The rise of Large Language Models (LLMs) and AI agents is transforming software development, introducing new paradigms in automation and human-machine collaboration. This thesis, conducted in collaboration with Poseidon, a company specializing in low-code automation, explores the state of the art in LLMs and AI agents, analyzing their architectures, datasets, and benchmarks, with a focus on programming efficiency and business workflows. Building on this foundation, the study presents a case study in which an SQL agent is developed and evaluated for natural language database querying, enhancing user interaction. The evaluation employs both standard and custom metrics, utilizing a specifically designed dataset to assess the model's performance comprehensively and demonstrate its practical implications for accessibility and efficiency.