To be fair to CF, it’s really easy to write server-side rendered CRUD apps in it, and there’s a lot of functionality baked it to support it. But RoR is just as easy, and lacks the, ah, legacy aspects of the CF language and runtime that are so troublesome. To put it another way, CF is a syntactic and semantic mess by design, a holdover from late-‘90s development, and attempts to improve it largely consist of throwing increasingly baroque structures on top of it that no one really uses in practice. (And the class structure the generated Java uses is enough to make a grown person weep and reach for the whiskey.)
But it’s also a development paradigm that is easily comprehended and encompassed by a single developer, which is not something you can say about modern development architectures. You don’t need to understand front-end toolchains or VDOMs, you don’t need to figure out how to make reverse proxies and API gateways work, you never have to use OpenShift or even understand what a container, let alone Kubernetes, is, and you don’t need to stand up logging and analytics servers. You can do a lot in CF with a little. Granted, oftentimes it’s a lot of damage, but there’s definitely a lure to having such a simple development environment even if you’re trading off little things like, say, code encapsulation and separation of concerns.
Having said all that, the “70% of major companies” argument is bunk. Choose any legacy technology and you’re likely to find it in 70% of major companies just through the inertia of “if it ain’t broke.”