How AWS’s New Semantic Layer Turns Enterprise Data from Queries to Understanding in the Agent Era

Introduction

On July 31, AWS open‑sourced the Context Ontology Accelerator (COA). It connects structured and unstructured data, uses AI to draft enterprise Ontology drafts, and after expert review stores the results in a knowledge graph that agents can consume through MCP. AWS’s goal is to make agent decisions more accurate, consistent, explainable, and auditable.

Semantic Layer in the Agent Era

Traditional semantic layers focus on metrics, dimensions, relationships, and business definitions. With agents entering enterprise workflows, the system must also answer how entities relate, which rules must be satisfied, which data supports a conclusion, and who is authorized to execute actions. Thus the semantic layer is shifting from a BI‑centric business model to a runtime business‑context infrastructure required by agents.

Limitations of RAG

Retrieval‑Augmented Generation (RAG) excels at finding answers that exist in documents, but many enterprise decisions span structured data, entity relationships, and business rules. AWS Builder Center’s supply‑chain example shows a need to locate a substitute supplier that matches region, category, risk score, and does not trigger blocking rules—information spread across suppliers, distribution centers, routes, risk scores, and procurement policies. RAG alone cannot assemble the full reasoning chain.

COA Workflow

COA breaks its workflow into three steps: Scan, Model, Serve.

Scan : Connects databases, data catalogs, and documents to extract business metadata and entities.

Model : Uses AI to generate an Ontology draft, which is reviewed and approved by domain experts; the final Ontology and optional Governed Metrics are published to a knowledge graph.

Serve : Exposes the assets through MCP, REST, and SPARQL interfaces for agents.

Runtime Capabilities for Agents

COA makes semantic assets callable at runtime. Queries that can be answered by Governed Metrics follow a deterministic path; relational queries invoke the Ontology/Virtual Knowledge Graph; more complex scenarios fall back to the agentic layer. MCP standardises access so the same set of metrics, entity relationships, and business rules can be reused by multiple agents without rebuilding the semantics.

From Semantic Layer to Context Layer

In June, AWS announced a “shared, governed context layer” that automatically maps relationships, business rules, and domain knowledge into a knowledge graph for agents. The July 31 COA release clarifies that custom Ontology capabilities will become a native part of AWS Context. This evolution expands the semantic layer’s boundary from pure metrics to a full business‑context infrastructure that includes entities, constraints, provenance, and authorization.

References

AWS What’s New – COA release (2026‑07‑31)

AWS COA Documentation – Context Ontology Accelerator

AWS Builder Center – Beyond RAG (2026‑08‑15)

AWS Prescriptive Guidance – Semantic Layer for Agentic AI

AWS ML Blog – AWS Context (2026‑06‑17)

Apache Ossie (Incubating) – official description