Infrastructure

AWS Wants to Own the Context Layer — and Let Agents Do the Maintenance

Amazon's new AWS Context service builds a self-learning knowledge graph from enterprise data, positioning AWS as the default infrastructure for AI agent memory. The competitive stakes are higher than the press release admits.

Written by Julian Park · Bureau Tech · Jun 19, 2026

The Surprising Part Isn't That AWS Showed Up — It's the Architectural Bet

Amazon didn't just enter the context layer market this week. It entered with a specific and contestable premise: that the knowledge graph connecting enterprise data to AI agents should learn from agent behavior automatically, rather than requiring data teams to maintain it by hand.

That's a meaningful architectural claim, not a feature list. And it's worth examining what AWS is actually betting on — and who benefits if the bet pays off.

What the Context Layer Actually Is

The context layer — the connective tissue between enterprise data stores and AI agents — has no standard service. Until now, building it has been bespoke work: mapping which tables exist, what columns mean, how data sources relate, and which sources are authoritative. Data teams rebuild these graphs manually every time the underlying data changes.

AWS Context, announced Wednesday at AWS Summit NYC, is designed to eliminate that maintenance burden. The service automatically builds a knowledge graph from existing data, infers relationships across datasets and business rules, and makes that context available to agents at runtime. Crucially, it refines itself over time based on which sources agents actually use and which produce correct results.

"Your agents now get smarter without you having to rebuild anything from scratch," said Swami Sivasubramanian, VP of Agentic AI at AWS, during the keynote.

Three Layers, One Stack

AWS isn't pitching a single product — it's pitching a stack:

- **AWS Context** synthesizes the knowledge graph agents query at runtime, combining semantic search with graph-level reasoning across structured and unstructured sources. - **Amazon S3 Annotations** (now generally available) lets users attach business context directly to individual S3 objects at the storage layer. - **AWS Glue Data Catalog skill assets** (in preview) attach domain knowledge — runbooks, query patterns, usage rules — at the catalog layer.

Each layer feeds the next. The architectural logic is coherent: annotate at storage, enrich at the catalog, synthesize at the graph.

Data stewards manage the graph through the AWS Management Console, reviewing inferred relationships and promoting them to production. Every query inherits the calling user's IAM and Lake Formation permissions, making agent data access auditable by identity — through controls enterprises already rely on.

Metadata is published in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine. Third-party catalog connections are supported.

The Competitive Picture

The context layer is now a genuinely contested category. Snowflake announced Horizon Context and Cortex Sense earlier this month. Microsoft is providing context through Fabric IQ's semantic ontology. Redis has a context platform optimized for retrieval. Pinecone's Nexus compiles enterprise data into task-specific artifacts before agents ever query.

AWS's structural argument is the one it always makes: if you're already running S3, Glue, and Lake Formation, AWS Context extends your existing identity model with no data movement required. Zero-integration friction is the pitch — not just cost consolidation.

That argument is strongest for enterprises already committed to AWS infrastructure. For organizations running hybrid or multi-cloud environments, the calculus is less obvious.

The Open Question

Holger Mueller, VP and Principal Analyst at Constellation Research, told VentureBeat that every agentic platform vendor needs a context capability — and AWS is no exception. But he flagged the real risk plainly: "The concern — as with all context offerings — is going to be performance, especially for transactional data."

A self-learning graph that's slow to reflect writes isn't just a performance problem — it's a correctness problem. Agents acting on stale context make bad decisions. That's the test AWS Context will face in production, and press release language won't settle it.

Key takeaways

AWS Context builds and maintains a knowledge graph automatically, inferring relationships across datasets, business rules, and domain knowledge — and refining itself through agent usage over time.
The launch bundles three services: AWS Context (the graph), Amazon S3 Annotations (business context at the storage layer), and AWS Glue Data Catalog skill assets (domain knowledge at the catalog layer).
All metadata is published in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine — no proprietary lock-in on the query side, at least on paper.
Competitors already in the space include Snowflake (Horizon Context, Cortex Sense), Microsoft (Fabric IQ), Redis, and Pinecone — making this a crowded architectural category with no settled standard.
Analyst Holger Mueller flagged performance on transactional data as the open question; self-learning graphs that lag on writes are a real operational risk for enterprises.

FAQ

What is the 'context layer' and why does it matter for AI agents?

The context layer is the infrastructure that connects AI agents to enterprise data stores — mapping what data exists, what it means, how sources relate, and which are authoritative. Without it, agents lack the structured understanding needed to query data accurately. Building and maintaining this layer has historically been bespoke, manual work with no standard service to automate it.

How does AWS Context differ from a traditional knowledge graph?

Traditional knowledge graphs require manual curation and periodic rebuilding as data changes. AWS Context is designed to infer relationships automatically from existing data and refine the graph over time based on how agents actually use it — which sources produce correct results, which parts get queried most. The goal is a graph that improves without requiring data teams to intervene.

Does AWS Context create vendor lock-in?

On the query side, AWS is using open standards: metadata is published in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine. Third-party catalog connections are also supported. However, the service is most frictionless for enterprises already running S3, Glue, and Lake Formation — which is precisely the lock-in dynamic AWS benefits from, even without proprietary APIs.

Who are AWS's main competitors in the context layer market?

Snowflake (Horizon Context and Cortex Sense), Microsoft (Fabric IQ), Redis (context platform optimized for retrieval), and Pinecone (Nexus, which pre-compiles enterprise data into task-specific artifacts) are all active in the space. The category has no settled standard, and multiple architectural approaches are competing simultaneously.

What is the biggest risk with a self-learning knowledge graph?

Performance on transactional data. A graph that learns from agent usage but lags in reflecting recent writes can serve stale context — meaning agents make decisions based on outdated information. Analyst Holger Mueller of Constellation Research specifically flagged this as the open question for AWS Context and context offerings broadly.

Citations

AWS enters the context layer race with a graph that learns from agents, not manual curationAWS Context automatically builds a knowledge graph from existing data, infers relationships across datasets and business rules, and refines itself through agent usage over time.
AWS enters the context layer race with a graph that learns from agents, not manual curationHolger Mueller, VP and Principal Analyst at Constellation Research, told VentureBeat that performance on transactional data is the key concern for AWS Context and context offerings broadly.
AWS enters the context layer race with a graph that learns from agents, not manual curationAll metadata from AWS Context is published in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, Spark, or any Iceberg-compatible engine, with no proprietary APIs.
AWS enters the context layer race with a graph that learns from agents, not manual curationCompetitors in the context layer space include Snowflake (Horizon Context, Cortex Sense), Microsoft (Fabric IQ), Redis, and Pinecone (Nexus).