AI RESEARCH
Beyond Local Code Optimization: Multi-Agent Reasoning for Software System Optimization
arXiv CS.AI
•
ArXi:2603.14703v1 Announce Type: cross Large language models and AI agents have recently shown promise in automating software performance optimization, but existing approaches predominantly rely on local, syntax-driven code transformations. This limits their ability to reason about program behavior and capture whole system performance interactions.