Survey of Governed Cloud Data and Agentic Pipeline Control

Cloud data platforms increasingly combine streaming ingestion, batch analytics, cross-cloud machine-learning workloads, semantic evidence stores, API gateways, and compliance-sensitive control planes. Agentic automation can connect these surfaces more quickly than manual operations, but it also creates new risks when a software agent can inspect data, propose infrastructure changes, and influence governed pipelines. This paper surveys governed cloud data and agentic pipeline control through a 2025-oriented taxonomy. We synthesize work on agentic cloud data pipelines, cross-cloud optimization, multi-cluster gateway governance, cloud-native infrastructure streams, hybrid semantic-relational retrieval, historical profile modeling, agreement-gated learning, and structured extraction. The survey identifies six recurring control dimensions: evidence grounding, policy expression, resource placement, confidence gating, bounded actuation, and audit recovery. An analytical coding of representative systems shows that pipeline governance is strongest when semantic evidence, structured constraints, and change authority are treated as separate layers. The resulting design guidance is conservative: agents should assemble evidence and propose changes, while policy compilers, verifiers, approval gates, and actuation ledgers preserve operational accountability.