On February 2, 2025, Andrej Karpathy — co-founder of OpenAI and former Director of AI at Tesla — posted a tweet that would reshape how millions of people think about programming.
“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 ‘Accept All’ always, I don’t read the diffs anymore.”
The tweet received over 4.5 million views within days. Within weeks, the term was on the front page of Hacker News, covered by The New York Times and The Guardian, and adopted by Microsoft’s VS Code team. By November 2025, Collins Dictionary named vibe coding its Word of the Year — beating out every other candidate.
But what does it actually mean?
The Simple Definition
Vibecoding is building software by describing what you want in plain English (or any language) and letting an AI tool write the code for you. You don’t need to understand the code. You don’t need to read it. You describe the result, the AI produces it, and you check if it works.
Think of it like this: traditional coding is writing a recipe step by step. Vibecoding is telling a chef what you want to eat and letting them figure out the recipe.
What Karpathy Actually Meant
Here’s the part most articles skip: Karpathy explicitly scoped vibecoding to throwaway weekend projects. His exact words were “It’s not too bad for throwaway weekend projects, but still quite amusing.”
That caveat didn’t travel with the term as it went viral. People started applying it to everything — startups, production apps, entire businesses. Karpathy later called it “a shower of thoughts throwaway tweet that I just fired off.” He was as surprised as anyone.
How It Went Viral
The timeline moved fast:
- February 2, 2025 — Karpathy’s tweet. 4.5 million views almost immediately.
- February 24, 2025 — Microsoft’s VS Code team signed off a blog post with “Happy vibe coding!” — a major legitimization moment.
- March 2025 — Merriam-Webster listed it as trending slang. Google search interest for “vibe coding” jumped 6,700% in a matter of weeks.
- Spring 2025 — Y Combinator revealed that 25% of their Winter 2025 batch founders had codebases that were 95%+ AI-generated. Vibecoding wasn’t just a meme — it was an economic force reshaping how startups get built.
- November 6, 2025 — Collins Dictionary named it Word of the Year — the first programming-adjacent term to ever receive the distinction.
- February 2, 2026 — On the one-year anniversary, Karpathy reflected on how the term had evolved far beyond his original intent.
The Debate
Not everyone loves the term. Andrew Ng — another AI pioneer — argued at the LangChain Interrupt conference in May 2025 that “vibecoding” misrepresents what is actually “a deeply intellectual exercise.” He preferred the term “agentic engineering.” (He notably still went on to build a “Vibe Coding 101” course with Replit.)
Among professional developers, the conversation split:
- Non-technical users embrace vibecoding as a label for building without coding skills
- Experienced developers increasingly use “agentic engineering” for structured, professional AI-assisted development
- Critics point to real incidents — security breaches, production failures, a DeFi protocol that lost $1.8M — as evidence that vibecoding without understanding what you’re building is dangerous
The truth, as usual, is somewhere in the middle. The tools are genuinely powerful. The risk is treating them as magic instead of as tools that still need a human checking the output.
Where It Stands Today
Vibecoding is fully mainstream. It has a Wikipedia page, academic papers reference it, and it’s standard vocabulary in startup and developer communities worldwide. According to Stack Overflow, vibecoding has fundamentally changed who can build software.
The tools have gotten dramatically better since Karpathy’s tweet. Claude Code, GitHub Copilot, Cursor, and others now handle complex multi-file projects, run tests, debug errors, and deploy to production — all from natural language instructions.
What started as a joke about accepting all AI suggestions without reading them has become a legitimate way for non-technical people to build real software. The key shift: you don’t need to vibe blindly. You can vibe with guardrails — understanding what you’re building at a high level, testing the output, and using the AI as a partner rather than a black box.
That’s what this course teaches. Not blind vibing. Smart vibing.
This blog post is part of the free vibe.lab course on vibecoding. Start from Chapter 1 to learn the full workflow.