Improving the speed and energy-efficiency of AI agents | MIT News

Source Domain: news.mit.edu

Researchers from MIT and Microsoft have developed an intelligent system called Murakkab to optimize agentic workflows, which are AI-powered systems that tackle complicated tasks by chaining multiple models and tools together.
This system enables developers to describe their intended application in plain language without needing to specify all technical details upfront. Murakkab automatically determines the best models, tools, hardware configurations, and computational resource allocations.
The automated optimization capabilities of Murakkab help in dynamically adjusting configurations based on user priorities, such as minimizing costs or maximizing speed, without compromising performance.
When tested on various workloads, Murakkab significantly reduced computational units, energy consumption, and costs while maintaining efficiency compared to traditional approaches.
The researchers aim to expand their system to more complex workflows and larger computing clusters to further reduce the energy and cost inefficiencies of agentic workflows on a broader scale.

You may have missed