Remember how fifteen years ago infobiz advertised training in digital professions? Everyone drew idyllic pictures of you pampering your body in a hammock stretched between palm trees, with a laptop on your belly, on which you do something for TWO hours a day – and receive from four to ten salaries of a decent factory worker.
There is a feeling that some novice developers imagine working with code this way. Because now there are neural networks that you can simply tell what you want – and in response get a ready-made application code. Especially since now it is called by the fashionable word “vibe coding”.
In this article, we looked at what it is and listened to the opinions of developers about this phenomenon.
What’s happened?
In 2025, the era of vibe coding began. And all thanks to Andriy Karpaty, a scientist in the field of machine learning, former director of artificial intelligence at Tesla and co-founder of OpenAI. In early February 2025, he wrote that language models have become so cool that you can now create products without coding and barely touching the keyboard.
How Vibe Coding Became Possible
The community began discussing the term “vibe coding” in early 2025. However, the idea itself is not new.
“While some are surprised, others are timidly trying, and still others are fiercely criticizing, I sit and think – seriously? From the first day I saw modern neural networks in action, I started using them not to generate variations of pasta about dad’s soup, but for real work. And now, some guy from Tesla called it vibe coding.” – eaterman99, member of the Habr community
Autocomplete — automatic code completion — existed in various IDEs and other development tools long before the neural network boom. In 2022, GitHub Copilot started doing magic — it suggested entire lines, code fragments. But even then, the neural network did not understand the context of the project, got confused in logic, and was only good for template tasks. And most importantly, the guy from Tesla and OpenAI did not call it “vibe coding” yet.
Three technological leaps changed everything:
1. AI has learned to “understand” the entire project
Early tools only saw the open tab in the editor. Today, thanks to Cursor, Snap2txt, and Github Copilot, AI analyzes the entire project. Snap2txt compresses the context, creating a “digest” of thousands of lines of code. Cursor looks for connections between modules and builds a dependency graph. Bots remember the conversation and do not lose the thread of the task.
2. “Interlocutors” replaced “prompters”
Copilot was more of a tool. GPT-4o and Claude 4 can act as colleagues, partners in pair programming. You can have a dialogue with them: “Explain why this method does not work with asynchronous calls” → “How can I rewrite it for async/await?” They will help decompose a large complex entity – for example, they will offer an architecture option, describe dependencies and endpoints for a microservice. You can delegate debugging to them based on a simple description – that is, give them a piece of code, say what error it returns, and ask them to find the bug.
3. Local LLMs
For corporate projects with strict security requirements, GigaCode, Continue.dev, and Tabby have emerged — they operate within the infrastructure. Businesses pay for hardware, not tokens, and do not worry about the risk of leaks.
And now, language models in numerous tools like Cursor, Windsurf, and Firebase Studio can write entire modules — and you don’t need to know the code yourself to do this, you just need to talk to them in plain human language. Or is it?
Vibe Coding Through the Eyes of Practitioners
What is the Internet without a holy war… And when it comes to phenomena that really radically change the landscape of the industry, everyone speaks out. And around vibe coding, different voices have also been heard. Let’s see what practitioners say.
Some see this as a revolution – liberation from boilerplate code and a dramatic acceleration of development.
Others are sounding the alarm: will we turn into “AI managers”, losing our hard skills?
Others snort: “Bullshit! Just a waste of time” and call vibe coding a toy for beginners and switchers.
Optimists: “Vibe coding transforms programming from a technical skill into a creative collaboration where anyone can bring their idea to life.”
Optimists, often from startups, see AI not as a tool, but as a partner. They note a radical increase in speed: a prototype that used to take a week can now be assembled in a day. The main advantage is freedom from routine: generating CRUD, basic API, tests, configs. AI has also become a “personal teacher” for them, instantly explaining complex concepts using their own code as an example.
“We are entering a new era where the web is becoming a space for creating dynamic digital experiences. AI, no-code tools, and automation are removing technical barriers, and anyone can create rich, interactive digital worlds. There is a surge in creativity that is redefining the boundaries of what is possible online.”— Jamie Marsland, developer at Automattic
“Vibe coding is about working with AI in a smooth and intuitive way, where AI takes over routine tasks, and developers can focus on creative problem solving. But vibe coding is not just for programmers. It gives you the opportunity to create, even if you don’t know how to code. It’s about removing barriers, making technology accessible and creating something meaningful. Vibe coding is proof that AI does not replace creativity, but helps to reveal it. It turns programming from a technical skill into a creative collaboration. Anyone can bring their idea to life.”— Bill Salak, CTO of edtech company Brainly
“Vibe coding has significantly sped up our product development process. Ideas that have been sitting in the backlog for years have come to life in just a week thanks to AI. I can create a working prototype of a new feature in a day. If I provide the AI with requirements for a new product, I know that in about 20 minutes I will have a functional prototype that can be sent for testing. AI allows us to quickly give users the features they want. Vibe coding is my go-to method for creating MVPs quickly.”— Zach Katz, president of nocode platform GravityKit
The voices of skeptics and caution: “Now two engineers can create as much bad code as it used to take fifty.”
