Once upon a time, God said, “Let there be light”—and there was light. Now you can say, “Let there be this app,” and the app will appear right before your eyes, all thanks to the magic of AI.
This is what vibe coding promises.
“I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works,” Andrej Karpathy—a co-founder of OpenAI and former AI director at Tesla— tweeted last month, describing his new programming approach on social media.
Instead of meticulously writing code, he explained that he simply talks to an AI assistant and lets it do the heavy lifting. He called this “vibe coding”—and all the AI guys loved it.
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. It’s possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper…
— Andrej Karpathy (@karpathy) February 2, 2025
The term immediately found product-market fit, and in just a few weeks, Merriam-Webster Dictionary had added “vibe coding” to its database, defining it as “writing computer code in a somewhat careless fashion, with AI assistance.” By March, Y Combinator CEO Garry Tan reported that in the accelerator’s Winter 2025 startup batch, “for 25% of the batch, 95% of lines of code are LLM-generated.”
Some think this is the future of technology, when anyone can bring into reality any idea using natural language, and giving machines the responsibility to code. Others see the trend as the tech equivalent of a jam session, in which people start with an idea, and let the flow—and the interaction with their chatbot—guide them through a final result.
Of course, critics see this trend as a recipe for disaster.
“Vibe coding is like buying a kit to build a race car, paying your drunk uncle who ‘knows a thing or two about racing’ to build the kit for you, then telling all your friends that you built it,” said one Reddit user.
From typing code to talking to the code
At its core, vibe coding involves telling an AI what you want to build in plain language, then letting it generate the actual code. Rather than wrestling with syntax or poring over documentation, developers can simply say, “Create a login form with email validation and password requirements,” and watch as the AI handles the implementation details.
The vibe coding approach flips traditional software development on its head. Instead of writing code line by line, vibe coders focus on high-level instructions and iterative refinement through natural conversation with AI assistants.
“It’s a new way of coding where you don’t write a single line of code,” noted tech educator Matthew Berman. “You do not read the code, you simply try to use it, see if it works—and if it doesn’t, you explain what’s wrong… you do not look at the diffs, and you basically just vibe.”
This is wild.
Vibe coding with AI just completely changed the game.
People can’t stop creating games with Grok, Windsurf, Cursor & Claude 3.7 Sonnet.
10 wild examples:
1. Fortnite meets Minecraft on Three.js 🤯pic.twitter.com/towhyg0oYv
— Min Choi (@minchoi) March 21, 2025
Karpathy calls this the “Accept All” mentality. “I ‘Accept All’ always, I don’t read the diffs anymore,” he said, referring to how he accepts AI-suggested code changes without scrutinizing them.
This approach—where the AI is trusted to write working code without the developer fully understanding every line—is what distinguishes true vibe coding from simply using AI as a sophisticated assistant.
In AI-assisted coding, users rely on AI to help them code, but they debug errors and understand what the AI is doing. The AI simply is a very fast typist that gets the job done. Vibe coders on the other hand don’t debug, don’t code, and don’t know—nor care—about what happens under the hood. All they do is talk, review, and talk again.
In a pure vibe coding session, the prompter doesn’t check on the code. They just point out the errors and their ideas for improvement after running what the machine created. It’s the machine, not the human, the one who is supposed to fix everything and improve the code.
Turning coding into a creative flow state
Beyond the technical aspects, vibe coding has spawned its own distinct culture and aesthetic. The term “vibe” isn’t accidental—it captures the emphasis on creativity, flow, and even playfulness that practitioners bring to coding sessions.
On TikTok and YouTube, videos of vibe coding often feature developers in moody lighting with RGB keyboards, ambient music playing softly in the background as they chat with their AI assistant. The scene looks like anything but traditional software engineering.
This fusion of coding with a chill, creative atmosphere has made programming appear more approachable and emotionally engaging. Riley Brown, a social media influencer who calls himself a “Sr. Vibe Coder” on LinkedIn and is now fully dedicated to spreading awareness on this trend, has amassed over a million followers across platforms by posting videos where he builds functioning apps through voice commands alone. Other influencers like Matthew Berman, Nuno Maduro, and The Primeagen have also shared their sessions on YouTube and Twitch.
