AtumAI: A Principled Framework for Agentic Generation of Datacenter Control-Plane Policies

A new framework called AtumAI aims to automate the design of datacenter control-plane policies using agentic AI, addressing the inefficiencies of manual policy design. The framework is introduced in a paper on arXiv (ID: 2608.02569) with the abstract highlighting three key shortcomings of off-the-shelf agentic AI: lack of formality, transferability, and systematic exploration. AtumAI makes the process formal by structuring the problem, transferable by leveraging knowledge across tasks, and systematic by exploring a broader design space. The paper discusses the challenges of designing control-plane policies due to the rapid growth of the hardware-software stack, vast interdependent design space, and the months-long prototyping time. AtumAI generates policies from a stated goal, promising to improve efficiency and reduce manual effort. The framework is presented as a principled solution, though specific implementation details and results are not provided in the abstract. The paper is authored by researchers and published on arXiv, a preprint server, indicating it is a recent contribution to the field of AI-driven systems management.