GA4GC: Greener Agent for Greener Code via Multi-Objective Configuration Optimization
Jingzhi Gong, Yixin Bian, Luis de la Cal, Giovanni Pinna, Anisha Uteem, David Williams, Mar Zamorano, Karine Even-Mendoza, W. B. Langdon, Hector Menendez, Federica Sarro
TL;DR
This work tackles the environmental and scalability costs of autonomous coding agents by introducing GA4GC, a multi-objective optimization framework that jointly tunes agent runtime and code performance. Using NSGA-II to explore a configuration space over LLM hyperparameters, agent constraints, and prompt templates, GA4GC discovers Pareto-optimal setups that substantially reduce runtime while maintaining or improving correctness on the SWE-Perf benchmark. Key findings include up to 135× hypervolume improvement and a 37.7% runtime reduction, with temperature emerging as the most influential hyperparameter and actionable strategies for runtime-focused, performance-focused, or balanced deployments. The approach enables context-aware, greener industrial deployment of coding agents and provides practical guidelines for practitioners performing green SBSE in real-world settings.
Abstract
Coding agents powered by LLMs face critical sustainability and scalability challenges in industrial deployment, with single runs consuming over 100k tokens and incurring environmental costs that may exceed optimization benefits. This paper introduces GA4GC, the first framework to systematically optimize coding agent runtime (greener agent) and code performance (greener code) trade-offs by discovering Pareto-optimal agent hyperparameters and prompt templates. Evaluation on the SWE-Perf benchmark demonstrates up to 135x hypervolume improvement, reducing agent runtime by 37.7% while improving correctness. Our findings establish temperature as the most critical hyperparameter, and provide actionable strategies to balance agent sustainability with code optimization effectiveness in industrial deployment.