AWS Announces Knowledge Graph Service for AI Agents

Amazon's cloud service AWS announced three new services to build a "context layer" that enables AI agents to leverage enterprise data. The announcement was made at AWS Summit NYC held in New York, with the core service "AWS Context," the general availability of "Amazon S3 Annotations," and a preview of the Skill Asset feature in "AWS Glue Data Catalog" all released simultaneously.

A context layer is a mechanism positioned between the large volume of data an enterprise possesses and AI agents, enabling agents to understand "what they should reference." Until now, there was no standard service in this domain, requiring each enterprise to implement custom solutions. AWS adopted an approach that automatically learns from agent usage patterns to address this challenge.

AWS Context is a service that automatically builds knowledge graphs from existing enterprise data. A knowledge graph organizes the relationships between data points like a map, automatically inferring information such as what data exists in which tables and what relationships exist between data sources. Swami Sivasubramanian, Vice President of Agentic AI at AWS, explained, "As agents continue to use the service, they learn which data sources produce accurate results, and the knowledge graph itself improves automatically."

Data administrators can verify the inferred relationships through the AWS Management Console, assign business definitions and usage rules, and then reflect them in the production environment. Access permission management leverages AWS's existing IAM and Lake Formation mechanisms, enabling tracking of who accessed which data. Metadata is stored in Amazon S3 in Apache Iceberg format, allowing reference from standard engines like Athena and Redshift, with a design that avoids dependency on AWS-proprietary APIs.

Amazon S3 Annotations, announced alongside these services, is a service that allows business meaning and supplementary information to be directly attached to individual files in storage. This enables context information to be embedded at the time data is stored, leading to improved accuracy when AI agents reference it. The Skill Asset feature in AWS Glue Data Catalog is in preview stage, and details will be revealed in future announcements.

The development of context layers has become a competitive field with multiple vendors participating. AWS entered the market with a unique approach: "a graph that learns automatically from agent usage without manual data reorganization." As the adoption of agentic AI expands, how much of the cost and effort required for enterprise data management can be automated will be a critical factor determining future adoption.