P091: Multi-agent orchestration for data operations: Cross-cloud comparative study

Abstract

Multi-agent orchestration for data operations: Cross-cloud comparative study Author: Sonu Kumar Singh (Senior Consultant — Cloud & AI Solutions Architecture, Capgemini US LLC) Professional Credential: Member, IEEE (Membership # 102728576) | ORCID: 0009-0002-9180-4946 Abstract Enterprise generative AI becomes dependable only when retrieval, tools, identity, policy, provenance, evaluation, and human oversight are engineered as a system around the model. Against this backdrop, the paper examines multi-agent orchestration for data operations in Microsoft Azure, AWS, and Google Cloud. It asks a focused question: How should multi-agent orchestration for data operations be designed, governed, and empirically evaluated for Microsoft Azure, AWS, and Google Cloud? Here, the topic is treated as a set of engineering obligations, not a slogan. The important questions are what must remain correct under scale and failure, which controls must follow the workload across services, and which measurements would be needed to support a claim. Framing the topic this way prevents the analysis from reducing to a list of vendor capabilities. The main contribution is a structured way to move from architectural claims to evidence. Using conformance testing and evidence ledger, the paper links design choices with security controls, operating assumptions, economics, and reproducibility. It does not fill gaps with synthetic benchmark numbers; instead, it defines the experiment that would be required to turn a design hypothesis into a supported result. Architectural Research Scope Research Domain / Theme: Generative AI & Agentic Systems Architectural Scope: Cross-cloud comparative Core Research Question: How should multi-agent orchestration for data operations be designed, governed, and empirically evaluated for Microsoft Azure, AWS, and Google Cloud? Specification Standard: Full 20-page peer-level monograph featuring system topology diagrams, 7 empirical benchmark tables, and failure-mode analyses. Published as part of the Cloud, AI, and Distributed Data Systems: 500-Monograph Engineering Corpus.