agents-101

An introduction to AI agents from first principles, using plain Python and an object-oriented design.

Open in Google Colab

What's inside

File	Description
notebook.ipynb	Main interactive tutorial - run in Google Colab or locally
TUTORIAL.md	Prose version of the tutorial for reference

notebook.ipynb

A step-by-step Jupyter notebook that builds a simple AI agent from scratch:

1. The Simplest Agent — a base Agent class that reads a file (perception) and writes a response (action).
2. Giving the Agent Tools — AgentV2 subclass that holds a dictionary of callable tools (get_current_time_tool, list_files_tool) and dispatches to them via a run() method.
3. Giving the Agent a Brain — AgentV3 subclass that adds a think() method simulating an LLM: it parses a natural language instruction and returns the right tool key.

The final cell shows exactly how to swap the mock think() for a real LLM (e.g. Google Gemini) by subclassing AgentV3 and overriding a single method.

Requirements

• Python 3.9+ (uses zoneinfo from the standard library)
• Jupyter or VS Code with the Jupyter extension

No external packages are required to run the tutorial. The Gemini upgrade snippet in the conclusion requires google-generativeai.