Vibe coding went from a casual tweet to a cultural phenomenon in under a year. It changed how millions think about programming. Here is what it actually means, where it came from, when it works, when it fails, and what replaced it.
Vibe coding is a style of AI-assisted programming where you describe what you want in natural language and let the AI write the code, accepting the output without fully understanding every line.
The term was coined by Andrej Karpathy -- co-founder of OpenAI, former head of AI at Tesla, and one of the most influential figures in machine learning -- in a tweet on February 2, 2025. He described his experience coding with AI tools where he would "just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works."
The key insight was that for certain types of projects, you could build functional software without understanding the implementation details. You focus on the "vibes" -- what you want the software to do and feel like -- and let the AI handle the how. Karpathy specifically noted he would "give in to the vibes, embrace exponentials, and forget that the code even exists." For a hands-on walkthrough, see our complete vibe coding guide.
From tweet to Wikipedia page to industry-defining paradigm -- the timeline of how vibe coding changed software development.
Andrej Karpathy posts about a new coding style he calls "vibe coding." He describes using AI to write code he does not fully understand, focusing on intent rather than implementation. The tweet resonates immediately with millions of developers experimenting with ChatGPT, Copilot, and Cursor.
Vibe coding becomes the most discussed topic in software development. Non-technical founders build MVPs over weekends. Students submit AI-generated assignments. The term gets its own Wikipedia page. Y Combinator sees a surge in solo-founder applications where the entire codebase was vibe coded. AI coding tools like Cursor and Windsurf see record adoption, with Cursor surpassing $1B in annual recurring revenue.
Vibe-coded apps start failing in production. Security researchers find widespread vulnerabilities in AI-generated code. Companies discover that codebases built on vibes are nearly impossible to maintain or extend. The term "vibe coding debt" emerges to describe the unique technical debt created by accepting code you do not understand. Experienced developers begin pushing back against the hype.
Karpathy himself declares vibe coding's casual era over, introducing "agentic engineering" as the professional evolution. The new paradigm keeps AI at the center but adds engineering discipline: specifications before prompts, tests before implementation, review before acceptance. This marks the maturation of AI-assisted development from novelty to professional practice.
Vibe coding is not universally good or bad. It is a tool with a specific effective range. Understanding that range is the difference between a successful prototype and a security disaster.
Not everyone is enthusiastic about vibe coding. The criticism comes from experienced engineers who have seen the real-world consequences.
The most fundamental criticism. When vibe-coded software breaks at 2 AM in production, the developer who accepted AI output without understanding it cannot diagnose the issue. They end up pasting the error back into the AI, starting a cycle of increasingly confused prompting that often makes things worse. Understanding your code is not optional when you are responsible for keeping it running.
Critics argue that relying on AI from the start prevents developers from building the mental models they need to evaluate AI output. Junior developers who vibe code exclusively never develop the intuition to recognize when AI-generated code is subtly wrong -- they lack the experience to know what "right" looks like. This creates a dependency loop where AI is needed not just for speed but for basic competence.
AI models trained on public code repositories have learned countless insecure patterns: hardcoded secrets, SQL injection vulnerabilities, missing input validation, insecure default configurations. When developers accept this output without review, they ship vulnerabilities directly to production. Security researchers have found that AI-generated code has a higher rate of certain vulnerability classes than human-written code, precisely because the training data includes so many examples of insecure patterns.
Vibe coding showed the world what is possible. Agentic engineering shows how to do it professionally. Learn the frameworks that let you build apps with AI at speed without the vibe coding debt.
Start Learning for FreeAndrej Karpathy, the co-founder of OpenAI and former head of Tesla Autopilot, coined the term in a tweet in February 2025. He described it as writing code by "vibes" -- describing what you want in natural language and letting the AI handle implementation, accepting code you do not fully understand. The term immediately went viral and became the defining phrase of AI-assisted development in 2025.
No. Vibe coding is a specific style of AI-assisted coding where you accept AI output without fully understanding or reviewing it. Professional AI-assisted coding (sometimes called "agentic engineering") involves directing AI with clear specifications, reviewing every diff, running tests, and maintaining architectural control. The difference is like the difference between dictating a letter and hiring an editor versus publishing whatever autocomplete suggests.
You can build a prototype or MVP with vibe coding, but production-quality software requires more discipline. Many founders have launched successful MVPs using vibe coding approaches, but those products invariably need professional engineering oversight to scale, secure, and maintain. The smart approach is to use vibe coding speed for prototyping and then apply engineering rigor before anything touches production.
The five main dangers are: (1) security vulnerabilities from AI-generated code trained on insecure patterns, (2) technical debt that accumulates invisibly when you accept code you do not understand, (3) architectural inconsistency from AI making different design decisions in different prompts, (4) inability to debug your own application because you did not reason through the code, and (5) false confidence from working demos that hide fundamental issues.
Vibe coding in its raw form will not replace traditional programming, but AI-assisted development will transform it permanently. The future is not "no code" or "all AI" -- it is developers who understand engineering fundamentals using AI to execute faster. The developers who thrive will be those who can direct AI effectively while maintaining the judgment to know when AI output is wrong, insecure, or poorly architected.