Redazione RHC : 16 September 2025 07:38
There’s a new phenomenon in the programming industry: vibe coding. Developers are increasingly using artificial intelligence agents to speed up their work, but they’re also encountering unpredictable problems. The stories of programmers who shared their experiences demonstrate that automated coding can make things easier or turn into a disaster.
Carla Rover, who has worked in web development for over 15 years and is now founding a startup with her son to build machine learning models for marketplaces, admits she was moved to tears when she had to start the entire project over again.
She trusted the AI-generated code and skipped a detailed review, relying instead on automated tools. When errors emerged during manual analysis and third-party audits, it became clear that the project was beyond saving. She believes treating AI as a full-fledged employee is a dangerous illusion. It can help shape ideas, but she’s not ready for independent accountability.
Rover’s experience is supported by large-scale statistics. According to a Fastly study, out of nearly 800 developers surveyed, 95% spend extra time fixing AI code, with the majority of the workload falling to senior specialists. They identify a wide range of issues, from dummy libraries to removing necessary program parts and vulnerabilities. This has even given rise to a new job title in companies: “Vibe code cleanup specialist.”
Feridun Malekzade, who has worked in development and design for over 20 years, describes the process with humor. She actively uses the Lovable platform, including for her own projects, and compares vibe coding to working with a stubborn teenager: you have to repeat the request many times, and in the end, the result partially meets the task, but is accompanied by unexpected and sometimes destructive changes. According to her calculations, half the time is spent formulating requirements, about 20% on generation, and up to 40% on correction. At the same time, AI is unable to think systematically and is prone to solving problems head-on, creating chaos when scaling functions.
Carla Rover observes that AI often finds inconsistencies in the data and, instead of admitting a mistake, starts providing convincing but false explanations. She describes the experience as like dealing with a toxic colleague. There’s even a social media meme about models like Claude responding to criticism by saying, “You’re absolutely right,” which is echoed by Austin Spyres of Fastly. He warns that AI is aiming for speed but ignoring correctness, leading to beginner-level vulnerabilities.
NinjaOne’s Mike Arrowsmith also discusses security. He argues that vibe coding undermines the foundations of traditional development, where multi-stage checks help identify flaws. To reduce risks, the company introduces “safe vibe coding” rules: limited access to tools, mandatory code reviews, and automated security checks.
However, despite all the criticism, the technology has firmly established itself in practice. It is ideal for prototypes, interface drafts, and routine tasks, allowing developers to focus on scalability and architecture. Rover admits that thanks to AI, it was able to process the interface faster, and Malekzadeh says that productivity is still higher than without the use of generators. Many developers call it an “innovation tax”: hours of debugging are needed, but the benefits in terms of speed and convenience outweigh the costs.
The conclusion is clear: Vibe coding is no longer an experiment, but has become the new norm. Experienced programmers know that AI cannot be released into production unsupervised, but they have already adopted it as a tool to accelerate processes.
The future of development now looks like this: a human sets the direction, an AI writes the code, and then the same human checks and fixes everything that’s been done.
Redazione