LARGE LANGUAGE MODELS IN AI-AUGMENTED DEVSECOPS PIPELINES: EXPLORING SECURE SOFTWARE ENGINEERING PRACTICES AND ARCHITECTURAL CHALLENGES

Abstract

The increasing adoption of Large Language Models (LLMs) has transformed AI-augmented DevSecOps by enhancing secure software engineering practices throughout the Software Development Life Cycle. Despite their growing use, limited research has explored how practitioners experience the integration of LLMs within secure software development environments and the architectural challenges that accompany their implementation. This study employed a qualitative research design using semi-structured interviews with 20 software engineers, DevSecOps practitioners, security architects, cloud specialists, AI engineers, and technology consultants. The collected data was analyzed using Braun and Clarke's Reflexive Thematic Analysis. Five major themes emerged: LLMs as intelligent enablers of secure software engineering, the continuing importance of human oversight, architectural integration challenges, governance and explainability for responsible AI adoption, and the future of collaborative AI-enabled DevSecOps. The findings reveal that LLMs significantly improve secure coding, vulnerability analysis, documentation, and operational efficiency while introducing concerns related to model reliability, data privacy, governance, software supply chain security, and regulatory compliance. The study proposes a practitioner-informed action framework that emphasizes human-centered governance, secure architectural integration, and continuous validation, thereby contributing practical and theoretical insights for the responsible adoption of LLMs within enterprise DevSecOps ecosystems