I kinda want authors to define "vibe coding" in their specific context when they write articles like this.
Is it "try to one-shot something via a Web UI" or "Figure out proper spec with Claude in Plan mode, let it implement according to the plan"? Both give completely different results.
And it doesn't matter a bit if LLM produced code isn't deterministic. We have 100% deterministic tools to check the code. Have had for decades. Agentic LLMs might produce bad code, but they can also run deterministic checks on the output and fix them immediately before even bothering the user.