AWS shows how data mesh can become the foundation for agentic AI in production

AWS has published a detailed technical guide to building a governed serverless data mesh, an architecture that gives agentic AI a reliable, secure and scalable foundation for production work.

Why agents need a data mesh

Agentic AI differs fundamentally from chatbots: it builds chains of actions on its own, calls tools and makes decisions without a human in the loop. To do this reliably, an agent must get up-to-date enterprise data in real time, and strictly within security policies.

The problem is that enterprise data is traditionally scattered across dozens of systems, teams and formats. Centralized data warehouses and data lakes solve part of the problem but create a bottleneck: data goes stale, schemas drift apart, and a single engineering team cannot keep up with requests from every department.

Data mesh takes a different approach: each domain (finance, marketing, logistics) owns its data and publishes it as a "product", with documentation, clear schemas and access contracts. An AI agent connects to such a product directly, with no intermediaries.

Architecture on AWS

AWS offers a serverless stack that makes it possible to implement a data mesh without managing servers:

• Amazon S3: decentralized storage; each domain keeps its data in separate buckets
• AWS Glue Data Catalog: a single catalog of schemas and metadata for all domains
• AWS Lake Formation: access control at the table, column and row level
• Amazon Athena: serverless SQL queries over data in S3 without additional ETL pipelines
• AWS IAM: a role-based model in which each agent gets the minimum permissions it needs

The key point: the agent does not need to know where the data physically lives. It addresses a logical product through the Glue catalog, not S3 buckets directly. This lets teams move and restructure data without changing the interface for agents.

In practice the cycle looks like this: the agent receives an IAM role, Lake Formation checks permissions on every request, Athena runs the SQL, and the result returns to the agent. Everything is serverless, and every action is auditable.

Security and control

Agentic AI is unpredictable: it can generate a query that overloads the system, or try to read data outside its own domain. AWS addresses this at several levels.

Lake Formation policies block any query outside the permitted scope, even if the agent technically has access to S3. CloudTrail keeps a full audit log of every action. Athena quotas cap the amount of data scanned per query, which protects against accidental or deliberate overload.

Another important detail is data semantics. The Glue Data Catalog stores not only schemas but also field descriptions: the agent understands the meaning of what it reads, rather than just receiving a set of bytes with no context. Schema versioning ensures that a structural change in one domain will not break the agents working with that product.

What this means

Data mesh is ceasing to be a theoretical concept and is becoming a practical requirement for anyone building production AI. Companies without governed data access will end up with agents that are either unreliable or unsafe, and more often both at once.

AWS offers a ready-made serverless stack that covers most tasks out of the box.