04-build-coding-agent

Module 4 — Create a Coding Agent

▶️ Workshop Video
📹 Demo TODO App with Z-ai

---

Overview

• Build your own coding agent that can scaffold and extend projects.
• Use a Django template as the base project.
• Learn how agents act as project bootstrappers.
• Explore multiple agent orchestration frameworks.
• Outcome: a Django app created and modified by your AI agent.

---

Prerequisites

• Python 3.8+
• Jupyter Notebook
• OpenAI API key (required)
• Anthropic API key (optional)
• GitHub account (optional, for Codespaces)

---

Environment

• Run locally with Python + Jupyter, or in GitHub Codespaces.
• Configure API keys in environment variables or Codespaces secrets.
• Install required libraries:

   pip install jupyter django uv openai toyaikit

---

OpenAI API Recap

• Review of responses API for system + user prompts.
• Function calling: extend model with custom Python functions.
• Prepares for agent tool integration.

---

Django Template Project

• Use a prebuilt Django template repo or create one from scratch.
• Add Makefile for install/migrate/run commands.
• Create templates (base.html, home.html) with TailwindCSS + Font Awesome.

---

Building the Agent

• Define tools: read/write files, execute shell commands, search files, view file tree.
• Create a developer prompt describing the Django project.
• Run agent using ToyAIKit.

---

Alternative Frameworks

• OpenAI Agents SDK — define tools with @function_tool, run with OpenAIAgentsSDKRunner.
• PydanticAI — lightweight agent library with tool support, easy provider switching.
• Z.ai — OpenAI-compatible API with alternative LLMs.

---

Learning Outcomes

• Understand how coding agents function as intelligent project bootstrapper tools.
• Gain practical experience in setting up Django with AI assistance.
• Learn to integrate tools into agents across multiple frameworks.
• Build agents that can adapt to different LLM providers.