Build, Orchestrate and Improve Agents with Databricks

Thursday, September 24, 2026 | 9:00 AM-10:30 AM PT

Take a hands-on deep dive into building and shipping production-grade AI agents on Databricks. You'll deploy your own custom customer-support agent as a Databricks App — guided by an AI coding assistant, with no infrastructure to set up — then watch enterprise governance follow your own identity straight through the live app with on-behalf-of-user authentication and Unity Catalog masking.

From there, you'll do exactly what real agent teams do: break the agent on planted data-quality bugs, measure the damage with MLflow LLM-as-judge evaluations over real traces, fix it, and prove the improvement with numbers — not vibes. You'll even give your agent memory with Lakebase and query its own conversation history.

In 90 minutes you'll run the full agent hardening loop — build, govern, break, measure, and fix — and leave knowing how to take an agent from prompt to production on Databricks.

What You'll Learn:

• Deploy a custom AI agent as a Databricks App, guided by an AI coding assistant — no infrastructure setup required
• Govern it with on-behalf-of-user auth and Unity Catalog masking, so access controls follow each user's identity into the live app
• Give the agent tools over governed data (Unity Catalog functions + Vector Search) and conversation memory in Lakebase
• Surface real agent-quality failures by breaking the agent on planted data bugs
• Measure agent quality with MLflow LLM-as-judge evaluation over real traces
• Fix the agent with a prompt change and prove the improvement with numbers