The Impact of Large Language Models on Software Development Productivity: A Systematic Review and Critical Analysis

This paper presents a systematic review of the impact of Large Language Models (LLMs) on software development productivity. Since their widespread adoption in late 2022, LLM-powered coding assistants have been posited as revolutionary tools capable of delivering order-of-magnitude productivity gains. However, empirical evidence presents a complex and often contradictory picture. This review synthesizes findings from 37 peer-reviewed studies, industry case studies, and controlled experiments to analyze the multifaceted effects of LLMs across the software development lifecycle. We ground our analysis in established productivity frameworks, such as SPACE and DORA metrics, to provide a holistic assessment beyond simple output measures. Our findings reveal that while LLMs demonstrably accelerate specific tasks like boilerplate code generation and information retrieval, they also introduce significant risks, including degraded code quality, security vulnerabilities, increased cognitive load for verification, and a potential reduction in team collaboration. Notably, we explore the "productivity paradox," where developers perceive significant efficiency gains even when objective measures indicate a slowdown. We conclude that the impact of LLMs is not a universal enhancement but a complex trade-off, heavily contingent on developer experience, task complexity, codebase size, and the quality of human-AI interaction. The paper identifies critical research gaps, particularly the need for longitudinal and team-based studies, and offers recommendations for practitioners on leveraging LLMs effectively while mitigating their inherent risks.