According to Andrej Karpathy, who coined this term, vibe coding refers to a programming technique that allows individuals to depend on artificial intelligence (AI) to write code for their desired software’s specific features. It’s not a fad, but a new programming way that has changed the role of coders. So, what are exactly the benefits of using vibe coding, and does it present any potential challenges? This article will help you answer these questions.
Who Can Vibe Code?
Theoretically, anyone can vibe code, regardless of their coding expertise and experience.
- Those whose jobs are often involved in programming (e.g., software engineers or data analysts) can benefit from vibe coding. Accordingly, they can let AI do repetitive yet simple coding tasks while focusing on more complex aspects of a software development project that LLMs haven’t handled.
- Vibe coding also helps aspiring developers who want to learn how to code and bring their idea to life. The case of Prasad Naik is a typical example. As a licensed professional mechanical engineer, Naik isn’t a software developer and had no previous experience with JavaScript. Yet he successfully built two software solutions using JavaScript for internal use, thanks to ChatGPT’s step-by-step guidance.
- Non-programmers can offer natural language prompts for AI coding tools to convert their ideas into possibly working features and functionalities. With vibe coding, even those without a programming background can ‘code.’
FURTHER READING: |
1. Chatbot Integration With ChatGPT: A 5-Step Guide |
2. What is Generative AI? A Beginner's Guide in 2025 |
3. 10 Leading Generative AI Tools You Should Invest in 2025 |
When Should You Vibe Code?
Vibe coding allows anyone to code, as we mentioned. But it’s just a theory. LLMs, albeit advanced and intelligent, can still generate code with errors and bugs. When you vibe code your software, it means you just offer a guide to AI coding tools and say yes to all its changes or decisions without reviewing or adjusting the code, which may be incorrect and contain error messages. Like the way Karpathy described vibe coding:
I ask for the dumbest things like “decrease the padding on the sidebar by half” because I’m too lazy to find it. I “Accept All” always, I don’t read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I’d have to really read through it for a while. Sometimes the LLMs can’t fix a bug so I just work around it or ask for random changes until it goes away.
With this claim, vibe coding isn’t always a good choice for every programming project due to its high risk of generating defective code lines. So, when is it ok to vibe code? Here are some situations you can use AI tools for vibe coding:
- You vibe code for… fun, prototypes, and a short-term purpose. Do you want to experiment with AI and self-develop a simple app despite no coding knowledge? If yes, testing your app idea with AI coding tools and letting it build without the intention to maintain the code for long-term use is a good way.
- The algorithms that generate your app’s code have possibly been trained on existing code for similar tasks. For example, you build a mobile app using common technologies to create to-do lists and send reminders 30 minutes before each item is due. AI tools are already trained on this functionality. Also, any incurred bug is already fixed to ensure the relevant code is always correct. This helps you “accept all” the code AI tools create without worrying about its quality, as well as build a working solution.
- Vibe coding is a good option if your project is low stakes and doesn’t require high security levels. As we mentioned, the code generated by vibe coding might contain bugs and security issues. Without reviewing the code or having a technical background to correct the bugs, your defective code may harm others. Particularly, this code can accidentally leak confidential information through security holes, result in money loss, or damage reputation. So if you build software that may be used by different people and heavily involves data security, don’t code with the vibes.
Benefits of Using Vibe Coding
The term “vibe coding” has just been introduced. Therefore, there’s no official discussion about its advantages or disadvantages in a professional conference or committee. However, with its increasing popularity, it’s not too hard to observe the potential benefits of using vibe coding:
1. Speed Up Common Coding Tasks
Vibe coding is obviously helpful in creating routine or boilerplate code that involves popular programming languages and simple, common patterns. As LLMs are trained on large amounts of such code, they can create syntactically correct and functionally reliable code for common situations. For this reason, LLMs can speed up common and routine coding tasks, allowing programmers and coders to focus on more complex problem-solving and context-relevant tasks of software development.
2. Democratize Software Development
Significant advancements in deep learning make AI tools, typically large language models like ChatGPT, smarter. They can automatically learn everything in a somewhat rapid fashion, understand your conversational languages, and create outcomes based on your input requests. For this reason, they use what they learned from vast amounts of code to let you code with the vibes. This makes coding accessible to anyone, even when they have no programming background or experience.
3. Provide Opportunities for Learning and Creativity
One widely recognized benefit of using code is that it opens an avenue for people of all technical expertise to test their new ideas and learn new technologies. Jason Touleyrou, a data engineering lead at Corewell Health, leveraged AI-powered coding tools to experiment with his new ideas quickly. He believed that speed to ideation is essential, and with AI’s assistance, he can test his idea’s feasibility with new programming languages. This allowed him to define his project idea clearly before moving to the coding and testing phases.
