I recently saw an IBM presentation on how AI will help eliminate COBOL once and for all. Basically, they are trying to sell a tool that uses their WatsonX cloud service to convert COBOL to Java with the help of the current buzzword named generative AI. Let me tell you, COBOL has outlasted most of the tech that was invented since 1965 and I think it will outlast generative AI, too.
@nixCraft it it were as simple as mechanical translation, it would be done. way back in highschool we were hired one summer to write a mechanical translator converting a language called DB/C into C++. we did that. we made the generated code as idiomatic and tidy as we could, and we think we did a decent job.
@nixCraft the DB/C website is still around. it lists, we think, five customers. the company we spent that summer at used to be on that list; it no longer is. their million lines of code were successfully migrated.
@nixCraft unfortunately, what we solved was the easy version of the problem. our employer was a logistics company with a million lines of code that had been written in-house for its own operational needs. as long as the program still fills those operational needs, migration successful!
@nixCraft for a finance company to do it... well, just imagine the bar of testing they'd have to meet to conclude that it's safe to switch over. truly, we're not sure such a large scale software testing effort has ever been attempted. it's an existential risk for them; mess something up and they cease to exist and their executives may never work in finance again.

@nixCraft keeping COBOL around is the lower cost option, and certainly lower risk, and not due to the difficulty of the translation itself

also, LLMs cannot translate code. that is a lie. lolsob.