Vibe Coding: What Is It, Is It Cheating, and Is It Here to Stay?

08/20/2026|,

Written by

Someone may have recently told you they built an app. A nephew, a coworker, a guy on a podcast. Yet you know this person has never written a program in their life before. They also say in the same breath that they can’t really write code. So how? Must you know programming languages to write an app? The answer today is no. Well, not quite no, but we will get into that. The word for what that person did is vibe coding.

Last November, Collins Dictionary made “vibe coding” its Word of the Year.

In December 2019, campaigning in Derry, New Hampshire, Joe Biden suggested that displaced coal miners could move into tech work by learning to code. As reported at the time, he said anybody who can go down thousands of feet in a mine can “sure as hell learn to program.” The fuller context was a retraining argument. He pointed at a nineteen-week community college program in Detroit that had trained women for IT jobs. Federal money was meant to do the same in coal country.

Politics aside, and whatever you make of it, it did not go over well. The objections came from every direction: from his own left, from a primary rival, from the coal industry, and from programmers. Timing had a lot to do with it. “Learn to code” had already curdled into an insult eleven months earlier. Still, in a way, he was unknowingly describing the future.

The odds are you almost certainly won’t do it, and this is not an article on how to program. The focus here is that coding, or programming, is something someone can do without having a career in programming or even Information Technology.

What code actually is

Let’s take a step back for a moment. What is code actually?

A computer does nothing on its own. Code is a set of written instructions telling it exactly what to do, in a language it can follow without guessing. Literally everything a computer does is based on some written code, or what we call a program. You have heard the term, code is just the new term for a program.

A program is a series of commands. They let a computer or device, like your smartwatch or smartphone, do its job. When you click on something, the computer checks its code to determine what to do next.

Now, understand this is a very basic and general explanation; there is a lot more to it, but the foundation is important here.

Where vibe coding came from

In February 2025, an AI researcher named Andrej Karpathy described how he’d been working. He said what he wanted in plain English, let the AI write the code, and barely read it. He called it vibe coding. You “fully give in to the vibes,” he wrote, “and forget that the code even exists.” In short, he types or even talks to an AI such as ChatGPT or Copilot. Then he tells it what he wants his program to do.

The name traveled further than the idea. You describe, the AI builds, you look at what came back, you describe the next thing. Think of it as if you are a manager for an employee. You sit across the desk and tell them you want a program that looks up the weather. You describe how you want it to look and how it should work for the person using it. In short, this is vibe coding.

Microsoft’s chief executive said in April 2025 that software or an AI wrote 20 to 30 percent of the code.

As a side note, on February 4 of this year, Karpathy walked back his own term. He’d meant it for “fun throwaway projects, demos, and explorations,” he said, and offered a different one, agentic engineering, for serious work where a human stays accountable. Since then he has moved on to a research role at Anthropic, and though he tried to change the term, it has stuck.

The cheating question. Is it cheating if you don’t spend years learning to code first?

No.

Cheating implies a test with rules, and there are no rules. No exam stands between a person and publishing an app, and there never has been. If you are writing something to do something of just yourself, the only gatekeeper or grader is you. Now, if you want to publish an app for the iPhone or a Windows desktop, there are more gatekeepers. There are rules and guidelines you have to follow. In those cases, the AI can work with them.

What’s happening is older than it looks. Coders or programmers have always learned by doing. Before AI, that meant learning the language first and building second. Vibe coding is the same approach through a different door: it lets someone who has never coded start much further along. The big difference is that they learn how to work with AI and manage it, rather than how to write the program.

Where the work actually goes

This management approach sounds like it removes the work. It doesn’t. It moves it. These AI tools have vast knowledge and no understanding, and they will do things unless you question them and check them carefully. Yes, you can start coding with no prior knowledge. But you have to be willing to admit what you don’t understand. First to yourself, then to the AI. This is not just something a new vibe coder will make, but seasoned ones as well, assuming there are seasoned ones. Remember, vibe coding is only about a year old.

I made the full case for that in the ai didn’t go rogue, it was just that focused. An AI will do things in the most efficient and fast way it can, this does not always mean the best.

Quick summary

Vibe coding means building software by describing it to an AI instead of writing the code yourself. It isn’t cheating, it works, and it’s already behind apps you use. The catch: AI-written code is syntactically correct more than 95 percent of the time. It passes a security review about 55 percent of the time. No app store currently has a rule about any of it. It is also here to stay.

Where vibe coding goes right

The best-documented case involves a school district.

Peninsula School District in Washington state used AI coding tools to replace software it had been buying. A workflow tool, an e-signature tool, a scholarship finder, a budget dashboard. It expects to cancel enough contracts to save between $220,000 and $250,000 a year. Two education trade outlets reported it with the district’s technology chief on the record, though the money is projected rather than promised.