The aesthetic emphasizes that coding can be fun and expressive rather than dry and technical. It’s a stark departure from the buttoned-up, precision-focused culture traditionally associated with software development.
Several high-profile tech figures have championed vibe coding, propelling it from niche concept to mainstream discussion in a matter of months.
Andrej Karpathy’s influence cannot be overstated. His initial posts describing his vibe coding workflow received thousands of likes and sparked widespread debate. Given Karpathy’s reputation in the AI community, his endorsement lent significant credibility to the approach.
Andrew Chen, general partner at venture capital firm Andreessen Horowitz, has been another evangelist of the trend. Chen described vibe coding as “brilliant and enormously frustrating.”
“Most code will be written (generated?) by the time-rich,” he said. “Thus, most code will be written by kids/students rather than software engineers. This is the same trend as video, photos, and other social media.”
random thoughts/predictions on where vibe coding might go:
– most code will be written (generated?) by the time rich. Thus, most code will be written by kids/students rather than software engineers. This is the same trend as video, photos, and other social media
– we are in the…
— andrew chen (@andrewchen) March 9, 2025
Amjad Masad, CEO of Replit, has reported that “75% of Replit customers never write a single line of code” thanks to AI features—a statistic that underscores how vibe coding principles are already changing who can create software.
Detractors speak
But not everyone is convinced. Many developers have raised serious concerns about vibe coding’s potential drawbacks.
Code quality is the main criticism. Vibe coding could be a great tool for inexperienced people, but experts argue it is still not good enough to produce quality results. Jonathan Blow, developer of acclaimed indie game Braid, weighed in on the trend in response to vibe coder Levelsio in a popular debate on X, which was triggered after the latter shared a flight simulator built entirely during a vibe coding session.
“I’ll put it this way: If you have never tried to make a game, I am sure it is fun to have a game-making experience. For those who have made games, even just simple ones, getting stuff on the screen is not impressive, it’s easy; it’s making the game good that is hard,” Jonathan Blow, developer of acclaimed indie game Braid, told vibe coder Pietier Levels (aka Levelsio) in a viral debate on X triggered after the latter shared a flight simulator built entirely with AI.
I’ll put it this way: If you have never tried to make a game, I am sure it is fun to have a game-making experience.
For those who have made games, even just simple ones, getting stuff on the screen is not impressive, it’s easy; it’s making the game *good* that is hard.
— Jonathan Blow (@Jonathan_Blow) February 25, 2025
Security vulnerabilities represent another significant risk. AI assistants don’t inherently understand security best practices that are specific to a particular application. Blindly accepting AI-generated code could introduce security holes that might remain undetected until exploited.
“AI makes shit up. Code written by AI is almost always flat-out wrong,” said Reddit user Dw444 in a recent post. “My employer pays for AI assistants we can use for work, and even the most advanced models are prone to start writing blatantly incorrect code at the drop of a dime. You really don’t want to use AI code in [production].”
There’s also concern about skill atrophy. If developers increasingly rely on AI for even moderate coding tasks, then their foundational skills might deteriorate over time.”Ease of use is a double-edged sword,” AI researcher Harry Law told Business Insider. “Beginners can make fast progress, but it might prevent them from learning about system architecture or performance.”
Debugging can also become particularly challenging when the original code wasn’t human-written. As Andrew Chen noted, while getting the first 75% of a project done with AI might be trivial, refining and fixing it afterward can become extremely difficult.
Overall, the best approach thus far appears to be AI-assisted coding.
“Suffice to say (vibe coding) is not mainstream and the first few attempts at generating useful code—in a few minutes—were completely wrong,” AI developer Julian Harris wrote while describing his experience with vibe coding.
But he concluded, “I found careful prompting with solid software engineering disciplines worked very well. With my background, I expect I’d have spent twice as long hand-writing it all, and honestly probably would’ve given up.”
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
Source: https://decrypt.co/311144/vibe-coding-using-ai-create-apps-games
