I think there is more nuance or spectrum than good or bad. Vibe is one extreme, but along the dial from traditional to pure vibe are degrees of involvement. I’d characterize the degrees something like:

0. No AI, just elbow grease
1. AI as just auto complete on steroids
2. AI generating more complete change sets, but still from focused, more surgical specs, and still a human review on everything
3. “Spec-driven development” where, as I see it, you’re engineering a multi-agent-role workflow to intersect different contexts and iterating to try to converge on carefully designed specs

In 3 of those 4, the human is fundamentally the one owning the output, and AI is an accelerator and potentially an influence, kind of like pair programming. And even the SDD workflow can be a human-in-the-loop approach, although the more agents produce autonomously, the harder it might be for a human to be effective at reviewing the output.

So I’ll agree that “use it or don’t” is a binary, but I’d just add that there’s still a spectrum of how it’s used.