CloudWeaver: A systems foundation for safe agentic cloud management

A team of researchers (Minghao Li, Ziqian Liu, Ziyu Mao, Daqian Ding, Yu Kang, Qingwei Lin, Tianyin Xu, Yiming Qiu) from Microsoft and the University of Illinois Urbana-Champaign have published a paper advocating for a systems foundation to enable safe, effective agentic cloud management. They argue that current autonomous management agents lack a fundamental substrate to handle multi-agent coordination, conflict resolution, and accountability—critical for production environments. To address this, they introduce CloudWeaver, a management substrate that works across existing cloud user interfaces and future agent-native interfaces.

CloudWeaver introduces two key mechanisms: (1) scoping agent sessions to local views of cloud resources, preventing agents from seeing or affecting unrelated resources, and (2) a coordination layer that manages concurrent operations on shared cloud resources, offering strong safety guarantees and attributable feedback when conflicting intents arise. The system was validated on a representative Azure API workload, demonstrating that it can preserve concurrency for independent operations while safely serializing conflicting ones. This work lays the groundwork for a new class of robust, multi-agent cloud management systems.