Vibe coding

Main article: How artificial intelligence helps develop software

2025

Almost 80% of Russian developers used vibe coding in their work

ICT.Moscow October 10, 2025 published the results of a survey on the use of the Vibe Coding method in professional development in 2025. Data show that vibe coding is quickly gaining a place in the working practice of representatives of Russian IT the community: most have already tested this approach, many have created new products with it. At the same time, a significant group of respondents who either do not use vibe coding for organizational reasons or reject it due to unsuccessful experience or distrust of artificial intelligence is noticeable AI ().

The concept of vibe coding, introduced in February 2025, describes an approach in which a developer does not write code line by line, but generates it through instructions to an AI agent.

The results of a survey recently conducted by ICT.Moscow among more than 400 specialists showed that the majority (76%) of respondents are already familiar with the direction of vibe coding and tested it in work tasks. Of this number, more than 83% (64% of all survey participants) were satisfied with the results. Among those who tried the approach, almost 21% (16% of all respondents) were able to create a new product using this programming technique.

The total share of developers who do not use this method as of October 2025 is 36%. These include, among other things, 21% of study participants who deliberately refuse to use vibe coding - due to unsuccessful experience with AI agents (12%) or fundamental distrust of the results created by AI (9%). This group also includes those who have not yet tried this approach (11%), and those who are not yet allowed to work conditions (4%).

The survey was conducted in August - September 2025 among users of the AI Knowledge Base. It was attended by 475 respondents.

The material is part of a series about tools used by Russian AI developers as part of a special project about AI.

Vibe coding destroys the mechanism of forming programming skills

Waib coding creates the illusion of the availability of programming, but in fact destroys the mechanism for forming these skills, which completely eliminates the development of structural and systemic thinking with the degradation of understanding the architecture, hierarchy and algorithms of the program, debugging methods and verification of the final result.

The result is a generation of pseudo-developers capable of generating code, but unable to understand, debug, optimize or support it.

Independent programming is critical not for knowledge of the syntax of the programming language, but for the formation of fundamental cognitive skills - algorithmic, structural, logical, abstract and systemic thinking with task planning and decomposition and hierarchical modeling.

The days of super-people are long gone. The assembler programmer did not just write code, he directly controlled the processor. He had to thoroughly know the command system of a particular processor, the architecture of its registers, how to address memory and the peculiarities of the operation of peripheral devices. Each line of assembly code corresponded to one machine instruction - it is very difficult, long and ineffective, but these skills were made from extra-human programmers.

An error of one byte in addressing could overwrite the system code or data and completely crash the system.

In conditions of limited computing resources and memory, it was necessary to "reinvent" mathematics, create special algorithms to optimize the code, which extremely developed cognitive skills.

The digital world (at the level of the foundation) that we know - created by the "gods" from the 1970s and 90s.

These programmers were the creators of the foundation: they wrote operating systems, compilers, device drivers, BIOS, the first video games not through procedural generation, as they are now, but literally "pixel by pixel" - all that the modern digital world still stands on.

Their work required a unique combination of formal logic, engineering accuracy, and deep knowledge of electronics.

The creation of an understandable and logical architecture required a programmer of enormous self-discipline and thoughtful design on paper before writing code.

A modern programmer in 2025 works at the top of a huge technological pyramid. There are dozens of layers between its Python or JavaScript code and hardware: a high-level language, its runtime, operating system, drivers, etc.

Now the developer does not need to know how the network socket is physically arranged in order to send an HTTP request. He uses a finished library. It does not need to manually manage memory - the compiler does it for it. Now the complexity has moved from the level of machine commands to the level of interaction of systems.

On the one hand, the intelligence of modern programmers is not aimed at combating the limitations of the machine, but at managing the complexity of distributed systems and quickly delivering business value. The loss of manual memory management and bitwise optimization skills is not degradation, but evolutionary adaptation to a new technological reality.

On the other hand, vibe coding significantly transforms even modern OOP, WEB and cloud programming.

Software developers write code in the usual language. Is there a future for such programming in Russia - TA opinions

Modern AI tools turn the description of a natural language problem into working code. It is enough to write "write a script for parsing the site" - neuronet and offer a ready-made solution, for example, on. Python Russia TAdviser He spoke with experts and market participants about the prospects of the so-called vibe coding in June 2025.

