AWS Context learns from agents automatically, aiming to replace manual graph curation

Amazon is entering the context layer race with a thesis that is both simple and kind of heretical in enterprise AI: stop rebuilding knowledge graphs by hand.

At AWS Summit NYC, Swami Sivasubramanian, vice president of Agentic AI at AWS, described AWS Context as a new knowledge graph service that gets smarter through agent usage over time. The pitch is direct. “Your agents now get smarter without you having to rebuild anything from scratch,” Sivasubramanian said. AWS also claims the service automatically builds a knowledge graph from existing data, infers relationships across datasets, business rules, and domain knowledge, and makes all of that available to agents and the organization at runtime.

If you have ever watched an enterprise try to keep AI “grounded” in real business reality, you already know why this matters. Context layer work is often bespoke. Graphs do not maintain themselves. Someone has to interpret data, define relationships, and decide what is authoritative. AWS is trying to commercialize the part that typically breaks adoption: the never-ending upkeep. Instead of requiring human re-curation after each new deployment, AWS Context is positioned as self-learning. The knowledge graph improves itself over time, according to Sivasubramanian, as it learns which sources produce correct results and which parts get used.

Under the hood, AWS Context is built to map relationships across existing enterprise data automatically. That includes what tables exist, what columns mean, how sources relate, and which sources are authoritative. AWS says it combines semantic search with graph-level reasoning. The output is a graph that spans structured and unstructured knowledge, and then exposes that context so it can be used during agent runtime.

But AWS is not only selling a graph. It is selling a workflow for how enterprise teams manage it. Data stewards, AWS says, manage the graph through the AWS Management Console by reviewing inferred relationships, promoting them to production, and attaching business definitions and usage rules. That is a key nuance for buyers. Fully autonomous systems tend to freak out risk teams. AWS is describing a human-in-the-loop governance layer, even if the inference and learning are automatic.

For security and auditability, AWS adds a practical detail: every query inherits the calling user's IAM and Lake Formation permissions. In other words, agent data access is auditable by identity through controls enterprises already rely on. For regulated industries, that is not a “nice to have.” It is often the difference between an agent pilot that survives and one that gets locked down before it scales.

Then there is the packaging. AWS Context publishes metadata in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine. AWS is also claiming “no proprietary APIs,” and it supports third-party catalog connections so context from systems outside AWS can be pulled into the same graph. In enterprise terms, that translates to less lock-in anxiety for procurement and platform teams, and fewer architectural dead-ends for data engineers.

This is not a single-service story either. AWS is framing a stack of three products as a context intelligence stack for AI agents. The centerpiece is AWS Context. Amazon S3 Annotations is described as enabling users to attach rich business context at the storage layer, directly to individual S3 objects. AWS Glue Data Catalog skill assets, in preview, are for attaching domain knowledge at the catalog layer by linking runbooks, query patterns, and usage rules to data assets across the estate. AWS Context then synthesizes both into the knowledge graph agents query at runtime, combining semantic search with graph-level reasoning across structured and unstructured sources. Each layer feeds the next, according to AWS, which is a clever way to sell continuity across the stack.

The competitive pressure here is real, because “context layer” is quickly becoming the architectural category every AI stack claims it can solve. Snowflake announced its context approach earlier this month with Horizon Context and Cortex Sense services. Microsoft is providing context via its Fabric IQ platform that provides a semantic ontology for data. Redis has developed a context platform that optimizes data for retrieval. Vector database vendor Pinecone has its Nexus context offering that compiles enterprise data into task-specific artifacts before agents ever query them. AWS’s differentiated angle is that enterprises already running S3, Glue, and Lake Formation can extend an existing identity model, with no data movement required. AWS’s structural argument is zero-integration friction. Not just cost consolidation, but a lower-risk path to adopting agent tooling.

There is also a market reality hiding inside AWS’s claims. As Holger Mueller, VP and Principal analyst at Constellation Research, told VentureBeat, context offerings always face performance concerns, especially for transactional data. He said “we will see” performance issues for transactional workloads. That is the second-order risk boards should track: it is one thing to build an elegant graph. It is another to keep latency stable when agents start hammering the system in real time.

So the strategic stake is bigger than “another AWS service.” If AWS successfully reduces the operational burden of maintaining context, it could become the default foundation under agentic platforms across enterprise teams already standardized on AWS data services. And for peers, the lesson is clear: the race is not only about model intelligence. It is about who owns the durable, governable context that turns enterprise data into trustworthy agent behavior.