Software Engineering Practices: In the era of AI / LLMs

Abstract

The software development landscape is going through a major shift as generative artificial intelligence (AI) tools, including code assistants and autonomous agents, become increasingly widespread. Recent industry surveys indicate that more than 75% of developers currently use or plan to adopt AI-based solutions, with approximately half of professional developers utilizing these tools daily. Furthermore, organizational adoption is nearly universal, with reports suggesting that over 97% of companies incorporate AI into their development workflows for tasks such as code generation, documentation, code review, and automated testing. The anticipated benefits of these technologies are considerable. Empirical studies report productivity gains of up to 50% in coding speed and 33% in code refactoring efficiency. However, these advantages are accompanied by notable risks. AI-generated code can propagate security vulnerabilities and exacerbate technical debt. Moreover, evidence from a 2025 randomized controlled trial (RCT) shows that experienced developers using AI assistants were, on average, 19% slower when completing real-world programming tasks, challenging the assumption of universal efficiency gains. This paper discusses the current state of generative AI adoption in software engineering and synthesizes best practices for employing its potential while controlling associated risks. Following the structure recommended by the Journal of Computer Science and Technology Studies, the discussion is organized into the following sections: Introduction, Literature Review, Methodology, Results and Findings, Conclusion, Statements and Declarations, and References [11].