Anthropic's new AI tool can write 67-year-old COBOL code :-O
Posted by jacque on February 24, 2026, 2:15 pm Valued Poster
This is concerning in many ways.
Anthropic's new AI tool can write 67-year-old COBOL code, sending 115-year-old IBM's stock tumbling by 13% — IBM stock has worst day in 26 years, down 25% MoM and counting Bruno Ferreira
What do airlines, banks, and insurance companies have in common? Besides being an absolute pain to deal with, they all rely on COBOL and IBM mainframe computers as core infrastructure. The computing giant's stranglehold over those markets may finally begin to crack, though. Anthropic has announced COBOL-specific functionality for its Claude AI bot, and IBM's investors responded with a resounding 13% drop in stock prices.
For practical purposes, COBOL only runs on one type of system, supported by one set of people: IBM's mainframes and its engineers. That means the company has enjoyed multiple decades of telling clients how many zeros the gigantic bills will have tacked on for the next period. The unforgiving nature of that stranglehold means that any attempt at breaking it is most welcome by its existing customers, and a serious threat to IBM as a business.
If you've ever interacted with social security, public administration, healthcare, government, finance, insurance, automotive, retail, or airlines, you've touched a COBOL system at some point in your transaction, even if it was 30 layers deep. Similar to gravity, the language is invisible and yet affects every part of the modern world.
A cynical view of the situation would say that the systems running COBOL are meant to be 100% accurate 100% of the time, a notion that doesn't lend itself very well to the "probabilistically correct" that LLMs can offer. Even still, as I've attested myself, a good bot is a power multiplier in the hands of a competent developer, and can also lower the barrier of entry for young folks trying their hand at wrangling old systems.
The language harkens back to the 1960s, proposing itself as a human-readable language targeted at business transactions, using full decimal-point math as default, in contrast to the default floating-point math in other languages. True to its proposal, it revolutionized business computing, becoming entrenched across almost every sector of note, and was never truly replaced.
The situation doesn't just revolve around IBM's monopoly, though. Most well-versed COBOL programmers are retiring and dying, making their skills rarer and more expensive. The COBOL systems invariably run business-critical operations that cannot afford any downtime whatsoever, and are chock-full of proprietary data formats and business logic that is not documented, and understood only by a few greybeards — if at all.
If you're wondering why COBOL just wasn't up and replaced with something else, know that any rewrite attempt must (a) reverse-engineer miles-long business logic; (b) reverse-engineer the underlying data structures; (c) reimplement said logic and structures while being careful to always use fixed-point decimal math; and (d) execute a perfect transition with minimal to zero downtime.
Even when all those conditions can be true, COBOL systems are often so interconnected that it's unfeasible to replace just the one, as is the case with airlines. And heaven help you if you're in the financial sector, as you'll have to undergo extremely long-winded tests and audits, adding months to any deployment.
There's a well-known joke among programmers that almost certainly originated from COBOL: "When I wrote this code, only God and I understood what it does. Now... only God knows."
I tried learning it years ago but quickly decided that I'd rather not and learned DOS, Linux, & Ubuntu instead. (Most of which I've since forgotten). Ubuntu saved me after a nasty data hostage situation on my laptop followed a foolish mistake on my part. DOS is always handy in certain situations, but they happen so infrequently now that I just don't remember them well enough anymore.
In any case, I still recall the endless waiting for IT to write new code for the main frame computer in a manufacturing business I worked for as a Junior Industrial Engineer. We were trying to implement a computerized system in the manufacturing unit so that we could computerized the payroll insteadbof having 3 - 4 people doing piece work payroll by hand. Our new program would have allowed it to be done by one person, releasing the other three to do other office functions that needed help.
It took the IT department so damn long to just barely get a partial functioning that we ditched a year's worth of work on it in favoring of buying Microsoft computers (using DOS, windows was a year away), hiring techs to create the base of the program, then using Access to build our database that could interface with the base program. The whole thing was done in 7 months.
That was years ago when you could readily find someone well versed in COBOL and it still took forever to get so little functional code written to work with the system already operating there.
Re: Yeah, a LOT of systems are gonna fail when there's no one left who understands COBOL
When I was learning code and systems I remember looking at it but never worked with it. I was pushed toward newer languages because that was what would run the world. IBM definitely held a heavier hand than the newer big tech as far as legacy systems go.
I taught myself in Command line and Terminal type systems. In 1999 I decided to get credentials and pursue it professionally. Y2K was mostly resolved but uncertain. My curiosity about legacy systems workarounds and migrating to newer computing models was a big motivator.
I was also very impressed with GUI advances but wasn't intimidated by command line terminals.
I still work in MS, Apple, Linux environments. Mostly as Administrative and Network stuff. I should clarify besides my own stuff I am IT for friends and family. Charity work.
I did work writing code for a while but it was mostly project based. I acquired a broad view of inter-connectivity with Language, Applications, Systems, and more.
Life stuff changed things for me. I think about taking some refreshers and maybe get new credentials too. Mostly for learning purposes. Maybe earn a few bucks too.
Some older experience may need to be revived soon. A Jack of all trades is master of none, but oftentimes better than a master of one.
... And many thanks for the masters in skill for setting our standard of work to instill.
on February 24, 2026, 2:15 pm
