Genie Code: Building Data Products from A to Z

Date: August 6, 2026
Duration: ~120 minutes
Instructor: Matt Giglia (matthew.giglia@databricks.com)
Product Area: Genie

---

Overview

Genie Code is so much more than just a single coding agent like Claude or Codex. Furthermore, for many of our customers it may be the first and only tool they'll use to start vibe coding due to cost or regulation. Learn the right way to use Genie Code including using the appropriate Databricks Editor, when to use skills, sharing sessions, creating session summaries for long term project memory, MCP servers, providing business context and using the power of Unity Catalog to enhance the vibe coding experience.

We'll build a complete data product from start to finish using only Genie Code.

Three Takeaways

1. Genie Code is the best tool to use for building in Databricks — it's not just a coding agent, it's a platform-aware partner with deep Unity Catalog integration
2. Learn how to effectively use Genie Code so that you can better demo with it and teach it to your customers
3. Context is the most important part of Genie Code — use the power of Databricks to build tools that Genie Code itself can use

---

Agenda

#	Topic	Duration	Type
1	Resetting Genie Code (skills evaluation, MCP server setup)	10 min	Setup
2	Databricks Editors for Genie Code Context	10 min	Conceptual
3	Sharing Genie Code Sessions	5 min	Conceptual
4	Proper Declarative Automation Bundle Setup	15 min	Architecture
5	Architectural Plans, Project Memory & Session Summaries	10 min	Architecture
6	Vibe Session 1: Infrastructure for AI and Apps	30 min	Hands-on
7	Vibe Session 2: AI Search Endpoints	15 min	Hands-on
8	Introduction to Databricks AppKit	5 min	Conceptual
9	Vibe Session 3: Databricks Apps with Genie Code	20 min	Hands-on

---

The Use Case: WanderBricks Intelligence Platform

We'll build a complete data + AI + app stack for WanderBricks, a vacation rental platform. The source data lives in samples.wanderbricks (available on all Databricks workspaces — no setup required).

What we're building:

Session	Deliverable
Vibe Session 1	Schemas, Volumes, SDP Pipelines (bronze→silver→gold), Metric Views, Genie Agent, Lakebase Project, Feature Tables
Vibe Session 2	Vector Search index on property descriptions + reviews for semantic discovery
Vibe Session 3	AppKit application (property search, host dashboard, or support agent)

The Flywheel: Along the way, we'll build tools that Genie Code uses to build better things — a business definitions index, a support resolution search, and an auto-generated data dictionary.

---

Pre-requisites

Before the workshop, ensure you have:

• Dev catalog with USE CATALOG and CREATE SCHEMA permissions
• Serverless compute enabled on your workspace
• Git credentials configured in Databricks (Settings → Developer → Git credentials)
• Read access to the samples catalog
• Model serving permissions (for AI Search endpoint)
• Databricks Apps permissions (for Vibe Session 3)
• Lakebase access (for app state persistence)
• Vector Search endpoint access

---

Getting Started: Branch Guide

This repo uses long-lived lesson branches as starting points. Each branch represents a clean state for that section of the workshop.

For Students

1. Clone this repo to your Databricks workspace
2. Checkout the branch for your current lesson:

   git checkout lesson/01-foundations
   

3. Create your personal feature branch:

   git checkout -b <your-name>-genie-<session>
   

   Example: jane-genie-vibe-infra
4. Work with Genie Code from there!

Branch Map

Branch	Start Here For	What's Already Done
lesson/01-foundations	Topics 1–5 (setup, editors, sharing, bundles, project memory)	Documentation only
lesson/02-vibe-infra	Topic 6 (Vibe Session 1: Infrastructure)	Empty bundle scaffold + PROJECT_MEMORY.md
lesson/03-vibe-ai	Topic 7 (Vibe Session 2: AI Search)	Infra fully built + deployed
lesson/04-vibe-apps	Topics 8–9 (AppKit + Vibe Session 3: Apps)	AI search endpoint running
main	Reference only	Complete answer key — everything finished

Never work directly on main or the lesson/* branches. Always create a personal feature branch.

---

Repository Structure

genieCodeWorkshop/
├── README.md                          # This file
├── PROJECT_MEMORY.md                  # Architectural decisions & workshop plan
├── LICENSE                            # MIT
├── docs/
│   ├── conventions/                   # Best practices reference
│   │   └── genie-code-best-practices.md
│   ├── 01-foundations/                # Lesson 1 guides
│   ├── 02-vibe-infra/                 # Lesson 2 guides
│   ├── 03-vibe-ai/                    # Lesson 3 guides
│   ├── 04-vibe-apps/                  # Lesson 4 guides
│   └── reference/
│       └── wanderbricks-analytics-handbook.md
├── media/
│   ├── screenshots/
│   │   ├── setup/
│   │   ├── editors/
│   │   ├── bundles/
│   │   └── vibe-sessions/
│   └── diagrams/
├── workshop-infra/                    # Instructor pre-setup (deploy before class)
│   ├── databricks.yml
│   └── src/
└── wanderbricks-platform/             # Student bundle (grows across sessions)
    ├── databricks.yml
    ├── src/
    └── fixtures/
        └── sessions/

---

Reference Materials

• Best Practices & Conventions — comprehensive guide to all patterns taught in this workshop
• WanderBricks Analytics Handbook — business definitions and KPIs (also used as the "tool-building" demo)
• PROJECT_MEMORY.md — all architectural decisions for this workshop

Further Reading

• Why A Frontier Data Agent Outperforms General Coding Agents in Quality and Cost — Databricks AI Research (July 2026). Across 400+ real tasks, Genie Code was both the most accurate and cheapest agent tested, delivering correct answers at less than half the cost of general coding agents. Deep semantic understanding of enterprise context means Genie Code skips brute-force schema exploration that drives other agents' errors and high costs.

---

License

MIT — see LICENSE