Evolution of programming paradigms and the vibe coding phenomenon: evaluation of opportunities, risks and research perspectives

DOI: 10.31673/2412-9070.2026.318119

Abstract

The rapid development of Artificial Intelligence (AI) systems and Large Language Models (LLMs) has led to a fundamental shift in software development approaches, engendering new paradigms of human-computer interaction. This article is devoted to an in-depth analysis of the historical evolution of programming — from direct hardware manipulation via machine code and early compilers to the establishment of high-level languages, object-oriented programming, and contemporary AI-driven methodologies. Special attention is paid to the concept of "vibe coding" — an emergent development approach that involves creating software products primarily through highly agentic, iterative interactions with AI assistants using natural language prompts, often with minimal manual coding or even without the developer's complete understanding of the generated algorithms and internal system architecture.
Throughout the research, the capabilities, advantages, and drawbacks of this approach are comprehensively evaluated. It has been established that deploying generative AI in a vibe coding format significantly lowers the barrier to entry into the IT industry for individuals without formal technical backgrounds. It facilitates unprecedented acceleration in prototyping and the development of Minimum Viable Products (MVPs). Empirical studies reveal a 26–30% overall increase in task completion productivity and a surge in iteration speed, particularly benefiting junior developers. Conversely, the study identifies and systematizes critical shortcomings and hidden risks associated with vibe coding that jeopardize long-term project viability. The existence of a "productivity paradox" is demonstrated, revealing that highly experienced engineers (Senior developers) can spend up to 19% more time executing complex tasks due to the cognitive overhead required to audit and debug hidden logical errors and AI model "hallucinations."
Furthermore, acute cybersecurity vulnerabilities are scrutinized: according to the Veracode report (2025), aggregated statistical data indicates that up to 45% of tested AI-generated code snippets contain critical vulnerabilities that align with the OWASP Top 10 classification. Addi-tionally, it is proven that excessive, uncritical reliance on AI without strict architectural oversight leads to the accumulation of a specific "Dark Debt," progressive architectural fragmentation (Context Rot), and the gradual degradation of fundamental engineering skills among developers (Skill Atrophy).
Based on the synthesis and analysis conducted, it is substantiated that vibe coding possesses absolute utility as a powerful tool for ultra-fast prototyping, automating routine tasks, and generating boilerplate code. However, this paradigm is fundamentally incapable of completely replacing the rigorous discipline of classical software engineering in the context of building critical, highly scalable, and security-oriented enterprise systems. Key directions for future scientific and applied research are delineated, including the development and implementation of "Context Engineering" methodologies, the creation of integrated Agentic AppSec systems for real-time semantic code verification, and a profound investigation into the socio-technical impact of AI on the educational and professional progression of future engineers.

Keywords: programming paradigms, artificial intelligence, vibe coding, technical debt, cybersecurity, code generation, large language models, software engineering.