Agentic AI-Based Automated Test Case Generation and Execution: A Survey

As software systems grow in scale and complexity, traditional manual testing and rule-based automation increas-ingly struggle to keep pace. Generating executable test cases from requirement documents and UI designs remains largely a manual, error-prone task — one that is slow, dependent on individual expertise, and prone to missing edge cases. Recent advances in Large Language Models (LLMs) have opened new possibilities here: these models can interpret loosely written requirements and produce structured test scenarios without task-specific training. However, most existing approaches treat test generation as a single inference step, with no mechanism for orchestration, execution feedback, or multimodal input. This paper surveys AI-driven testing methods — covering deep learning approaches, LLM-based generation, and emerging multi-agent architectures — and systematically identifies where current systems fall short, particularly around scalability, closed-loop refinement, and han-dling of visual UI inputs. Based on these findings, we propose a conceptual agentic AI framework comprising three coordinated agents: a perception agent that processes both textual and visual inputs, a reasoning agent that generates test cases covering functional, boundary, negative, and edge-case scenarios, and an execution agent that runs the tests and feeds results back into the pipeline. The proposed framework aims to substantially reduce manual QA effort, improve test coverage, and move toward software testing that can largely sustain itself. Index Terms—Software Testing, Agentic AI, Large Language Models (LLMs), Natural Language Processing (NLP), Computer Vision (CV), Test Case Generation, Multimodal Learning, Soft-ware Development Lifecycle (SDLC)