New System Optimizes AI Agent Speed and Energy Efficiency

The increasing complexity of AI agentic workflows presents a significant challenge for efficiency and resource management. These workflows, which involve chaining together various AI models and external tools to execute intricate tasks—such as analyzing video content and answering questions about it—are becoming foundational to modern AI applications. However, their fragmented nature often results in considerable inefficiencies, leading to unnecessary computation, higher energy consumption, and increased operational costs. To tackle these issues, a collaborative effort between researchers at MIT and Microsoft has led to the development of Murakkab. This novel system is engineered to optimize both the design and deployment of these multi-step AI workflows. According to Gohar Chaudhry, an electrical engineering and computer science graduate student at MIT and lead author of the paper, agentic workflows are rapidly becoming the backbone of cloud provider operations, making energy usage a paramount concern. Murakkab's core innovation lies in its ability to streamline the process of creating agentic workflows and automatically optimize their implementation. This intelligent approach allows cloud data centers, which deploy these applications for customers, to gain better insight into the workflow's internal structure. This visibility is crucial for allocating hardware resources in the most efficient manner possible. Without such optimization, developers would struggle to manually configure workflows optimally due to the vast number of possible configurations. The system's benefits extend beyond just technical performance. By enabling cloud providers to intelligently make these workflows more resource-optimal, Murakkab offers a win-win scenario, reducing both energy waste and financial expenditure. This is particularly relevant given the growing demand for AI agents and the associated computational load they place on infrastructure. The research highlights that it is currently very easy to over-allocate resources, leading to significant waste. Murakkab represents a critical step towards more sustainable and cost-effective AI deployment, ensuring that as AI agents become more prevalent and sophisticated, the underlying infrastructure can support them efficiently. The research was supported by the Semiconductor Research Corporation and the U.S. Defense Advanced Research Projects Agency, underscoring the strategic importance of optimizing AI agent performance.