The AI Spotlight
AI Coding

Vibe Coding Went From One Tweet to Word of the Year. Here’s How to Do It Without Getting Burned

Andrej Karpathy coined it in February 2025. By November, Collins Dictionary crowned it. Vibe coding lets anyone build software by describing it, and it comes with very specific traps.

A
The AI Spotlight DeskNov 6, 2026 · 4 min read
Advertisement
AdSense slot · articleTopplaceholder until ca-pub ID is set

On February 2, 2025, Andrej Karpathy, a founding member of OpenAI and former director of AI at Tesla, posted a short note on X. “There’s a new kind of coding I call ‘vibe coding’,” he wrote, “where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.”

He described talking to an AI coding tool, accepting its changes without reading the diffs, pasting error messages back in until they went away, and ending up with a working weekend project he did not fully understand. It was half a confession and half a joke.

Nine months later, on November 6, 2025, Collins Dictionary named “vibe coding” its Word of the Year, citing a huge rise in usage since February. As the BBC reported, it beat a shortlist that also reflected a year obsessed with technology.

How did a throwaway phrase become a whole way of building software? And should you do it?

Advertisement
AdSense slot · articleMidplaceholder until ca-pub ID is set

What vibe coding actually means

In the strict sense Karpathy meant it, vibe coding is building software by describing what you want in plain language and letting AI write the code, without reviewing or understanding that code yourself. You judge the result by whether it works when you run it, not by reading it.

That is different from AI-assisted programming in general. Developer Simon Willison made the distinction sharply in a March 2025 post: if an AI writes code and you review, test and understand it, that is not vibe coding, it is just using an AI as a typing assistant. The “vibe” part is specifically about not looking.

In everyday use, though, the term has stretched to cover anyone building apps through prompts, from non-programmers making their first tool to startups shipping products where most code is AI-generated.

Why it took off so fast

Three things lined up at once.

The tools got good enough. Coding assistants and agents became capable of generating whole working features, not just autocompleting lines. Browser-based app builders let people describe an app and get a running prototype in minutes.

The audience got huge. Most people with an idea for an app have never written code. Vibe coding offered them a way in, and they took it.

Startups leaned in. In March 2025, Y Combinator managing partner Jared Friedman said that for about a quarter of its Winter 2025 batch, 95% of the code was written by AI, according to TechCrunch. Those founders were technical, and many chose AI generation anyway because it was faster.

Where vibe coding is genuinely great

  • Personal tools. A script to rename your photos, a tracker for your workouts, a dashboard just for you. If it breaks, nobody else gets hurt.
  • Prototypes and demos. Show an investor, a client or your team what you mean in an afternoon instead of a month.
  • Learning by building. Seeing working code for your own idea is a strong way to start understanding how software fits together.
  • Throwaway experiments. Testing whether an idea is worth building properly.

Where it will burn you

The trouble starts when vibe-coded software leaves the “just for me” zone.

Security. Code you have not read can contain exposed API keys, missing authentication, open databases or injection vulnerabilities. AI models can and do write insecure code, and an app that “works” can still be wide open.

Money. Apps that call paid AI or cloud APIs can rack up real bills if a loop runs away or a key leaks publicly.

Maintenance. When something breaks months later, nobody understands the code, including you. Each AI fix can introduce a new problem somewhere else.

User data. The moment real people sign up, you take on responsibility for their emails, passwords and payments, along with privacy laws that do not care how the code was written.

A safe vibe coding playbook

You do not have to choose between vibes and discipline. Use both, and match the rigor to the stakes.

  1. Vibe the prototype, engineer the product. Go fast while it is only you. Slow down before strangers use it.
  2. Never paste secrets into code. Keep API keys in environment variables or a secrets manager, and set spending limits on every paid API.
  3. Ask the AI to review its own work. Prompts like “list every security issue in this app” or “where could a user see another user’s data?” catch a surprising amount.
  4. Use boring, proven building blocks. Hosted authentication, managed databases and established payment providers handle the hardest security problems for you.
  5. Save your progress. Use version control so a bad AI change is one click to undo, not a lost weekend.
  6. Get a human review before launch. One hour with an experienced developer looking at authentication, data access and billing is cheap insurance.

The bottom line

Vibe coding is real, it is useful, and it has permanently lowered the barrier to making software. Karpathy’s original framing is still the best guide: it is fantastic for “throwaway weekend projects.” The skill that matters now is knowing when a project has stopped being one, and switching from vibes to verification before your users find the bugs for you.

Sources

Up next · 4 min readThe $5.6 Million AI Model That Wiped $589 Billion Off Nvidia in a Single DayIn January 2025 a Chinese lab called DeepSeek released a reasoning model that rivaled the best in the world, and Wall Street panicked. Here is what the famous price tag really covered, and what the market got wrong.Keep reading →

More stories you'll want to read