According to Maxim Lapshinov, head of the Ramax Group's mobile application development group, at the current stage, vibe coding is perceived rather as another kind of hobby in which even those who have never encountered code writing directly can take the first successful steps. And as a hobby, this approach is more suitable for small projects with fairly simple internal logic. But even in this case, there is a possibility that AI will not be able to get out of the vicious circle of correcting its own errors, and the final code will not pass a critical review from an experienced developer, the interlocutor of TAdviser believes.

𝓲

Фото: Freepik

Vibe coding elements are already used in Russia. So, they are implemented in the tools Yandex Code, Sber AI, etc. As noted by the technical director of MD Audit (Softline Group of Companies) Yuri Tyurin, domestic teams are trying to use this approach for routine tasks: generating template code, SQL queries, unit tests and documentation. This is especially useful for speeding up prototyping, DevOps and QA automation.

Vibe coding is promising as a way to increase developer productivity, primarily in the early stages and when working with typical tasks. However, it will not completely replace classical programming - especially where architectural solutions, optimization, security and infrastructure integration are important. Rather, vibe coding will become the programmer's everyday "co-pilot": lower the entry threshold and speed up work, but not cancel engineering expertise, Tyurin said.

Maxim Zakharenko, CEO of Cloud Computing, says that by mid-2025, vibe coding in Russia is already being tried - most often in prototyping, when writing tests, documentation or standard code. This is especially convenient in the start of projects when speed is important. But vibe coding cannot be called a new way of programming - rather, it is a useful assistant that speeds up work with routine, the expert believes.

Fedor Tsvetkov, head of marketing at FedAG (develops AI systems for automating routine tasks of employees), agrees with this opinion. Vibe coding remains dependent on professionals, he said: AI, when generating code, often makes mistakes that require refinement. Programming involves logic, architecture and optimization, and this AI will not cope without human input. This is a tool, not a replacement for the developer, Tsvetkov emphasized.

Artyom Potemin, CTO of Arcsinus, doubts that vibe coding will replace traditional programming by 2027-2028 - unless, of course, there is some breakthrough in the development of programming models. This is an up-to-date tool that can save a programmer a lot of time, but which does not eliminate the need to have professional skills. Vibe coding generates a huge amount of technical debt, especially if you do not follow it, he explained.

Ramil Zaineev, co-founder and CEO of AI solutions developer ZeBrains, sees great prospects in vibe coding. If initially unprepared use led to unmanaged projects due to loss of context and lack of accounting for architecture, then with configured servers, workflow and promptas, the quality of the code increases significantly, he argues. Vibe coding allows you to reduce development time - a serious startup with integrations and asynchronous requests can be created in 12-15 hours. For seniors, this is a tool for improving efficiency, for business - a way to increase productivity, Zaineev added.

However, before vibe coding becomes ubiquitous in Russia, a number of barriers must be overcome. One of them is associated with the developers themselves, points out Vitaly Secretenko, technical director of Korus Consulting Group of Companies: it can be difficult to move from the usual paradigm "write code" to the paradigm "set goals and test hypotheses." In this sense, within the framework of vibe coding, maturity, systemic thinking and openness to new approaches are needed, the TAdviser speaker believes.

In his opinion, for the large-scale distribution of vibe coding in Russia, the development of local models, data protection and compliance with corporate requirements is important. Some companies are already following this path, introducing isolated environments with LLM and training teams to work in a new paradigm, Secretenko said.

The head of the Yandex Go for Business development, Ilya Masandilov, in a conversation with TAdviser, stressed that vibe coding is extremely promising in Russia, but with big reservations. In 2025, the role of the developer is significantly transformed: often, instead of writing code (which is interesting and exciting), with more and more AI implementation, the developer has to check someone else's code, fix errors and look for problems.

At the same time, I am sure that with the growth of a possible context window for large language models, their efficiency in writing code will multiply, "Masandilov said.

Wyb-coding: IT cheating or a real breakthrough in programming

Vibe coding is a new approach to creating software in which developers formulate their tasks in Russian, English or other natural language, and artificial intelligence generates code. In this format, the specialist's focus shifts to the role of a reviewer: he does not write an algorithm, but tests the results by trial and error. The approach speeds up prototyping - creating the first working versions of programs to quickly test an idea in practice - and allows you to create your own applications even for beginners, which is already radically transforming the field of startups. On May 6, 2025, NITU MISIS told what is the essence of vibe coding, how it affects the industry, what are the possible risks associated with information security.

