November 2026 Summaries

AI does not eliminate the need for data governance; instead, it increases its importance because models and agents can amplify inconsistent data, compliance gaps, and weak access controls into confident but unreliable outputs. As AI agents replace dashboards as major consumers of enterprise data, governance must apply automatically at query time across cloud, on-premises, and partner systems, rather than relying on centralized storage, periodic reviews, or manual approval tickets. Key elements of AI-ready governance include federated policy enforcement, fine-grained attribute-based access controls, curated data products with ownership and lineage, and comprehensive auditing of agent activity. Without these controls, organizations risk shadow AI, exposed sensitive data in vector stores, conflicting business definitions, and regulatory violations. The piece argues that combining broad federated access with portable, automated governance helps enterprises move AI initiatives from experimentation into production more reliably.

AI agents require access to both enterprise data and its business context to produce reliable results, but sensitive on-premises information may be restricted by privacy, regulatory, security, and data-sovereignty requirements. The piece argues that organizations can address this challenge through hybrid architectures that either securely connect cloud-based agents to on-premises data or run models and agents locally in air-gapped environments for highly sensitive workloads. It recommends data federation to let agents query distributed sources where data resides, supported by role-based access controls and improved distributed-query performance, rather than centralizing all information in a warehouse. It also emphasizes context engineering through governed, discoverable data products that package raw information into curated datasets for AI use. Starburst is presented as a platform for creating this unified context and governance layer across cloud and on-premises systems, including Apache Iceberg storage, MCP-based agent access, natural-language data querying, lineage, auditing, and access controls.

Adopting a data lakehouse does not necessarily require retiring a cloud data warehouse, as organizations can use both platforms for workloads where each is most suitable. Cloud warehouses remain effective for highly structured, stable data and high-concurrency reporting, but their costs can rise through idle compute, over-provisioning, egress fees, proprietary storage, duplicated data, and maintenance-heavy pipelines. A lakehouse using Apache Iceberg and distributed SQL can provide features such as schema enforcement, ACID transactions, schema evolution, and time travel on open object storage, making it a potential fit for large-scale, semi-structured, rapidly growing, or multi-team data workloads. The recommended approach is a gradual, workload-by-workload transition supported by federated query access, allowing teams to query warehouse and Iceberg data through one SQL layer, compare cost and performance in place, and retain warehouse workloads where its managed performance, regulatory requirements, or write patterns remain advantageous.