AWS, Stardog offer semantic layer to ground AI agents in business data

Enterprise analytics has pursued the same two-decade goal: shrink the time between a business question and a trustworthy answer. The body traces this evolution from scheduled reports through dashboards, then self-service BI, to agentic analytics—where autonomous agents reason over live data on demand rather than visualizing pre-built datasets. Foundation models available on Amazon Bedrock have already cleared the first hurdle: they can plan, write queries, and iterate. The hard part sits underneath: the data itself.

The core challenge is data fragmentation. Enterprise records live across purpose-built systems—operational data in Aurora and RDS, analytics history in Redshift, unstructured content in S3. Each system defines the same business concepts differently. The "customer" in a CRM is not the same record as the "customer" in billing; "revenue" calculated by one regional team differs from another's. An AI agent given direct access to this fragmented landscape will write technically valid SQL that produces wrong, conflicting, or unexplainable answers, eroding confidence the moment two agents return two different numbers for the same question.

A semantic layer addresses this by capturing business context once as an ontology—a unified model of concepts, relationships, attributes, and rules—and mapping those concepts to rows in each live source. When implemented as Stardog does it (with ontology, stable identifiers for every entity, and rules that derive new facts and validate data), the result is a knowledge graph where entities are connected as business relationships rather than isolated rows. The agent queries this layer; the layer translates each query into SQL for the underlying systems at runtime. Data stays where it is; meaning is reused. The body emphasizes that a semantic layer does not replace RAG but complements it, with most production systems needing both accessible through the same agent.