AI-driven software development: comparative study of code generation tools

Abstract

Inteligența Artificiala a devenit un element esentțial al ingineriei software moderne, influențând modul ın care codul este proiectat, generat, validat și menținut. Progresele recente în modelele de limbaj de mari dimensiuni și în sistemele agentice permit fluxuri de lucru tot mai autonome, în care AI asistă raționamentul și implementarea în mai mulți pași.

Artificial intelligence has become a core part of modern software engineering, changing how code is designed, produced, validated, and maintained. Advances in large language models and agent-based sys- tems enable workflows that are increasingly autonomous, where AI can support reasoning, implementation, and multi-step execution. This thesis studies these changes from both theoretical and empirical perspectives. The theoretical component reviews the current ecosystem of AI-assisted coding tools, cloud-native infrastructure, multi-agent architectures, and evolving models of human–AI collaboration. It introduces a distinction between traditional digital skills and emerging AI usage skills, highlighting why critical think ing, verification, and human oversight remain essential for reliable outcomes. The experimental component evaluates GPT-5.2 by generating an Inventory and Stock Management command-line system under three prompting conditions: minimal instructions, partial context, and a full Business Requirements Document prompt. Business logic is taken entirely from a paired thesis to preserve methodological consistency. The results show that prompt completeness strongly influences correctness, architectural quality, and requirement alignment, indicating that effective AI-assisted development depends on structured guidance and informed human control.