I think you're probably right. But fwiw, as a non-Rubyist who values good style and is recently working in a one-off Ruby codebase, I've found it easy to use LLMs to write Ruby code that is idiomatic and leverages built-ins well. I use ChatGPT 5 Thinking for this, and I don't let it generate any code that I use directly. I ask it about ways to do things, including stuff built-in to the stdlib, and sometimes have it generate 3 or 4 implementations. Then I compare them, consult the language docs or stdlib API docs or the pickaxe book, and choose what seems the most stylish or composable. I then write it out by hand, bearing in mind what I just learned, and see what Rubocop has to say about its style.

I wouldn't say LLMs have been essential in this, but they've been pretty convenient. Because Ruby has relatively a lot of special syntax, it's also nice to be able to inquire directly about the meaning of some syntax in a bit of generated code-- especially when it involves characters that are (for some reason) ungoogleable, like an ampersand.

I think people who use LLMs to generate code and people that embrace agentic coding AIs and "vibe coding" will absolutely fall into the pattern you describe. But RTFMers and developers who care about craftsmanship will probably use LLMs as another discovery mechanisms for the stdlib, popular Gems, and popular style conventions.