Autonomous AI Agents for Software Development: From Code Generation to Automated Testing

This research paper presents a proposed framework for Autonomous AI Agents for Software Development, focusing on the transition from code generation to automated testing, debugging, and software verification. The paper introduces a modular, role-based architecture consisting of a planner agent, context retriever, coder agent, tester agent, debugger agent, and reviewer agent. These components work together through a bounded Plan-Code-Test-Repair workflow to understand software requirements, retrieve relevant code, generate changes, execute tests, diagnose failures, and improve code quality. The proposed framework emphasizes functional correctness, automated test generation, repository-level issue resolution, secure tool execution, human approval, and auditability. It also outlines an evaluation protocol using established benchmarks and metrics to assess code correctness, test quality, efficiency, cost, reliability, and security. This work synthesizes existing research on large language models, reasoning-and-acting agents, self-debugging, multi-agent systems, and automated software testing. It provides a conceptual architecture and evaluation plan for future implementation and experimentation. No implemented system or benchmark results are claimed in this paper. The research aims to support the development of more reliable, secure, and transparent AI-assisted software engineering systems.