CAMBRIDGE — Researchers from MIT and Microsoft announced on June 25, 2026, the development of Murakkab, a system designed to optimize the design and deployment of AI-powered agentic workflows. Murakkab enables developers to describe their workflow intent in plain language.

The system automatically selects models and tools, determines hardware configuration, and allocates computational resources based on user priorities, such as cost or speed. Murakkab also dynamically adjusts configurations during execution to adapt to these priorities. Additionally, the system provides cloud providers with visibility into multiple workloads, which facilitates the efficient sharing of computational resources. Murakkab is an Urdu word meaning a composition of things.

Gohar Chaudhry, an electrical engineering and computer science graduate student at MIT and the lead author of the paper on Murakkab, explained the increasing complexity of AI systems. Chaudhry stated that agentic workflows are quickly becoming the backbone of what cloud providers are doing. He also noted that over-allocating resources is easy and can waste energy and money.

In tests conducted on agentic workflows for tasks including video question-and-answer and code generation, Murakkab utilized approximately 35 percent of the computation required by other methods. During these tests, Murakkab consumed about 27 percent as much energy as other methods. The system operated at less than 25 percent of the cost of other methods in these evaluations. In one specific instance, Murakkab reduced the energy consumption of an agentic workflow by more than an order of magnitude, resulting in a 2 percent drop in accuracy. Murakkab identified a configuration for a video frame selection model that optimized performance for a video Q&A task.

Murakkab identifies which workflow components should run sequentially and which should run in parallel to improve performance. The paper on Murakkab will be presented at the USENIX Symposium on Operating Systems Design and Implementation. Adam Belay, an associate professor of EECS at MIT and a member of the MIT Computer Science and Artificial Intelligence Laboratory, is a co-author of the paper, and Ricardo Bianchini, technical fellow and corporate vice president at Microsoft Azure, is the senior author.

Why It Matters

The development of Murakkab addresses concerns regarding the resource efficiency of AI-powered agentic workflows, which are increasingly central to cloud computing operations. The system aims to optimize resource allocation, energy consumption, and computational cost. By improving efficiency, Murakkab could reduce operational expenses for cloud providers and lower the environmental impact associated with AI processing.