Claude Code Sub Agents Tutorial

Sub agents are amazing and you should learn how to use them.

Even if you don’t use Claude Code.

Why? Because — like many things from Anthropic — sub agents are an emerging pattern that you will see a lot of in the coming years.

Check out the video on YouTube

1:00 - Benefits of Claude Code Sub Agents

2:12 - Sub Agents Demo

2:56 - Creating Your First Sub Agent

6:31 - Building a Feature with Sub Agents

12:10 - The Context Problem with Sub Agents

13:27 - Manually Editing Sub Agents

14:46 - Test Runner + SWE Sub Agent Demo

17:27 - Running the Unit Converter MCP Feature

19:20 - Release Helper Sub Agent

22:52 - Releasing a Version Upgrade of my PyPI Package

Course Update: Added 200+ images

I’ve added ~250 images to my AI Engineer Roadmap course.

These figures compliment the various concept maps and code examples through out.

These took me a very very long time to create, and I’ll be increasing the price of the course later in August.

Here’s one particularly nice set of plots from the optimization lesson:

Topics from this week’s video

• Sub Agents in Claude Code

  • Introduction to Sub Agents and their purpose

    • Pre-configured AI personalities that can delegate tasks

    • Benefits include:

      • Context preservation via isolated contexts for each subagent

      • Reusability as shareable, text-based configurations

      • Flexible permissions for scoped access based on least privilege principle

• Demonstration Use Case: MCP Server Unit Converter

  • Creating Sub Agents

    • Product Manager Agent

      • Read-only access

      • Helps ideate and prioritize features

    • Software Engineer Agent

      • Full tool access

      • Builds out features like batch conversion

    • Test Runner Agent

      • Executes tests following project-specific instructions

      • Initial issues with using proper commands (e.g., uv run pytest)

• Testing the Sub Agent System

  • Persistent subagents stored in local project directory

  • Feature ideation and implementation process

  • Iterative refinement of subagent system prompts

• Troubleshooting & Improvement

  • Addressing test context hand-off issues between subagents

  • Improving test accuracy and reliability

  • Identifying inefficiencies in agent interactions

• Successful Sub Agent Collaboration

  • Achieved working batch conversion feature

• Creating a Release Manager Sub Agent

  • Automates versioning and changelog updates

  • Uses custom release checklist from the project

  • Demonstrates value in codifying organizational best practices

• Publishing the Package

  • Final prep steps: formatting, type-checking, and linting

  • GitHub Actions validate before release

  • GitHub Release triggers PyPI deployment v0.1.2 Unit Converter MCP

---

If you don't prioritize your life, someone else will.

Greg McKeown