Image source: Pixabay
Vibe Coding is a term coined by Andrej Karpathy (former Director of AI at Tesla, now at OpenAI) in early 2025 to describe a new programming approach using generative AI (Large Language Models, or LLMs). Instead of manually writing each line of code, developers simply enter a natural language prompt for the AI to generate a complete code block.
Popular tools like GitHub Copilot, Cursor Composer, and Replit Agent are helping many developers and startups quickly implement ideas on the web, mobile, and for software testing. The main strength of Vibe Coding lies in its rapid prototyping speed, making it suitable for the experimentation and idea demonstration phases.
Accelerated Development: It creates code quickly, which is a significant advantage for startups or individuals needing to demo an idea.
Reduced Tedious Work: The AI handles repetitive boilerplate code, allowing developers to focus on core architecture and features.
Easy Accessibility: New programmers can experiment quickly without needing a deep understanding of syntax.
Despite its appeal, Vibe Coding comes with several drawbacks and potential risks:
According to an August 2025 report by Fastly, over 95% of developers admitted they had to spend extra time fixing code generated by AI. The reason is that the generated source code often lacks accuracy and can even "hallucinate" functions, packages, or variables that don't exist. Furthermore, the AI may accidentally delete important content or skip necessary conditional checks, leading to unpredictable logic errors and significant risks when the code is deployed in a real-world setting.
In traditional programming, every code block is checked through multiple steps: code review, testing, security audits, and static analysis to ensure safety. When using Vibe Coding, however:
The AI generates code quickly but doesn't prioritize security standards. For example, it might automatically insert an API connection block without encrypting sensitive data or use an outdated library with known vulnerabilities.
It doesn't provide clear error notifications. If there's an issue, the AI often "hallucinates" or generates alternative code without a warning. This can cause developers to miss critical security flaws.
This is especially dangerous for startups/small businesses. Under pressure for a fast launch, many companies may deploy AI-generated code into production systems without a proper security testing process, opening the door for hackers to exploit vulnerabilities.
In other words, Vibe Coding prioritizes speed over safety, leading to a "security blind spot"—potential vulnerabilities that developers don't have time to detect.
Using AI to write code is like hiring an inexperienced intern: they can get the job done quickly, but someone with more experience is needed to fix it.
More Debugging: Developers have to run tests, debug logic errors, and fix every line the AI wrote incorrectly.
Rewriting Code: Often, the AI's code is cumbersome, lacks proper variable/function naming standards, or doesn't fit the project's architecture, forcing developers to rewrite it from scratch.
Eliminating "Garbage Code": The AI often generates unnecessary extra code, making the project heavy and difficult to maintain.
Long-term Time Cost: While it initially saves time writing code, developers later spend 30-40% of their time just fixing it, negating the initial speed advantage.
Vibe Coding is like having a new assistant on a construction project: he can quickly lay bricks and mix cement to get a wall up in a short time, but without close supervision, the bricks might be laid crooked, the cement mixed in the wrong ratio, and the wall could easily crack. Ultimately, the lead builder—the developer—still has to spend time fixing and reinforcing it to prevent the structure from collapsing.
If there is excessive reliance on Vibe Coding, large-scale projects can suffer from unorganized and difficult-to-maintain source code in the long run. Since the AI often doesn't focus on adding comments or building documentation, other developers will have a hard time taking over the project. Furthermore, the fragmented code structure makes adding or expanding new features complex, posing risks of delayed timelines and increased development costs.
Vibe Coding is an inevitable trend in the AI era, offering benefits in terms of speed and convenience. However, if not controlled, it turns developers into "AI garbage collectors," requiring significant effort to fix errors and face ongoing security risks.
.png)
CyberTech đồng hành cùng doanh nghiệp trên hành trình chuyển đổi số thông qua các giải pháp AI, phát triển phần mềm và công nghệ thông minh, góp phần nâng cao năng lực cạnh tranh trong kỷ nguyên số.
