Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Large Language Models present transformative potential for managing complex cloud infrastructure across multiple providers. While current applications primarily utilize LLMs reactively to generate infrastructure-as-code or analyze snapshots of system state, significant opportunities exist for more advanced implementations that maintain continuous awareness of infrastructure state across provider boundaries. Key challenges include context management constraints, preventing hallucinations in generated configurations, and maintaining consistency across diverse cloud environments. Emerging techniques like retrieval augmented generation, feedback loops, and multi-model verification architectures show promise in addressing these challenges. However, substantial risks accompany these innovations, including model hallucinations that could lead to catastrophic misconfigurations, security vulnerabilities from inadequate access controls, and economic barriers related to computational costs and organizational change management. The evolution toward autonomous infrastructure agents requires careful consideration of technical constraints, operational risks, security implications, and adoption challenges that organizations must navigate to realize the benefits of LLM-driven multi-cloud orchestration while maintaining reliability and security in enterprise deployments.