keep Reading

Coding

Object-Oriented Programming Was Invented to Simulate the Real World — Then It Took Over the Real World

8 min read

Coding

Agent-Oriented Programming Was a Failed Academic Idea in 1993 — Now AI Is Making It the Future of Code

7 min read

Coding

Python Is Named After a Comedy Show, Not a Snake — and It Was Designed to Be Boring

6 min read