AWS COA: Semantic Layer Evolves into Agent Runtime Context Infrastructure

Semantic Layer Evolves for Agent Era

Traditional semantic layers resolved metric inconsistencies across reports by centralizing definitions of metrics, dimensions, relationships, and business terms. However, agents require deeper business context: they must understand entity relationships, governance rules, and which metric version is authoritative.

RAG Alone Is Insufficient for Complex Business Decisions

RAG retrieves document fragments but cannot handle decisions spanning structured data, entity relationships, and business rules. AWS Builder Center's supply chain example illustrates this: finding an alternative supplier requires joining supplier, distribution center, route, risk score, and procurement rule data. Three semantic needs emerge: Governed Metrics for definitive KPIs, Semantic Models for stable definitions and relationships, and Ontologies for formalized entities, rules, and constraints.

COA Turns Semantic Assets into Agent-Callable Runtime Capabilities

COA operates in three phases: Scan (connects databases, catalogs, documents), Model (AI-assisted ontology drafting with human review, publishing to knowledge graph, governed metrics definition), and Serve (exposes capabilities via MCP, REST, SPARQL). It unifies deterministic metrics and entity/relationship/rule assets into a single service chain. A layered resolution strategy prioritizes governed metrics, then ontology/virtual knowledge graph, then agentic fallback. Cedar authorization and SQL firewall enforce access control.

MCP standardizes access so the same metrics, relationships, and rules serve multiple agents regardless of model or framework changes.

Toward a Shared, Governed Context Layer

AWS Context (announced June 2026) aims to auto-map enterprise data relationships into a knowledge graph for runtime agent use. COA's custom ontology capability will become a native managed feature of AWS Context. This extends the semantic layer beyond metrics and joins to include complex entity networks, rules, provenance, identity permissions, and traceable resolution. As model capabilities grow, governed semantic foundations become more critical to prevent automation of ungoverned guesses.

References

AWS What's New | COA Release (2026-07-31)

AWS COA Docs | 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 Documentation