Building an AI pilot on AWS is becoming easier and easier. Taking it into production with context, access control, traceability, and consistent results is much harder. At that stage, the challenge is no longer just the model or the prompt. It is the system’s ability to understand who is asking, which data sources it can use, which policies it must enforce, and how to respond with confidence. That is where metadata becomes critical.

In this session, I will share a practical framework for understanding how metadata enables more trustworthy agents on AWS. We will explore how metadata about the user, the message, the agent, and the underlying data sources influences query interpretation, tool selection, permission enforcement, end-to-end traceability, and response quality.

From an architectural and applied perspective, I will connect capabilities such as Amazon Bedrock, IAM, metadata catalogs, and governance mechanisms to show how to design agents that do not just work in a demo, but can be sustained in production with greater control, scalability, and business value.

Target audience: cloud architects, data engineers, AI builders, governance leads, and technical leaders. Intermediate level.