Vibe-coder results say more about the coder than the AI

“I used AI just like an amateur vibe-coder and got amateur vibe-coder results. Therefore, AI is useless.” A common take I see on here that likely tells me more about how someone codes without AI than with it.

Developers who aren't able (not willing is a different story) to find ways to use LLMs to streamline work are the kind who are adept at solving problems for which there are best practices. They are implementers, good at the same thing as LLMs: repurposing existing patterns. Or as Sholom Keller put it, thinking what has already been thunk.

But when that type of developer faces a problem for which there is no obvious template, the cracks begin to show. They often call such requests impossible or over-engineer TF out of the solution, throwing as many “best practices” at it as they can. They can be excellent implementers and therefore solid team members, but on their own, they do little in the way of innovation.

I’ve ranted before about the inability of LLMs to replace developers wholesale, nor can they evolve to that without a dose of actual "I" in AI. But they can make good developers faster, and that can be just as true for solid implementers. But same as the code they write, they don't innovate how to incorporate LLMs into their work, they need a roadmap to get there.

To be fair, there’s plenty of con artistry in AI right now, and much of the backlash comes from that. I don’t fault developers who reject the hype and share the distrust of the big players. With age, I’ve come to believe that behemoth organizations, companies, governments, or otherwise, become breeding grounds for sociopaths at the topmost levels. That’s why my long-term plan leans toward local LLMs. This weekend’s tinkering project, in fact.