During an intro at a new client, someone asked me: "How do you prevent people from vibe coding?" I said, "Why would you prevent them?"

The room went quiet. The question told me everything. The name had already done its damage. A team that should be evaluating AI-assisted workflows was instead treating them as something to police. Not because they'd tried the tooling and found it lacking. Because the word "vibe" made it sound like something you'd catch a junior doing instead of working.

That's the problem. Andrej Karpathy coined the term to describe a specific workflow: give the LLM a direction, iterate, don't read every line. Playful and honest. But the internet turned it into a verdict. "Vibe coding" is now shorthand for "not real engineering." The name sounds unserious, instinct-based, amateur. Once something has that label, you don't need to evaluate what it produces. The name already told you it doesn't count.

The field moved on. The name didn't.

AI-assisted development in 2026 looks nothing like what Karpathy described. Agents run multi-step workflows autonomously. They read codebases, plan changes across files, run tests, interpret failures, and iterate without a human typing a single prompt per step. The developer's role shifted from "describe what you want" to "define the constraints, review the output, own the architecture." That's engineering. Calling it vibes is like calling a CI pipeline a guess.

The gap between the name and the reality keeps widening. The tooling matured. The workflows hardened. The results are in production. But the label stayed frozen in early 2025, and it's dragging perception with it.

Why the name matters

Names shape adoption. A CTO evaluating whether to invest in AI-assisted development hears "vibe coding" and pictures a junior prompting ChatGPT and shipping whatever comes back. That's not what they'd be buying. But the name made the decision for them before the evaluation started.

The label is costing the industry adoption. Teams that would benefit from agentic workflows, AI-assisted architecture, and LLM-driven test generation never get there because the framing makes the whole category sound unserious. The people who dismiss it fastest are often the ones who'd benefit most. They're not evaluating the tool. They're reacting to the name.

Every abstraction layer in computing history faced this. Assembly programmers thought high-level languages were toys. Backend developers thought frontend frameworks weren't real engineering. The pattern is always the same: a dismissive name buys time for the ego while the industry moves underneath it.

The ask

Drop the term. Not because Karpathy was wrong. He was describing a specific thing, and for that specific thing, the name fit. But the field outgrew it. What we're doing now is AI-assisted engineering, and it deserves a name that doesn't make decision-makers flinch.

The longer we call it vibe coding, the longer it takes to get serious about it.