Improving the speed and energy-efficiency of AI agents

Researchers from MIT and Microsoft have developed an intelligent system that automatically optimizes the design and deployment of agentic workflows—AI-powered software systems that chain together multiple models and tools to solve complex tasks. Developers can now describe their desired workflow in plain language, and the system automatically selects optimal models, tools, and hardware configurations while adjusting for priorities like cost or speed. When tested on several workloads, the new approach significantly reduced computational resource requirements and energy consumption compared to traditional methods without sacrificing performance. The research will be presented at the USENIX Symposium on Operating Systems Design and Implementation.