I have no way of knowing how ubiquitous the concept of “vibe coding” is anywhere else. In the Bay Area, it seems everyone knows how to code. It’s almost a qualification for living here. I’m married to a software developer. He’s been writing code since Nirvana was on the radio the first time and it wasn’t the classic rock station. When I first moved to San Francisco 13(!!) years ago, it was rare to get through a conversation with a new acquaintance without them saying, “So I have this idea for an app.” Now, it’s AI-this and AI-that. So maybe my desire to dive into this is a little esoteric, but anyway, here we go.
What is vibe coding?
Basically, vibe coding means that you describe to AI, in plain English, the features you’d want in an app, and it builds the code for you. AI is translating your requests into the code needed to build software for you.
I do not know how to code, save from some clunky html for my website. So if I had an idea for an app or a project that required coding to implement, the concept of vibe coding seems like it would allow me to build it. This seems pretty great - a barrier removed. I wouldn’t have to bribe my husband to do it, or otherwise hire someone.
The term “vibe coding” was coined by computer scientist Andrej Karpathy, a co-founder of OpenAI and former AI leader at Tesla, in February 2025. He said, “There’s a new kind of coding I call “vibe coding”, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists… I’m building a project or webapp, but it’s not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
The amount of money is staggering. For an industry that didn’t exist just a few years ago, the estimated market value for vibe coding agents in 2024 was nearly $3 billion. Companies like Anthropic that are in a broader generative AI business have offerings specific for vibe coding; other companies are themselves built around the tools and platforms (agents) needed for a user to interact with AI and build a product using vibe coding only.
Limitations
Vibe coding might be useful for someone like me to develop a prototype—something that shows the bare bones of my idea. But I’d be stuck the moment I had to review the code, or make changes, or figure out how to even prompt the AI to do what I want it to do for complex ideas.
Even Gemini cautions: “The approach can lead to security vulnerabilities, poor maintainability, and challenges with complex, scalable systems. It is best suited for internal tools, prototypes, and low-risk projects.”
Imagine this: for some reason, you want to write a book in another language. Your options are: learn the language yourself to the level you’d need to write the book; hire a translator to take your words and faithfully replicate not just the words but their meaning, style, tone, all of it; or use Google translate, and because you don’t actually know the language, you can’t check its accuracy, if the meaning has been faithfully maintained, or if anyone who speaks that language would actually want to read your book.
With vibe coding, the danger is taking what AI spits out at you and running with it, not checking for bugs, not putting it into context, not ensuring the integrity of the code. The name itself should perhaps give pause: we’re working on vibes, here, rather than rigor. If vibes go in, then vibes come out: something superficially sufficient, a good approximation of what I want, the same vibes as what I want.
I’ve heard people say vibe coding is like having an entry-level programmer working for you. You still have to check its work. It still takes time to train, de-bug, give context. Which, as it turns out, means you do have to understand the code, after all.
Resources
https://en.wikipedia.org/wiki/Vibe_coding
https://www.businessinsider.com/software-engineers-on-vibe-coding-ai-tools-2026-1