What is Vibe coding and when should you use it (or not)?

The invention of Generative AI and Large Language Models has recently given rise to "vibe coding" as a new paradigm of software development in which developers use natural language to state their high-level intent rather than writing or even being aware of the computer code. While the great ease of use of this new paradigm presents outstanding opportunities for Rapid Application Development, it also creates a significant risk of misuse, especially with inexperienced developers as its adoption outpaces formal understanding of the underlying technologies. Currently, there is a lack of empirical research analyzing the fundamental behaviors and code quality of different LLM tools. Moreover, as the software industry is rapidly undergoing profound changes, there is no consensus among practitioners even on the definition of vibe coding. This paper provides a formal definition of the vibe coding paradigm, a list of its advantages and disadvantages, a systematic comparation of vibe coding to other AI-assisted programming approaches, and suggestions for the use of this new paradigm. We argue that vibe coding should be avoided as a primary tool for learning programming, particularly in academic settings, as it can obscure foundational concepts. Furthermore, it is ill-suited for large-scale or productionlevel projects due to significant debugging and maintenance overhead. Instead, we conclude that it should be used only as a very specific productivity tool for experienced developers engaged in rapid prototyping and experimentation.