𝓲

Фото: Freepik

As reported, vibe coding is when a developer works at the level of intent, not implementation, often taking the generated AI code without detailed analysis. At the same time, the novelty of the technology is manifested in two aspects. First, it thoroughly changes the development workflow: natural language becomes a full-fledged interface for writing code. Secondly, the technology makes it easier to enter programming, allowing even beginners to create applications without deep technical knowledge.

Vibe coding itself did not arise from scratch. Although the term gained popularity thanks to the former director of artificial intelligence at the company Tesla and co-founder OpenAI Andrei Karpati, the very idea of ​ ​ programming through dialogue with artificial intelligence has been formed and developed over the past three years. In 2025, the industry had a clear name for a phenomenon that has long been gaining momentum. And, perhaps, it was thanks to such an accurate wording that it became easier to realize why this approach turned out to be so close and convenient for developers around the world.

In addition to vibe coding, there is know-how coding - these are two fundamentally different approaches to creating software products that are often compared, but rarely delimited. The bottom line is that both methods complement each other by serving different segments of users, rather than competing with each other. Developers interact with AI assistants who generate code, but must have an understanding of programming to effectively verify results. Know-coding (also known as low-coding) is useful for novice programmers or teams with limited development resources. Know-how is based on visual modeling using existing libraries of algorithms that will be executed by some ready-made application. That is, the user does not always interact with the code directly or minimize it. Although it is already clear that vibe coding can become an incentive for the growth of know-how of the market and accelerate the popularization of this format.

The modern ecosystem of services for vibe coding includes several key categories:

• large language models (LLMs),
• code editors with LLM integration,
• voice interfaces,
• platforms for fast generation of simple projects and sites,
• Automatic code validation and testing tools.

The fastest immersion in vibcoding is possible using various "one-button" services, where you can literally implement a simple application right in the browser with one request - among these you can try Google Firebase Studio, bolt.new or same.new. You can also try classic chat modes with LLM, that is, as usual, just write LLM "help me create an application that you can...." For such purposes, among Russian LLMs, you can try Yandex GPT 5 and GigaChat 2.0 from Sberbank. Foreign analogues - DeepSeek, ChatGPT-4o (OpenAI), Claude 3.7 Sonnet (Anthropic). But they are suitable if the user has already figured out "where to copy the code" so that it works. And if you haven't figured it out, then they will help you do it. And if the user wants to speed up the vibcoding process as much as possible, then you can use neural networks right in the code editor. For this, Cursor, GitHub Copilot, Tabnine are actively used, and in Russia - SourceCraft Code Assistant from Yandex Cloud and GigaCode from GitVerse (including IDE GigaIDE). And for accelerated generation of landings using AI, platforms like Framer AI and Tilda AI are used.

𝓲

Фото: NITU MISIS

Despite the rapid growth and variety of solutions, the vibe coding industry faces a number of technological limitations:

• Modern LLMs do not always preserve the project context throughout the session, which leads to a rewrite or breakdown of previously generated code.
• Often, models "forget" about parts of the task, and the developer has to manually clarify and supplement with teams.
• LLMs do not guarantee protection against vulnerabilities, especially in complex projects that require deep programming experience.
• Automation of interaction with Git and analogues is still limited, which reduces the convenience of working in a team.
• Although the technology already allows you to automate testing, the interfaces and services for vibe coding have not yet made this process completely convenient and automatic. It is necessary to finalize requests to the neural network and explain to AI what needs to be tested.

One of the most popular applications of vibe coding was prototyping. Many startups resort to this method to quickly assemble minimally viable products (MVPs) that allow the hypothesis to be tested early. Classic concept-checking techniques such as "The Wizard of Oz" and "Fake Door" allow ideas to be tested with minimal resources. In the first case, the functionality of the product is simulated manually: for example, the user interacts with the interface, not knowing that "behind the scenes" operations are performed by a person, not an algorithm. In the second, a fictitious element (button, page) is created, which does not carry any function, but collects data on user demand. With the advent of neural networks, a third was added to these methods - the generation of prototypes using AI. Neural networks automatically create working versions of the product based on the description, reducing development time from weeks to hours.