What the district did that matters isn’t the code. It’s the fence around it. The tools stay away from anything that identifies a student. Sensitive data stays on district servers. Nothing gets deployed before somebody vets it. Their approach was right: they did not assume the AI’s output was correct; they verified it at each step of the process.

That fence shows up wherever this goes well, and it tightens as the stakes rise. The Linux kernel, the software running most of the internet, wrote its policy in a line. An AI may write the code. Only a human can sign for it, because signing is a legal statement and a machine can’t make one. NASA goes further. Its software engineering handbook says not to use generative AI on safety-critical code at all.

Who signs for it

Notice the split. The people who literally run the internet say use it, but sign for it. NASA, which puts people in space, says it does not recommend it for safety-critical code, but stops short of forbidding it outright.

Would I vibe code the flight computer on a launch vehicle? Yes. Would I trust it without testing and reviewing it? No. Would I also use an AI to check that work and look for what a human missed? Absolutely.

It’s a circle. The AI can write code faster than a human, and that isn’t the argument. It has no true understanding of the program’s purpose. Yes, it knows it’s supposed to run the thrusters on the Mars lander, but does it understand what that truly means? No.

Yet with proper guidance, it can. Someone reviewing it, questioning it, testing it, asking the AI to explain itself. That is what produces a program that could land on Mars.

Where vibe coding goes wrong

The trouble isn’t that AI writes bad code. It writes code that looks right and is right based on what it was told to write. The problem is the human. The human did not cover or check everything they should.

Stack Overflow surveys tens of thousands of working programmers every year. In the most recent one, the top complaint, from 66 percent of them, was “AI solutions that are almost right, but not quite.”

The security firm Veracode has run the same test for two years. Eighty coding tasks, four languages, more than 150 AI models, checking whether the output carries known security flaws. As of March 2026, the code is syntactically correct over 95 percent of the time. It passes a security review about 55 percent of the time. That second number hasn’t moved in two years. Read them together. The code runs and does what you asked, nearly every time. Whether it’s safe is about as likely as a coin flip, and looking at working software tells you nothing about which side you got. What this means is that you, as the manager, have to verify and question whether it’s safe. You can make secure programs that handle complex tasks and very personal information. You just have to make sure the AI is doing it right.

“Best Practices” is a term that, in simple terms, means doing it the right way, which has been shown to be reliable and effective. One thing a vibe coder may ask an AI is, “Is this app following Apple’s best practices?” The AI may need more details. But it will check its own work against what Apple says programs should and shouldn’t do.

When it lands on somebody else

In February 2026, that gap landed on people. A social network called Moltbook left its database open to anyone who went looking. Wiz, the security firm that found it, counted roughly 1.5 million keys to users’ paid AI accounts, 35,000 email addresses, and 4,060 private messages left exposed. Its founder had posted, before any of this, that he hadn’t written “a single line of code” for it. Where data like that travels next is the subject of what happens to your data when a company gets hacked.

The site patched the hole and is still running. No fine. No lawsuit. Nothing happened to anyone except the people whose messages were in there.

I am learning to vibe code; it is like the telephone, TV, and, most of all, the Internet. It will change the future. One feature coming to people soon, I expect this year, will be those settings people tell you to change on your phone. Go to Settings, then To this and that, and click this. What if you could say, “Hey Siri, I want to change my Voice Mail Greeting Message,” and instead of telling you how to get there, it gives you your options. You get that now when you call up a company and get an AI asking you questions before sending your call to a human. It may not be writing code, but the concept is the same: you are guiding the AI in what you want it to do, or it is guiding you to your goal.

Nobody inspects the kitchen

No major app store has a rule about apps built with AI. Not Apple, not Google Play, not Mozilla, not Microsoft. I read all four sets of current policies looking for one. No one asks who or what wrote the app. They have a system in place that checks your code for security and privacy and ensures that it does what you say it does. If it passes, they generally let it through, regardless of whether it’s actually good.

So, should I worry about vibe coding?

So, should you worry if a program is vibe coded? No, it is new, and like the Internet many years ago, or the TV before that, it needs time to find its place.

Is vibe coding going well or evil? No, it’s neither; it’s the person guiding the AI behind it.

Could I learn to vibe code? Yes, it’s not as easy as YouTubers make it sound, and you do not need to pay someone to get started unless you want to speed up the process. Just remember that the person is as new to vibe coding as you are, with maybe a year or less of experience.

So when the person in your life tells you they built an app, believe them. Then ask what happens to the information people put into it. If they’ve thought about that, they’ve learned the part that matters. If it catches them flat, you have your answer, and it cost you one sentence.

Verified Resources & Documentation

Keep Reading

Recent Posts

Categories