There are more voices of experienced programmers, architects, developers of large low-latency systems, cybersecurity specialists. They do not deny the phenomenon as such, but pay attention to the issues of security and supportability. All this is still the area of interest of people, and not a machine that is only looking for the shortest path to satisfying the user’s immediate request.
“Vibe coding changes HOW we work, not WHY. The goal is not to produce more software, but to create a better user experience. AI makes code generation easier and faster, but it does not guarantee better results. Without constant feedback from users, vibe coding can only speed up the creation of software that no one uses. This is not a new problem, but AI increases the risks and the stakes.”— Todd Olson, co-founder and CEO of Pendo
“The key to vibe coding is having a human operator who is actually smarter than the computer. You can’t just let robots run wild. Sure, generative AI speeds up development, but it also increases risk. Two engineers can now write the same amount of unsafe and hard-to-maintain code as fifty used to.”
“One of the problems with using language models by developers is that they make things up that sound plausible. At the same time, they can only deeply research, work out security aspects, apply best practices if they are specifically asked to do so. An advanced user with no development experience can use AI to create things that look and work the way they want. But if they don’t understand how it all works, then when they encounter bugs, they may find themselves completely powerless.”— Chris Reynolds, Pantheon Platform Software Engineer
In the context of this approach to programming, experts often talk about security issues. It is not a priority for the neural network that creates the code. But what is even more alarming is that it is most likely not thought about by a novice developer who has decided to “give in to the vibe” and “enter IT”.
“Vibe coding makes development more accessible, but it also introduces security risks that even experienced developers have trouble managing. When you’re collecting user data or handling sensitive information, you need more than just a good vibe. Security can’t be an afterthought or an annoying hindrance to building cool stuff. You need to understand the risks and set up security checks to avoid turning users into accidental victims.”— Willem Delbare, Co-Founder and CTO of Aikido Security
“Vibe coding shows how AI tools can democratize programming. It allows beginners to create apps without traditional programming experience, but it also highlights that without basic development knowledge, security and productivity suffer.”— Jamie Madden, founder of PetFun
“I think the term vibe coding is used by people who think that AI-generated code is safe and secure, and want to fill a niche in a market where they probably don’t belong. Every vibe coding project I’ve seen has had serious security holes, didn’t solve the problem, or just imitated existing and better solutions.”— Rhys Wynn, Freelance WordPress Developer
Moderate skeptics acknowledge the benefits of AI for low-level tasks (documentation, simple scripts), but are against its use for complex logic or working with old systems. They see a risk of accumulation of hidden technical debt in vibe coding and insist that the work of the neural network should be supervised by an experienced developer with fundamental knowledge – and not even one.
“Vibe coding is revolutionizing software development, increasing developer efficiency and making the field more accessible to people with limited technical knowledge. But a teenager who gets behind the wheel for the first time needs to be limited by traffic rules, safety systems, and generally supervised. The same is true for vibe coding – you need fundamental knowledge and rules so that everything does not turn into chaos.
The gap between vibe coding and industrial programming will narrow. But for now, the former is more suited to creating “web toys” — personal tools with narrow specialization and minimal security requirements, without the risks and responsibilities associated with deployment in a production environment.”
How to Stay Relevant as a Developer in the Age of Vibe Coding
The debate about vibe coding is reminiscent of the early 2000s debate about whether “business needs the Internet.” Or the 2010s debate about the need to go mobile. Today it is clear: AI will not replace the developer, but it will change their tools forever. According to GitHub, in 2025, 92% of developers will use AI assistants, and 70% of companies will have implemented them in their workflow. The question is not whether to use AI, but how to do it without compromising quality.
Here are some tips and thoughts that can help you navigate the AI world.
1. AI is an amplifier, not a replacement for intelligence.
A good developer with AI will do 3 times more. A bad one will generate 3 times more problems. Responsibility for the architecture, security and final quality of the code remains with the person. AI only processes the request – like a compiler processes code.
2. Improve your soft skills, including those related to neural networks.
You need to be able to communicate not only with colleagues and clients, but also with AI. Telling AI to “write a parser” is a failure. Saying: “Create a parser of nginx logs in Python that extracts IP, status code, and User-Agent. Example of a log: <sample>. Handle format errors. Write tests with coverage >90%” is a success.
3. Develop systems thinking. A junior writes code. A senior designs the system. AI hasn’t changed this — it’s just made the gap between them even more obvious. As long as a developer only writes code fragments for a task, he remains a junior. But if he himself writes the code with difficulty and gives it to a neural network — things are really bad. You can’t just skip a stage. And the value of a senior is that he sets precise tasks (to a live developer or AI), sees the system as a whole, and takes responsibility for the result.
4. Delegate what AI can do — but know its limits. Linus Torvalds, the creator of Linux, said in a recent interview with ZDNet: “I hate AI hype. But if a tool makes things faster, it’s stupid not to use it. The main thing is to remember WHO is in charge.”
Remember that the term in the title of this article was coined by a professional engineer. And it’s unlikely that he meant that now anyone can walk in from the cold, enter a two-sentence prompt, and make a couple million dollars on the app.