The use of code in artisanal projects and software for personal use has become available. For example, if a user has long wanted to make an application for accounting for products in the refrigerator, vibe coding allows you to implement such an idea in literally one or two evenings. Vibe coding made it possible to prototype ideas and test concepts faster. For example, when you need to finalize the visual part of the finished solution to show the customer what the final product will look like, it is enough to quickly make changes using AI. And with neural networks, you can create personal projects: what was previously developed in a week is now spent a couple of evenings - you just need to describe the task, and AI generates code.

However, the method raises serious concerns about information leakage, software support after its launch and the potential loss of fundamental programming skills by developers. The future is likely to include a hybrid approach that balances the benefits of both traditional methods and vibe coding.

Separately, experts note the educational potential of this method. Today's learning system is just beginning to figure out how to integrate LLM into the learning process. If in the humanitarian disciplines - philology, linguistics - there are already active discussions about academic integrity, then in the field of programming, vibe coding allows you to study code logic, approaches to solving problems and application architecture.

In addition, vibe coding allows you to automate tasks that go beyond classical programming. For example, with its help, you can quickly compose a SQL query - a special command for working with a database that allows you to extract the necessary information - write a formula for Excel or solve a unique problem related to a specific working context.

And finally, the most rapidly developing direction is corporate vibe coding. Already, companies see LLM integration as a way to optimize processes, speed up team development and improve team efficiency. The most optimistic economic effect according to forecasts will be achieved precisely in the B2B segment - when the vibe coding tools will begin to be used within organizations on an ongoing basis.

If you look into the future, you can imagine how at some point modern vibe coding technologies will intersect with neurointerfaces. Such a symbiosis is able to create fundamentally new scenarios: from designing the architecture of programs exclusively with the help of thought to intelligent testing based on mental modeling of user scenarios. One of the promising areas is already becoming the so-called AI agents - intelligent systems that can take over the comprehensive automation of the entire development cycle.

Tips for entering vibe coding:

• Master the basics of industrial engineering. The ability to correctly formulate a request for AI is the key to getting the desired result. For example, instead of the abstract "create a site for displaying weather," it is better to write "create an adaptive one-page application using React, which displays weather data from an external API."
• Consideration of the context constraint. Large language models have a limited "contextual window" - they may forget what was written earlier, especially when working on a complex project. One of the life hacks is to use separate chats for different modules or stages of development.
• Mastering basic debugging. Even the most advanced LLMs may be wrong - for example, to offer outdated library installation commands or incorrect module names. Sometimes it is faster to find and fix the error yourself than to try to fix it with AI for a long time.
• Choose the right technology stack. Theoretically, you can code a vibe in any language - at least in C++. But if the level of training is low, it is better to give preference to simpler and more common technologies, such as JavaScript, where it is easy to find documentation and examples.
• Understand what is being created. Vibe coding is a great way to see how ideas are implemented in code. It is necessary to follow what code the AI generates and try to understand how it works. This will help not only improve the result, but also gradually develop your own programming skills.

The main possibility of vibe coding is the almost complete absence of an entry threshold. With its help, you can create a simple working application without even having serious programming knowledge. However, the efficiency of work directly depends on how the user interacts with the neural network and which tools he chooses. For example, there was a case recently when a blogger on Twitter created a web service with monetization using vibcoding and he had real customers, but since he was without the background of creating any products, very soon his service was hacked and had to be closed. But it is known that this whole concept is definitely not a hype and it will find its application on the market, more and more new services, cool stories of success and failure will appear.

Radical simplification of programming: the era of AI coding

One of the most notable trends was the simplification of development software thanks to an approach called vibe coding. The essence of vibe coding is that modern (large language models LLM) can generate not only code fragments, but also entire projects based on dialogue with a person. Simply put, instead of detailed manual coding, the developer (or product manager without a technical background) describes the desired functionality in natural language, and the AI model writes code. The appearance of tools like (OpenAI ChatGPT code-oriented modes), Claude Code, Cursor and other similar assistants actually sets a new development format - through a chat with AI or even voice commands. Read more in. to article