New MIT development reduces energy use in AI systems
Researchers from MIT and Microsoft have developed a system called Murakkab to improve the speed and energy efficiency of agentic AI workflows.
Agentic workflows combine multiple AI models and external tools to complete complex, multi-step tasks, such as analysing video or generating code. MIT said these systems are becoming more important for cloud providers, but their fragmented design can waste computation, energy and money.
Murakkab allows developers to describe an AI application in high-level terms rather than manually specifying every model, tool, hardware choice and execution step. The system then identifies suitable models and tools, decides which components should run sequentially or in parallel, and selects hardware resources for cloud deployment.
The system can adjust configurations during execution based on user priorities such as accuracy, speed, latency and cost. It also gives cloud providers more visibility into workflows, allowing them to allocate computing resources more efficiently across multiple tasks.
In tests of video-question-answering and code-generation workflows, Murakkab met user requirements while using about 35% of the computational resources required by other methods. It also consumed about 27% as much energy and cost less than 25% as much as the comparison approaches.
In one case, the system reduced energy consumption by more than an order of magnitude with only about a 2% drop in accuracy. The researchers plan to expand Murakkab to more complex workflows and larger computing clusters.
Why does it matter?
Agentic AI systems are becoming more complex and resource-intensive, especially as cloud providers deploy workflows that combine many models, tools and hardware configurations. Murakkab points to a shift from optimising individual models to optimising the whole AI workflow and its cloud deployment. That matters because energy use, compute costs, and data centre capacity are becoming central constraints on AI growth.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!