In this session, we cut through the hype and walk you through the architecture, design patterns, and governance guardrails needed to put Data Agents into production. You'll leave with actionable techniques for grounding agents on governed semantic models, enforcing security boundaries, and keeping answers accurate and explainable at scale.

Many organizations are moving beyond proof-of-concept and asking the harder questions: how do we trust the answers? how do we control costs? how do we govern who sees what?
We demystify in this session how Fabric Data Agents process user questions, determine the most relevant data source: whether a Lakehouse, Warehouse, Power BI semantic model, KQL database, or Microsoft Graph and invoke the appropriate tool to generate, validate, and execute SQL, DAX, or KQL queries.

We will cover five concrete production patterns:

1. Semantic model quality as the deciding factor.
2. Security that travels with the data.
3. Multi-agent architecture patterns.
4. Observability and cost controls.
5. Rollout strategy.

By the end of this session, you will be able to:

- Explain how Fabric Data Agents resolve natural language questions into SQL, DAX, or KQL queries across Lakehouses, Warehouses, semantic models, and KQL databases and where that process breaks down in production.
- Assess the readiness of your semantic models for agent grounding, and apply a "prep for AI" checklist to close the most common gaps before deployment.
- Design a least-privilege security architecture for Data Agents using Entra ID credentials, Purview DLP policies, and OneLake row- and column-level security.
- Compose a multi-agent solution that connects a Data Agent as the conversational analytics layer with operations agents for real-time monitoring and automated response.
- Define a phased rollout strategy: from single-domain pilot to org-wide deployment, with the governance controls, cost guardrails, and observability mechanisms needed to sustain trust at scale.