Get Started with AI Agents on Databricks

Unity Gateway extends Unity Catalog governance to AI traffic. Unity Catalog already governs your models, MCP services, functions, and connections as securable objects; Unity Gateway adds a central control plane over the traffic to them, enforcing routing, rate limits, budgets, usage tracking, and **service policies** (guardrails) on every request and response.

This lab introduces the fundamentals through a short lecture and then puts them into practice with a hands-on demo. In the demo, you govern a real agentic application, one that generates Databricks-native diagrams as code, by routing it through Unity Gateway and progressively hardening its governance: creating a model service, granting permissions, adding inference tables, rate limits, a hallucination guardrail policy, traffic splitting across Claude models, and fallback routing. Finally, you explore the `system.ai` and `system.ai_gateway` schemas to see how telemetry and observability complement payload-level inference logging.

By the end of the lab, you will understand how asset, traffic, and behavior governance combine into a single control plane for AI, and how to apply those controls to a production-style generative AI application on Databricks.

Note: For SCORM lecture files, please ensure that you close the SCORM window after completing the content. Do not click the ‘Next Lesson’ button, as doing so may prevent the SCORM module from being marked as complete.

本课程讲授如何在 Databricks Data Intelligence Platform 上使用纯 SQL 构建生产就绪的 ETL 管道。学员将学习以下内容：使用 Auto Loader 的流式处理表（用于增量摄取）、具有增量 刷新 功能的物化视图（用于白银到黄金的转换）、AUTO CDC（FLOW AUTO CDC）（用于声明式 SCD Type 1 和 Type 2 维度管理），以及 Apache Spark Declarative Pipelines（SDP）——将上述对象组合成具有数据质量期望、PRIVATE 表和统一监控功能的生产多表管道。本课程以一个真实的零售数据集为例，贯穿金银铜架构（铜 → 白银 → 黄金）的完整流程。使用 Lakeflow Jobs 进行编排作为附加模块提供。

注意：对于 SCORM 课程文件，请在完成内容后关闭 SCORM 窗口。请勿点击“下一课”按钮，否则可能导致 SCORM 模块无法标记为已完成。

AI 已经改变了开发者被要求构建的内容：智能体、副驾驶及智能工具正从实验阶段迈入生产阶段，工作重心也从编写代码转向协调模型、数据和服务。在本课程中，您将了解在数据所在之处构建智能体应用程序如何消除拼接碎片化技术栈所带来的基础设施开销。通过一个以账户简报为主题的引导式演示，您将看到 Databricks Apps、Lakebase 和 Agent Bricks（Genie Spaces、Knowledge Assistants 和 Supervisor Agents）如何在一个统一的、受治理的 serverless 基础之上协同工作，从而交付一个生产就绪的智能体应用程序。