AI-Driven Software Testing: An Adaptive Agentic Framework for Context-Aware Test Generation, Validation, Prioritization, and Refinement

Abstract

Software testing is a critical activity in the software development life cycle, but conventional test automation still requires substantial human effort for test design, maintenance, execution, and analysis. Recent advances in artificial intelligence (AI), machine learning (ML), natural language processing (NLP), and large language models (LLMs) have created new opportunities to automate and improve several testing activities. This paper presents a structured review of AI-driven software test automation, emphasizing test-case generation, test prioritization, defect prediction, test execution, failure analysis, and automated test repair. The study also examines the emerging role of LLMs in understanding requirements and source code and in producing developer-readable tests. AI-assisted approaches can improve coverage, fault detection, and testing productivity, but generated tests may contain hallucinations, weak assertions, flaky behavior, or inadequate contextual understanding. Therefore, human supervision, execution-based validation, coverage measurement, mutation testing, and secure integration with CI/CD pipelines remain important. The paper proposes a practical AI-driven testing workflow that combines requirement analysis, intelligent test generation, automated execution, feedback, and continuous refinement