Build an AI Agent From Scratch — Free 12-Step Course in 30 Lines of Python (Claude Code Internals)
Learn how AI coding agents like Claude Code work by building one yourself in 30 lines of Python. This free 12-step open-source course has surpassed 28K GitHub stars and takes just 5 minutes to set up.
Understand How AI Coding Agents Work — In Just 30 Lines of Python
Claude Code, Cursor, Kiro… AI tools that write code for you are everywhere. But have you ever wondered how they actually work under the hood?
shareAI-lab's open-source project learn-claude-code answers that question. With 28,000 GitHub stars and 5,000 forks, it has become a global hit among developers and AI beginners alike.
The Core Structure: One Loop
The inner workings of an AI coding agent are surprisingly simple — three steps repeating endlessly:
This loop continues until the AI decides "nothing more to do"
The entire structure takes just 30 lines of Python. "That's it?" is a common reaction. But stacking features on this simple foundation creates a fully functional AI coding agent.
12 Hands-On Sessions
The key value is that the project never shows the finished product all at once. Each session adds exactly one feature, letting you naturally understand why it's needed.
Sessions 1–3: Foundation
• Session 1 — Basic loop + terminal command execution ("Bash alone is enough")
• Session 2 — Add file reading, searching tools ("one tool = one handler")
• Session 3 — Task planning with to-do lists ("AI without a plan wanders")
Sessions 4–6: Getting Smarter
• Session 4 — Sub-agent delegation: split big tasks and hand them off
• Session 5 — Load knowledge on demand (no need to memorize everything upfront)
• Session 6 — Auto-summarize long conversations to free up memory
Sessions 7–9: Multi-Agent Collaboration
• Session 7 — Detect task dependencies and process them in order
• Session 8 — Run long tasks in the background (no idle waiting)
• Session 9 — Multiple AI agents collaborate as a team via messaging
Sessions 10–12: Autonomous Operation
• Session 10 — Establish team communication rules
• Session 11 — Agents autonomously pick up tasks
• Session 12 — Isolated workspaces for conflict-free parallel work
Can Non-Developers Follow Along?
Honestly, the project is best suited for those with basic Python knowledge. But every session includes documentation in English, Chinese, and Japanese, with everyday analogies for each concept.
For example, Session 4 explains sub-agents like this: "If a team lead does everything alone, they'll burn out. Split big tasks into small ones and delegate each to a fresh agent."
Even without coding experience, it helps you understand "so this is how the AI coding tool I use works under the hood." If Python is new to you, our free learning guide is a great place to start.
The same team's Kode Agent (5K stars) — learn-claude-code concepts turned into a real product
5 Minutes From Install to Running
Open a terminal and enter these commands:
# 1. Download the project
git clone https://github.com/shareAI-lab/learn-claude-code
cd learn-claude-code
# 2. Install dependencies
pip install -r requirements.txt
# 3. Set up your AI API key (Anthropic key required)
cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY
# 4. Run the first session!
python agents/s01_agent_loop.pyA web interface is also available:
cd web && npm install && npm run dev
# Open localhost:3000 in your browserUnderstanding the Internals Transforms How You Use AI Tools
As the AI coding tool market explodes, the natural question follows: "So how does this actually work?" Using tools like Claude Code daily without understanding the internals limits what you can do with them.
This project bridges that gap. Understanding AI agent architecture means:
- Better prompts — knowing the internal steps helps you give more effective instructions
- Smarter tool selection — you can compare tools based on their underlying architecture
- Build your own — create custom agents tailored to your company or personal projects
With 872 new stars per day, growth this fast for an educational project is rare. If you're interested in AI coding, this is well worth exploring.
To learn more about AI and automation, visit our free learning guide.
Related Content — Get Started With AI | Free Learning Guide
Sources
Stay updated on AI news
Simple explanations of the latest AI developments