AI-DRIVEN AUTONOMOUS CODE OPTIMIZATION USING REINFORCEMENT LEARNING AGENTS

Abstract

Modern software development faces challenges in performance tuning, especially in large-scale systems where manual code optimization is time-consuming and error-prone. Reinforcement Learning (RL)-based autonomous agents have emerged as a promising solution to automatically explore optimization strategies without human intervention. This research proposes an AI-driven framework that leverages RL agents to analyze source code, identify inefficiencies, and apply optimal transformations to enhance runtime performance. The system integrates program analysis, rewarddriven optimization, and continuous self-improvement through interaction with execution environments. Experimental results demonstrate that RL-driven optimization achieves up to 35–60% performance gains across benchmark programs. The findings highlight the potential of autonomous AI systems to revolutionize compiler design and software engineering.