And generally anonymous computer scientists, mathematicians and other super smart people — a group that includes many women at a time when men dominated the workplace — are creating what will become the Common Business-Oriented Language, better known as COBOL.
A lifetime later, COBOL stands as an ancient language by the standards of digital technology. Yet it still helps to run computers used for banking, insurance and government — three industries not exactly known for their speed of change.
Now, though, AI, the newest hotter-than-hot technology, promises to help public agencies and others upgrade their COBOL codebases, as shown by a recent announcement from Anthropic.
A blog the company published on Monday promoted the use of the company’s Claude Code AI for “COBOL modernization.”
An Anthropic spokesperson told Government Technology the capability is not new, and that the blog was meant more as a marketing message than a product launch announcement.
Even so, the blog underscores the need for ongoing COBOL expertise in the public sector and reminds us that what was fresh in 1959 is something that few people have experience with in 2026.
“The developers who built these systems retired years ago, and the institutional knowledge they carried left with them,” the blog explains. “Production code has been modified repeatedly over decades, but the documentation hasn't kept up. Meanwhile, we aren't exactly minting replacements — COBOL is taught at only a handful of universities, and finding engineers who can read it gets harder every quarter.”
It’s hard to get a fix on how many COBOL experts are left in the public sector or elsewhere. For instance, NASCIO, the main group for state CIOs, does not keep such a figure, and general estimates of how many people in the U.S. can program COBOL vary widely, from as low as 24,000 to as high as 2 million.
Deaths and retirements keep reducing the roster of COBOL programmers. But COBOL persists. For example, the Anthropic blog said that 95 percent of ATM transactions in the U.S. run on COBOL.
According to a 2022 paper from the Association for Information Systems, while COBOL has “largely been abandoned in classrooms,” some 800 billion lines of the language still help run various systems.
In the public-sector space, tech leaders and government technology suppliers continue to to work on upgrading COBOL-based systems for courts and other tasks — and that includes Anthropic.
"From SNAP to the DMV, state and local governments keep essential services running for millions of people, and we're excited to help them modernize the legacy systems behind those services," Michael Lai, state and local government lead at Anthropic, told Government Technology via email. "Claude can take on the labor-intensive parts of IT modernization — documentation, unit tests and more — that made these projects slow and prohibitively expensive, supercharging government IT teams to tackle previously unsolvable challenges while saving taxpayer money."
The COVID-19 pandemic shined a light on how much government work relies on COBOL, and since then officials have taken steps — usually expensive steps — to get away from the language.
Oregon offers an example.
The state’s Employment Department recently spent at least $106 million on a new system to replace a COBOL-based operation — this happened after the state was all but drowning in jobless claims during the pandemic, and struggled to keep up.
Anthropic is among the tech companies that wants to play a bigger role in such efforts. Its sales pitch revolves around the difficulty of getting past COBOL relative to similar efforts.
“COBOL modernization differs fundamentally from typical legacy code refactoring,” the blog says. “You aren’t just updating familiar code to use better patterns, you’re reverse engineering business logic from systems built when Nixon was president.”
The company says its AI can automate much of that work while documenting inherent risks that come with upgrades and documenting “workflows that no one can remember. With AI, teams can modernize their COBOL codebase in quarters instead of years.”
