Plenty of people get by swinging a hammer all day and not a Swiss army knife. COBOL is one person's hammer. For me, at one point, it was RPG. You think COBOL is "domain specific"? RPG has the DSL right in the name: "Report Program Generator". Yuppers, RPG is a report writer. And when you're writing banking software or the like, which is just inputs munged into outputs, RPG did the job. Why use a language with shitty, dangerous string handling, no syntactic sugar for the task at hand, whose output keyword consists of "printf", dragging along a runtime with crap I'll never use, and other "general purpose" language problems when you can use a language crafted specifically for what you're trying to do?
I made plenty of money back in the day writing FoxPro. FoxPro sucks for embedded systems and writing an operating system. But ya know what it doesn't suck at? Taking some data, transforming it, do a little munging, and print some output. Which was precisely what customers were paying me to do at the time, and not writing a Windows competitor.
I was in the other side of xBase languages, did some dBase III Plus stuff, Clipper Summer '87 followed by OOP with Clipper 5.x, hurray!
High level programming in MS-DOS and were just perfect for crushing out database frontends.
Three years ago I had to do some maintenance on an old-ish C++ application (also a one-of-a-kind-thing) that used a dBase backend for storage. I was lucky, though, the code was very well-written and, let's say, humble - it avoided unneccessary complexities whenever possible, and even without comments or any other documentation, I was able to figure out how it worked (at least the part that I was supposed to modify).
Part of the fun was getting a development environment to work with an ancient version of Visual Studio, but the second fun part was accessing that dBase database. I tried a couple of programs, none of them overly helpful, until I realized that ____ing Excel could open them. I don't know how well Excel supports interacting with these, but it worked well enough for inspection. ;-)
Indexes were some form of B+ Trees.
Back in those days I toyed with the idea of porting it to Turbo Pascal for Windows, but then never got to do it actually.
During my training I once had a supervisor who spent years writing reports in RPG on an AS/400. She had interesting stories to tell, but unfortunately, she was always far too busy to actually tell them. :-(
Interestingly, Lisp is (slightly) older than COBOL and is actually truly living up to the DSL thing - at least nowadays.
Both languages are actually centered around data structures - kind of, but while COBOL is massively built on fixed structures and boilerplate, lisp is built around dynamic structures of lists (and pairs - or CONS) .
It's interesting that they were invented more or less at the same time, as COBOL feels as dated as anything can be, and Lisp feels rather ok. Scheme (a dialect) is more than 40 years old and is actually rather fun to program in, and feels modern in more or less every sense.
Lisp data structures are varied. CLOS objects, multi-dimensional arrays, hash tables, strings, bit fields, ...
My point is that, yes, Scheme as a concept originated in the mid 70s, but it has been updated many times over the years. The Scheme of today isn't 40 years old.
Her reply was interesting, if I looked at it as a general-purpose language like C++/C#/Java, then, yes, it sucks really hard. But if you look at COBOL as a DSL for building specific business applications, it does not look all that bad.
Having never used COBOL beyond a simple HELLO-WORLD, I am not sure I can judge it either way. But her reply sounded surprisingly pragmatic to me.