<![CDATA[Vibe coding is a software development approach that intensively leverages Large Language Models (LLMs), where developers act as directors and validators rather than direct code writers. This study aims to: (1) characterize vibe coding scientifically, (2) measure its impact on productivity and code quality, (3) identify associated risks, and (4) formulate practical adoption recommendations. A Systematic Literature Review (SLR) of 47 publications from Scopus, IEEE Xplore, and ACM DL (2021–2025) was conducted, combined with an eight-week controlled experiment on two homogeneous groups of fi-nal-year Informatics Engineering students (control n=30, experimental n=30; Mann-Whitney U=427, p=0.83). Results showed the vibe coding group completed an average of 8.3 features (SD=1.2) vs. 5.4 features (SD=0.9) in the conventional group, a statistically significant difference (t(58)=10.47, p<0.001, Cohen’s d=2.70). However, the vibe coding group exhibited a significantly higher bug rate (5.1 bugs/100 LOC vs. 3.8 bugs/100 LOC; t(58)=3.84, p<0.001) and lower self-understood code pro-ficiency (5.9 vs. 8.4 out of 10; t(58)=-9.12, p<0.001). The study concludes that vibe coding effectively increases short-term productivity but risks technical skill erosion and hidden technical debt without rigorous software engineering practices.]]>
Copyrights © 2026
