Vibe coding and context engineering with ADK

Опубликовано 26 ноября 2025, 17:00

Go from an empty folder to a fully deployed AI agent using only natural language. This video introduces "vibe coding," a powerful technique to build applications from scratch using the Gemini CLI.

Follow along with Sita and Amit as they demonstrate how to scaffold a new AI agent with the Agent Development Kit (ADK) in plain English. You'll then learn "Context Engineering," the secret sauce for moving your agent from a prototype to production. See how to provide documentation, coding standards, and feature requirements to guide the AI in generating high-quality, maintainable code.

Finally, watch as they deploy the entire agent to Google Cloud Run with a single command.

Stick around for the "Eng Talk" segment, where Amit and Sita answer top community questions:
- Is RAG obsolete with million-token context windows?
- How do you manage context overload?
- What metrics should you use to track context retrieval?
- What are the best tips for token optimization?

Chapters:
0:00 - Introduction
0:34 - Core Concepts
3:39 - Setting up Gemini CLI in VS Code
5:05 - Loading ADK Documentation for the Agent
7:03 - Vibe coding an Expense Tracker from Scratch
11:36 - Refactoring Code into Tools and Models
16:10 - Debugging Python code with Gemini CLI
25:15 - Why You Need Context Engineering
26:36 - Creating Context Files: Gemini.md, PRD, & Summary
32:02 - Building Advanced Features with Context Engineering
35:52 - Preparing for Cloud Run Deployment (Docker & Server)
37:44 - One-Command Deploy to Google Cloud Run
39:30 - Recap
40:48 - Community questions
45:55 - Resources

Resources:
Learn more about the Agent Development Kit → google.github.io/adk-docs
Install the Gemini CLI → github.com/google-gemini/gemin...
Code from this tutorial → github.com/amitkmaraj/expense-...
A2A Protocol official site → a2a-protocol.org/latest

Subscribe to Google for Developers → goo.gle/developers

Products mentioned: Google ADK, Gemini CLI, Google AI