By Michelle Smith
To keep up with modern business needs, data strategy is shifting toward autonomous, real-time systems that deliver intelligence at decision time. Agentic AI is moving modern data teams beyond simply explaining what happened: teams now automate workflows that analyze what is happening, anticipate next steps, and recommend or take action. Below I outline the key trends shaping this era, from data agents and semantic layers to hybrid architectures and next-generation governance.
Putting data agents to work
Data agents are AI-powered software agents that access governed enterprise data and tools to answer questions and perform defined tasks. Rather than navigating reports and filters, a user can ask natural-language questions—such as “Why did sales decline last quarter?”—and receive a direct analysis.
- Analytical and organizational agents help users find trusted information by connecting to organizational data, answering natural-language queries, analyzing patterns, and surfacing relevant insights without manual navigation of databases, dashboards, or reports.
- Data engineering and governance agents operate behind the scenes to prepare, integrate, monitor, and manage the data that powers those insights. Agentic data engineering can generate transformations, map schemas, document datasets, monitor freshness, and suggest fixes.
However, data agents are only as good as the data they consume. Reliable insights and predictions require high-quality, well-governed data and sufficient context—making metadata increasingly important.
Metadata quality as the new data quality
Metadata sits at the center of meaning, trust and discoverability: it explains what a field means, where it comes from, who owns it, its lineage, freshness and sensitivity. Good metadata combines clear names and descriptions, shared business definitions, sources and ownership, relationships, and lineage.
For example, if an agent sees an ARR field showing $5.2 million, that number alone does not tell the agent how ARR is defined, what is included in the calculation, which system produced it, or how current it is. Metadata supplies that context so the agent can interpret the metric correctly and explain the source behind the answer.
Business context as the AI interface: semantic models, ontologies and knowledge graphs
Granting an agent access to data does not mean it understands the business. Two complementary layers provide the necessary context:
- Semantic models deliver analytical meaning: approved metrics, dimensions, calculations, hierarchies and relationships. If a sales leader asks, “How did ARR change in EMEA last quarter?”, the semantic model provides the approved ARR calculation, governed EMEA hierarchy and the company fiscal calendar.
- Ontologies and knowledge graphs supply entity meaning: they help an agent understand how customers, contracts, products, employees and organizations relate across systems. The same customer may appear under different identifiers in a CRM, billing platform and support system; a knowledge graph can establish that these records represent the same business entity and define relationships between entities.
Together they offer both analytical and organizational context: semantic models explain how the business measures something, while ontologies explain what things are and how they relate. This matters because an agent can generate valid SQL yet still return the wrong business answer if it selects the wrong metric, entity, time period or level of detail.
Protocol-first data access (MCP and similar approaches)
Organizations are starting to expose governed data and actions to AI agents through standardized interfaces, reducing the need for bespoke integrations. MCP (Model Context Protocol) is an emerging example that allows compatible AI clients to discover and invoke defined tools. A data platform could expose tools that let an agent find a certified dataset, retrieve a metric definition, inspect a schema, or run an approved query.
A protocol-first approach reduces duplicated integration work and creates explicit contracts around agent capabilities. It can also make authentication, governance and observability more consistent across integrations while simplifying the replacement or addition of AI clients and tools. Still, the protocol is just the connection layer: semantics, governance, permissions and security must be designed and enforced separately.
Identity passthrough as a make-or-break for enterprise AI on data
As agents access enterprise data, their permissions must reflect whom or what they act for. For user-initiated requests, agents can use delegated access so existing user permissions continue to apply. Autonomous agents may use their own identity scoped by least privilege.
In either case, agents should access only the data and actions required for their tasks. Identity-aware access reduces overexposure of sensitive data and establishes a basis for auditing and governance. When agents delegate work across tools, services or other agents, identity can drift or be lost, so it is critical to preserve the correct principal and permissions at every handoff.
Open table formats: from storage to catalogs
Open table formats such as Apache Iceberg, Delta Lake and Apache Hudi make it easier for multiple engines and tools to work with the same underlying data, reducing reliance on a single data platform. Organizations can store data once and make it available to multiple compatible analytics and AI tools instead of maintaining separate copies.
As data becomes more portable, differentiation shifts up the stack. The catalog increasingly becomes the control plane for discovering data, tracking lineage, applying governance and determining how AI systems can access it. As AI becomes a new consumer of enterprise data, the catalog grows in importance as a control point.
Building the foundation for intelligent decisions
These shifts point to a broader transformation: the future of data intelligence relies on creating a trusted, connected foundation that AI can understand, access and act upon—not on any single technology.
Success requires more than connecting agents to data. Organizations need trustworthy context, consistent business meaning and strong governance behind every answer. For BI teams, that means prioritizing certified semantic models, verified data and reusable metrics that both people and AI agents can trust.
The future of data intelligence is not just about faster answers; it is about building the trusted foundation that enables people and AI to make better decisions together.
