Building a Coding Agent: Python/Django Edition
Build your own project bootstrapper: a coding agent similar to Lovable, but for Python/Django projects. Learn how to create a tool-using agent that can bootstrap complete applications from natural language instructions.
Timestamps
Click any timestamp to jump to that moment in the video
Overview of the workshop objectives and what we'll build
Introduction to the presenter's background in AI and coding agents
Understanding Lovable as inspiration for building our own coding agent
Outlining the approach to build a coding agent for Python/Django projects
Setting up OpenAI API keys and configuring budget limits
Setting up the development environment with GitHub Codespaces
Understanding the difference between chatbots and agents, and how tools enable agentic behavior
Crafting effective system prompts and user prompts for the coding agent
Preparing a Django project template as the base for the agent to work with
Implementing core tools: file reading, writing, and search capabilities
Building and testing the agent tools interactively in Jupyter notebooks
Creating the critical developer prompt that guides the agent's behavior
Executing the coding agent using ToyAIKit framework
Reviewing what we've built so far and taking a break
Introduction to OpenAI's Agents SDK for building production agents
Creating a simple joke agent to understand Agents SDK patterns
Migrating our Django coding agent to use OpenAI Agents SDK
Introduction to PydanticAI as a production-ready framework for agents
Rebuilding the coding agent using PydanticAI framework
Comparing models and switching to Claude for better code generation
Testing the agent by building a complete Anki cards application
Exploring reasoning models from Z.AI for advanced agent capabilities
Overview of the AI Engineering Buildcamp course and learning journey
Core Tools
What You'll Learn
- Understanding the difference between chatbots and agents
- Building a project bootstrapper from scratch for Python/Django projects
- Implementing agent tools (file operations, search)
- Crafting effective system and developer prompts
- Using ToyAIKit for interactive agent development
- Migrating to OpenAI Agents SDK
- Building production-ready agents with PydanticAI
- Comparing different LLM providers (OpenAI, Anthropic, Z.AI)
- Testing agents with real-world projects
- Understanding how project bootstrappers work internally
Expected Outcome
A fully functional project bootstrapper: a coding agent that can build Django applications based on user instructions, implemented using multiple frameworks (ToyAIKit, OpenAI Agents SDK, PydanticAI), with the ability to read/write files, search code, and generate complete applications from scratch