The Dual-Edged Sword: A Comprehensive Analysis of Generative AI in the Modern SDLC

Generative artificial intelligence (GenAI) tools, particularly large language model (LLM)-based code assistants such as GitHub Copilot and Amazon CodeWhisperer, have rapidly permeated modern software engineering workflows, promising substantial improvements in developer productivity and code generation efficiency. However, the empirical evidence regarding their benefits and associated risks remains dispersed across heterogeneous studies with varying methodologies and evaluation criteria. This paper presents a systematic review of 15 peer-reviewed and industry publications from 2022 to 2025, examining the impact of GenAI on four critical dimensions of software development: developer productivity, automated testing and bug detection, security vulnerabilities in generated code, and legacy code refactoring. Our analysis reveals a pronounced productivity paradox: while controlled experiments demonstrate that developers using AI assistants complete tasks 55–57% faster, concurrent security evaluations indicate that 40–74% of LLM-generated code contains exploitable vulnerabilities mapped to common weakness enumerations (CWEs). This work contributes a unified analytical framework for assessing GenAI tools in software engineering, synthesizes quantitative evidence across productivity and security dimensions, and identifies critical gaps necessitating future research to ensure that efficiency gains do not compromise software reliability and security.