Conference Agenda

The rapid adoption of Generative Artificial Intelligence, especially Large Language Models (LLMs), is reshaping backend development by automating essential tasks such as code generation, automated testing, documentation, and API design. Despite their widespread use, the combined implications of LLMs on productivity, code quality, and security remain insufficiently consolidated in the current body of research. This study presents a Systematic Literature Review (SLR) aimed at analyzing how LLMs influence backend development processes, focusing on opportunities for efficiency as well as emerging security risks. Using the PICO methodology and PRISMA guidelines, 36 peer-reviewed studies from the Scopus database were evaluated.

Findings reveal that LLMs are predominantly integrated into hybrid development workflows, where they support developers by generating initial code for endpoints, validation layers, and database operations. These tools consistently improve productivity, particularly in high-complexity and high-risk domains such as finance and healthcare. However, the evidence also shows that AI-assisted code tends to contain a significantly higher density of vulnerabilities—including injection flaws, improper authentication logic, weak input validation, and misconfigured authorization checks—when compared to traditional development practices. The review also highlights a tendency among developers to overtrust AI-suggested code, which exacerbates security risks.

The study concludes that while LLMs are powerful enablers for accelerating backend development, their responsible adoption requires rigorous manual review, security-focused prompt engineering, and standardized metrics for quality evaluation. These insights provide a consolidated foundation for practitioners and researchers seeking to integrate LLM-based tools safely and effectively.