AI-powered software testing tools: A systematic review and empirical assessment of their features and limitations

TL;DR

This paper systematically reviews AI-powered software testing tools, categorizes their features, and empirically evaluates two tools, revealing their strengths in efficiency and limitations in handling complex UI changes.

Contribution

It provides a comprehensive taxonomy of AI testing tools and empirically assesses their effectiveness and limitations in real-world scenarios.

Findings

AI tools improve test efficiency and reduce maintenance

Limitations include difficulty handling complex UI changes

Current tools suffer from false positives and limited domain understanding

Abstract

Context: The rise of Artificial Intelligence (AI) in software engineering has led to the development of AI-powered test automation tools, promising improved efficiency, reduced maintenance effort, and enhanced defect-detection. However, a systematic evaluation of these tools is needed to understand their capabilities, benefits, and limitations. Objective: This study has two objectives: (1) A systematic review of AI-assisted test automation tools, categorizing their key AI features; (2) an empirical study of two selected AI-powered tools on two software under test, to investigate the effectiveness and limitations of the tools. Method: A systematic review of 55 AI-based test automation tools was conducted, classifying them based on their AI-assisted capabilities such as self-healing tests, visual testing, and AI-powered test generation. In the second phase, two representative tools were selected for the empirical study, in which we applied them to test two open-source software systems. Their performance was compared with traditional test automation approaches to evaluate efficiency and adaptability. Results: The review provides a comprehensive taxonomy of AI-driven testing tools, highlighting common features and trends. The empirical evaluation demonstrates that AI-powered automation enhances test execution efficiency and reduces maintenance effort but also exposes limitations such as handling complex UI changes and contextual understanding. Conclusion: AI-driven test automation tools show strong potential in improving software quality and reducing manual testing effort. However, their current limitations-such as false positives, lack of domain knowledge, and dependency on predefined models-indicate the need for further refinement. Future research should focus on advancing AI models to improve adaptability, reliability, and robustness in software testing.