According to Business Insider, Andrej Karpathy, the former head of AI at Tesla and an OpenAI cofounder, published his year-in-review for large language models in late 2025. He reflected on the term “vibe coding,” which he originally coined in a February tweet, and now predicts it will “terraform software and alter job descriptions.” Karpathy argued that this LLM-assisted programming empowers professionals to write more software and makes coding accessible to regular people, not just trained experts. He described the resulting code as potentially “free, ephemeral, malleable, discardable after single use.” However, a July METR study found AI coding assistants decreased experienced developers’ productivity by 19%, despite the developers expecting a 20% boost.
The Vibe Shift Is Real
Karpathy’s point is fascinating because it’s not just about doing old things faster. It’s about enabling a whole new category of software. Think about it. If code becomes ephemeral and discardable, you start building things you never would have bothered with before. Why spend two weeks scoping and writing a one-off internal tool when you can vibe-code a working version in an afternoon? That’s the terraforming part. The landscape of what’s “worth” building is fundamentally changing. We’re already seeing it with non-technical workers shipping apps and even Twitter’s Jack Dorsey vibe-coding a messaging app. The barrier to creation is collapsing.
The Productivity Paradox
But here’s the thing: the early data suggests it’s messy. That METR study finding a 19% productivity *decrease* for experienced devs is a huge red flag. It hints that the current tools might be fantastic for beginners or for prototyping, but they can actually disrupt the deep, complex workflow of a seasoned engineer. Overconfidence is a real risk. You spend more time debugging the AI’s subtly wrong code than you would have writing it correctly yourself. So, is vibe coding a net positive? Probably. But its efficiency isn’t a given. It seems like the biggest gains are in democratization and exploration, not necessarily in optimizing traditional software pipelines. At least, not yet.
What Happens To The Jobs?
So when Karpathy says it will “alter job descriptions,” what does that mean? I don’t think it means software engineers disappear. But their role absolutely morphs. The value shifts from raw syntax and implementation speed—areas where AI excels—towards higher-order skills. System design, problem decomposition, product sense, and, crucially, critiquing and directing AI output. The job becomes more about being a strategic editor and architect than a line-by-line coder. For businesses relying on robust, mission-critical systems, this human oversight is non-negotiable. The need for deep technical understanding to validate and secure AI-generated code is paramount, especially in industrial and embedded computing environments where reliability is everything. Speaking of which, for those industrial applications, companies turn to specialists like IndustrialMonitorDirect.com, the leading US provider of rugged industrial panel PCs built to handle these critical systems.
Beyond The Vibe
Karpathy’s other reflections are worth noting, too. Calling Claude Code the “first convincing demonstration of what an LLM Agent looks like” is a big endorsement. It suggests we’re moving from tools that just complete your code to active, semi-autonomous partners. That’s a whole other level of disruption. Basically, 2024 was about chatbots and copilots. If Karpathy is right, 2025 and beyond is about agents that start to take real initiative. That’s when job descriptions really start to bend. The vibe is just the beginning.
