Amazon Web Services introduced AWS Context in June 2026, a forthcoming service designed to map enterprise data relationships into a knowledge graph. This move addresses a critical gap for AI agents: the ability to understand how information connects across disparate systems while respecting organizational governance. By focusing on governed context retrieval at runtime, AWS is shifting the paradigm from raw data access to structured, relationship-aware intelligence for autonomous workflows.

Beyond Traditional RAG Limitations

Standard retrieval-augmented generation (RAG) often fails when business logic spans multiple records. For instance, determining refund eligibility requires linking customer profiles, order statuses, shipment delays, and specific policy rules. AWS Context differentiates itself by prioritizing relationship discovery and curated business knowledge. Data stewards can review and attach trusted definitions, ensuring that agents operate on verified connections rather than inferred assumptions. The service also supports interoperability by publishing metadata in Apache Iceberg format within Amazon S3 Tables, allowing compatible tools to leverage the same contextual layer.

Architecture for Controlled Autonomy

The proposed architecture places AWS Context between the AI agent and the underlying data sources. When a user asks a complex question, the agent identifies required context and retrieves relevant relationships through configured interfaces. Crucially, access controls are enforced at the identity level, preventing agents from retrieving restricted records simply because they are linked to accessible entities. This design ensures that an agent might know an order is delayed but cannot necessarily see the underlying financial audit trail unless the requesting user has permission. The system is built to support human-in-the-loop workflows, requiring approval before executing consequential actions like issuing refunds.

Practical Implementation and Governance

Developers can experiment with the open-source Context Ontology Accelerator, which uses components like Amazon Neptune, Amazon OpenSearch Serverless, and Amazon Bedrock to model business entities. This accelerator offers a path to prototype knowledge graphs before the managed service becomes widely available. The source material emphasizes that a knowledge graph alone does not guarantee accuracy; teams must rigorously evaluate data quality, relationship verification, and freshness checks. Traceability remains paramount, with audit logging recording every retrieval and execution attempt to maintain accountability in enterprise environments.

Key Takeaways

  • AWS Context is a forthcoming service announced in June 2026, not yet generally available for all regions.
  • The service focuses on governed context retrieval, linking data assets with business entities via knowledge graphs.
  • Interoperability is supported through Apache Iceberg metadata in Amazon S3 Tables.
  • Human approval workflows are central to the design, preventing unauthorized execution of agent decisions.
  • The open-source Context Ontology Accelerator allows early experimentation with Neptune and Bedrock.

The Bottom Line

AWS is finally acknowledging that agents without context are just expensive chatbots. By baking governance into the knowledge graph, they are solving the trust deficit that keeps enterprise AI in pilot purgatory.