Agentic AI-Driven Software Engineering: Transforming Requirements Analysis, Code Generation, and Automated Testing
Abstract
ABSTRACT The integration of Artificial Intelligence (AI) into Software Engineering (SE) has undergone a fundamental paradigm shift, evolving from passive code completion assistance to fully autonomous, multi-agent cooperative systems. Agentic AI represents a transformative approach where specialized, goal directed AI agents leverage Large Language Models (LLMs), tool integration, and long-term memory to independently execute complex software development lifecycle (SDLC) tasks. This paper presents a comprehensive review of Agentic AI-driven software engineering, focusing specifically on three core pillars: requirements analysis, code generation, and automated testing. We analyze the underlying architectural frameworks, coordination protocols, and reasoning mechanisms that enable agentic workflows to decompose ambiguous specifications, generate syntactically and semantically sound code, and autonomously execute self-healing test suites. Furthermore, this review identifies critical open challenges, including context window constraints, error propagation, non-deterministic execution, security vulnerabilities, and human-in-the-loop governance. Through detailed analysis, quantitative comparisons, and structural frameworks, this study highlights the engineering significance of agentic software development and outlines key future research directions.
KEYWORDS: Agentic AI, Software Engineering, Multi-Agent Systems, Code Generation, Requirements Analysis, Automated Testing, Large Language Models, Autonomous SDLC.
Full Text:
PDF 54-66Refbacks
- There are currently no refbacks.