Limitations of Vibe Coding
Vibe coding is only beneficial when you implement a hobby project, test new ideas, or learn new technologies with throwaway code. In fact, due to AI’s visible limitations, coding professional software with the vibes still presents more challenges than benefits. If you take the definition of vibe coding by heart (that means ignoring the AI-generated code’s quality), you’ll soon realize that using vibe coding for a complex, professional project is now hard to implement.
Discussions of experienced programmers and engineers on Reddit or StackOverflow also indicated that vibe coding is terrible for their work at times:
Accordingly, vibe coding now comes with the following challenges:
1. Vibe coders don’t understand the code and care for its quality
One noticeable feature of vibe coding, as mentioned by Karpathy, is ignoring the meaning and quality of the code generated by LLMs. Accordingly, vibe coders don’t need to completely understand why these AI tools generate such code and whether AI comprehends their input requests correctly to produce appropriate outcomes. Without a full understanding of the code, you hardly detect bugs, defects, and security vulnerabilities.
Even when you blindly let AI coding tools create tests and check the performance of their generated code lines (without caring whether these tests are technically correct), not understanding how these tools work also makes debugging, maintenance, and security ineffective.
This limitation of vibe coding is also a reason why one member in StackOverflow criticized vibe coding as not a real help for non-programmers. Without coding knowledge or experience, they struggle to debug or fix ‘disgusting’ code. Even for experienced programmers, depending heavily on AI-generated code without comprehending it can result in serious tech debt in the future.
2. AI code’s quality doesn’t always reach standards
Even when AI tools can write technically correct code, it doesn’t mean the code is well-structured or performs optimally. It can leverage intricate logic, be organized in a hard-to-modify way, or have inconsistent naming conventions.
The reasons behind this may stem from both user prompts and the nature of AI-powered models themselves. On one hand, users struggle to give clear prompts for LLMs to fully comprehend. This, however, turns out to be a common problem encountered by many software engineers. According to Taher Vohra, clarifying what you want AI to do is much harder than doing it yourself.
On the other hand, even with clear and well-intended requirements, AI models may produce poor-quality outcomes at times. The primary goal of AI is to fulfill your input requests as effectively as possible based on vast amounts of code they’re taught. They don’t inherently own the ability to dig into the root cause, examine long-term code maintainability, and follow traditional software engineering practices (e.g., modular design or documentation). Therefore, they tend to create immediate or simple solutions.
3. Vibe coders may ignore security issues
AI tools make software development easier than ever before. With natural language prompts, you can get code lines in seconds. However, the nature of coding with the vibes can lower the guard of users, especially non-programmers, about security issues.
AI-generated code looks safe superficially. But underneath, it can contain serious security vulnerabilities. This can be because the knowledge AI models learn from may be outdated, incomplete, or incorrect. Therefore, their code may have hidden flaws like outdated encryption. Not to mention that non-programmers may not detect security weaknesses as they have no coding knowledge to recognize them.
Further, AI may come up with unusual or complicated ways of generating code. Pure vibe coding, therefore, can make it hard for human coders to debug, update, and add new features on top of this code later.
Involving Human Oversight in AI Coding (But It’s No Longer Vibe Coding)
Using AI to generate, autocomplete, and debug code is increasingly popular, with a 25.8% annual growth from 2024 to 2030. But AI coding assistants aren’t always perfect. Therefore, it’s not surprising that AI-generated code may have errors and bugs.
Simon Willison, an independent software engineer, provided his point of view about vibe coding when being interviewed by Ars Technica. He found vibe coding interesting in experimenting with a specific idea and observing whether it works. However, “vibe coding your way to a production codebase is clearly risky,” said Willison. It’s because most of the work software developers are involved in developing existing systems, where understanding the underlying code and ensuring its quality are essential.
So, one practical approach various seasoned developers advise AI users to do is just consider AI’s code as a first draft and take advantage of its ability to create boilerplate code. AI’s code is imperfect. So we need experienced developers to review and refactor it (if any). This will ensure it is well-structured and easy to maintain, and complies with the best software engineering practices. However, if there’s proper overview and oversight over AI’s code, it’s no longer called ‘vibe coding’. So when AI hasn’t reached the professional level of human engineers, vibe coding is only a decent option for hobby projects and throwaway code.
Conclusion
This article has given you the potential benefits and challenges of using vibe coding in practice, based on our observations. In fact, coding with the vibes isn’t a good option for complex, context-relevant tasks, but just for hobby or throwaway weekend projects. It’s because AI hasn’t reached the professional level of human engineers to analyze the broader context and long-term maintainability of AI-generated code. Despite some limitations, AI still proves helpful if you know how to take advantage of its strong points and combine them with human oversight.
What do you think about vibe coding? Share your idea with us on Facebook, X, and LinkedIn! And don’t forget to subscribe to our blog to receive more interesting news!
Also published on
Share post on
Read more topics
