AWS Context and Genie Ontology Raise the Stakes: Own Your Context Layer

In the space of two days this week, two of the most important platforms in enterprise data announced their own context layers for AI agents. Databricks unveiled Genie Ontology at its Data + AI Summit, and on June 17 AWS introduced AWS Context, a knowledge graph service meant to give every agent in an organization a shared, governed picture of its data. Our view: this is good news for enterprises, but only for those that treat their business context as an asset they own, not a feature they rent.

What is AWS Context?

According to the AWS Machine Learning Blog, AWS Context maps relationships across existing data into a knowledge graph and offers agentic search so agents can reach governed relationships, business rules and domain knowledge at runtime. It infers relationships automatically, then lets data stewards review them, promote them to production and add business definitions and usage rules. It watches which sources give correct results and which join paths agents use, ranks sources by actual usage, and shares what one agent learns with others. Queries inherit the calling user's IAM and Lake Formation permissions.

The service was listed as coming soon in AWS's roundup of Summit New York announcements, which also included a managed knowledge base for retrieval-augmented generation in Amazon Bedrock. Databricks described Genie Ontology as a permission-aware ontology that ranks knowledge snippets and combines them with Unity Catalog definitions and glossaries, Constellation Research reported on June 16.

Why does the context layer matter so much?

Because it is where an enterprise's meaning lives. Models are interchangeable and getting cheaper. The definitions of an active customer, a recognized sale, a high-risk claim or an approved supplier are not. They encode years of decisions, exceptions and regulatory interpretation. When an agent acts on data, that context decides whether it acts correctly.

AI Connect practitioners have been making this case all year. In a AI Connect expert panel on the context layer for enterprise AI in May, AtScale's Mark Palmer predicted that large platform providers would fight over who controls semantics. This week's launches are that fight arriving. Precisely's Chief Architect of SaaS, Jeremy Simson, put the stakes plainly on the same panel: "AI is just another user on your network who has access to your data and is going to make decisions for you."

What is the case for letting the platform own context?

It is a strong one, and enterprises should take it seriously. Building a governed knowledge graph from scratch is slow and expensive. A platform that infers relationships automatically, learns from agent behavior and enforces existing permissions removes much of that cost. AWS says the service builds on the knowledge graph technology behind Amazon Quick, which it says already handles millions of requests a day. Native integration with a company's catalog, lakehouse and identity system means fewer moving parts and fewer places for policy to drift.

For a company that runs almost all of its data on one cloud, a provider-managed context layer may be the fastest route to agents that work, and speed to production is worth a lot.

Where should enterprises draw the line?

The risk is that the most valuable layer in the AI stack becomes the hardest to move. Context that is inferred, ranked and refined inside a provider's service can quietly become proprietary, even when the underlying data sits in open formats. If the learning about which sources are trustworthy and which joins are correct lives only in a vendor's graph, switching platforms means losing institutional knowledge, not just migrating tables.

AWS appears to have anticipated this concern. Its blog says key metadata is published to Amazon S3 in Apache Iceberg format and can be queried by any Iceberg-compatible engine, and that the service is designed to connect to third-party catalogs. That is the right design, and buyers should hold every vendor to the same standard. We would set four conditions before committing to any platform's context layer:

• Exportable semantics. Business definitions, relationships and rules must be readable and portable in open formats, including the parts the system learns on its own.
• Human ownership of meaning. Automated inference should propose; named data stewards should approve. The approval record belongs to the enterprise.
• Auditability outside the agent. Snowflake's Sam Mittal, speaking in a AI Connect webinar on June 18 about agentic-ready data, said the record of who authorized an action and what data it used "lives in the platform architecture, not the agent prompt." Enterprises should be able to query that record independently.
• Multi-platform reach. Few large companies run on one platform. A context layer that covers only one vendor's data will give agents a partial and sometimes misleading picture.

None of this argues against using AWS Context, Genie Ontology or similar tools. It argues for using them deliberately. The work of curating meaning, which our coverage of always-on semantic governance describes, is the enterprise's job no matter which tool assists it.

The platforms have now told us what they think the next control point in enterprise AI is. They are right. That is exactly why enterprises, not platforms, should own it.

Frequently asked questions

What is AWS Context?

A coming-soon AWS service that builds a governed knowledge graph from existing data and offers agentic search so AI agents can use business relationships, rules and domain knowledge at runtime.

Can AWS Context metadata be used outside AWS services?

AWS says key metadata is published to Amazon S3 in Apache Iceberg format and can be queried by Athena, Redshift, Spark or any Iceberg-compatible engine.

Why does owning the context layer matter?

Business definitions and learned relationships encode years of decisions. If they live only in one vendor's service, switching platforms means losing that knowledge.

Opinion pieces reflect the views of The AI Enterprise Magazine Team.

Filed under: Context, Semantics & RAG, AI Governance & Trust, Agentic AI