AWS Enters the Context Layer Race with a Graph That Learns from Agents
For enterprises building AI agents on top of complex data ecosystems, one of the most stubborn problems has nothing to do with the models themselves. It has to do with context. Getting an AI agent to understand which data sources exist, how they relate to each other, and which ones are authoritative has historically been bespoke, manual work — a slow, expensive process that breaks down the moment the data changes. Amazon Web Services is now making a direct play to solve that problem at scale.
At AWS Summit NYC, Amazon announced a three-part context intelligence stack designed to serve as the foundational context layer between enterprise data stores and AI agents. The centerpiece of the announcement is AWS Context, a new knowledge graph service built around a premise that sets it apart from competing offerings: the graph should learn automatically from how agents use it, eliminating the need for continuous human re-curation.
What Is the Context Layer and Why Does It Matter?
The context layer sits between raw enterprise data and the AI agents that act on it. Without a robust context layer, agents lack the structured understanding they need to reason reliably. They may query the wrong tables, misinterpret column names, ignore business rules, or fail to recognize that two datasets from different systems describe the same entity. The result is agents that produce incorrect outputs — not because the underlying model is flawed, but because the data environment around it is opaque.
Building and maintaining this layer has until now required significant engineering investment. Teams must manually map relationships between data sources, document domain knowledge, encode business rules, and continuously update those mappings as enterprise data evolves. There is no standard service that automates this process end to end. That gap is exactly what AWS is positioning itself to close.
AWS Context: A Knowledge Graph That Builds and Improves Itself
AWS Context is the anchor product in Amazon's new context intelligence stack. Rather than requiring data teams to manually construct and maintain a knowledge graph, AWS Context builds one automatically from an organization's existing data. The service maps what tables and data sources exist, what their columns mean semantically, how different sources relate to one another, and which sources should be treated as authoritative for a given domain.
According to Swami Sivasubramanian, Vice President of Agentic AI at AWS, the ambition behind the service goes further than initial graph construction. "This service automatically builds a knowledge graph from all your existing data," he said during the keynote. "This service infers relationships across your data sets, business rules, and domain knowledge, and makes all of it available to your agents and your organization at runtime."
Crucially, AWS Context is designed to improve over time without human intervention. The graph learns from agent interactions — tracking which data sources consistently produce correct results and which do not — and updates its internal model of the data environment accordingly. As Sivasubramanian put it, "Your agents now get smarter without you having to rebuild anything from scratch."
Under the hood, AWS Context combines semantic search with graph-level reasoning. This dual approach allows the system to surface not just syntactically similar data, but data that is conceptually related or governed by shared business logic. The result is a context layer that gets richer and more accurate the more it is used, compounding value over time rather than degrading as data drifts.
The Full Stack: Amazon S3 Annotations and AWS Glue Skill Assets
AWS Context does not stand alone. Amazon announced two companion products that extend the context intelligence stack at different layers of the data pipeline.
- Amazon S3 Annotations, now generally available, allows organizations to attach structured metadata directly to objects stored in S3. This enriches unstructured data at the storage layer, giving downstream services like AWS Context richer signals to work with when building and refining the knowledge graph.
- Skill assets in AWS Glue Data Catalog, currently in preview, allow teams to register reusable agent skills — discrete capabilities that agents can call — directly within the data catalog. By anchoring skills to the same catalog that governs data assets, AWS makes it easier for agents to discover and invoke the right capabilities in the right data context.
Together, the three products form a layered architecture: S3 Annotations enrich data at rest, AWS Glue Data Catalog organizes both data assets and agent skills, and AWS Context binds it all together into a dynamic, self-updating knowledge graph that agents can query at runtime.
A Contested Market with a Differentiated Bet
AWS is not the first vendor to recognize the context layer as a strategic architectural category. A growing number of startups and established data infrastructure players have released products targeting this space, each with their own approach to graph construction, maintenance, and agent integration. The competitive landscape is crowded, and differentiation increasingly comes down to the assumptions baked into the architecture.
Amazon's architectural bet is clear: manual curation is the wrong long-term foundation for enterprise knowledge graphs. As data volumes grow and agent deployments scale, human-maintained graphs will struggle to keep pace. By building a system that learns from agent behavior automatically, AWS is wagering that the organizations most likely to win with agentic AI will be those whose context layer improves in proportion to how much they use it.
That bet aligns with a broader pattern in AWS product strategy — taking capabilities that customers have historically built themselves and offering them as managed, continuously improving services. Whether the self-learning graph lives up to its promise at enterprise scale will depend on real-world deployments, but the architectural direction is a meaningful departure from the status quo in context layer tooling.
What This Means for Enterprise AI Teams
For data engineers, AI platform teams, and enterprise architects, the AWS announcement signals that the context layer is rapidly moving from a custom engineering problem to a managed infrastructure concern. Teams currently investing in manual graph construction and maintenance should watch closely how AWS Context performs in production, as it may significantly reduce the operational burden of keeping a knowledge graph current and accurate.
More broadly, the announcement reinforces a shift in where enterprise AI competition is playing out. Model performance is increasingly table stakes. The organizations that deploy AI agents most effectively will likely be those with the strongest data context infrastructure underneath — and that infrastructure layer is now a primary battleground for cloud providers, data platform vendors, and AI tooling startups alike. AWS has just made its opening move.