Microsoft Azure, working alongside researchers from MIT, has developed a new artificial intelligence system named Murakkab designed to enhance the efficiency and speed of agentic workflows. These complex AI-powered processes combine multiple models and tools to perform intricate tasks but have traditionally suffered from significant computational overhead, energy consumption, and associated costs.
What Happened
The Murakkab system intelligently automates the configuration of agentic workflows based on high-level user intent rather than requiring developers to manually specify every detail. By dynamically selecting the optimal AI models, tools, hardware configurations, and resource allocations, Murakkab adapts in real time to meet performance targets such as speed, accuracy, and cost constraints. It was tested on workloads including video question answering and code generation, where it demonstrated remarkable reductions in computational unit usage—down to about 35 percent compared to conventional methods—and lowered energy consumption to roughly 27 percent, with less than 25 percent of the typical cost. The system adjusts hardware deployment and scheduling live during cloud execution, maximizing efficiency based on user priorities.
Key Facts
Murakkab was developed through collaboration between Microsoft Azure and the Computer Science and Artificial Intelligence Laboratory at MIT. Officially unveiled in a paper presented at the USENIX Symposium on Operating Systems Design and Implementation, the platform addresses the challenge of configuring agentic workflows, which orchestrate multiple AI agents and tools to complete multi-step tasks such as video analysis and data processing.
The research team includes Gohar Chaudhry, an MIT graduate student and lead author; Adam Belay, MIT associate professor; and Ricardo Bianchini, Microsoft Azure corporate vice president and technical fellow. The system’s core innovation eliminates the need for developers to hard-code workflow specifics upfront, instead enabling descriptions in plain language that Murakkab translates into efficient workflow and hardware plans. Supported in part by the Semiconductor Research Corporation and DARPA, the platform provides cloud providers with visibility into multiple workloads, allowing smarter resource sharing and deployment management.
What This Means
Murakkab represents a significant step forward in reducing the environmental and economic impact of deploying AI-driven applications on large cloud platforms. As agentic workflows become increasingly integral to cloud services, inefficient resource use not only inflates operational costs but also contributes to higher energy demands with associated environmental concerns. By optimizing workflow configurations dynamically and at scale, Murakkab offers cloud providers a tool to meaningfully cut wasteful spending and carbon footprint while maintaining user experience quality.
For developers and enterprises, this system reduces technical complexity and accelerates innovation by removing the need for exhaustive manual tuning of AI components and hardware setups. It also future-proofs applications against rapid AI model improvements and hardware advancements, since Murakkab can adapt workflows automatically with no developer intervention. Ultimately, this could lead to broader adoption of AI services by lowering the cost barriers and ecological costs currently associated with sophisticated AI applications.
Background
Agentic workflows, combining multiple autonomous AI agents and external tools, have emerged as essential components behind advanced AI applications. However, the complexity of configuring these systems has grown, as developers must determine which models and tools to employ and how to allocate hardware resources efficiently. Traditional approaches require upfront, fixed decisions that cannot easily incorporate improvements such as new AI models or hardware accelerators, making workflows prone to inefficiencies and waste.
What Comes Next
The research team plans to extend Murakkab’s capabilities to manage more complex agentic workflows and larger cloud computing clusters. Additionally, they aim to explore how this system can optimize new types of AI-driven applications, further enhancing operational efficiency and sustainability at scale.
Sources
This article is based on reporting and publicly available information from the following sources:
Read more Artificial Intelligence stories on Goka World News.