AI School for Developers

From “I know ChatGPT” to “I govern what my AI sees.”

Follow these tracks to learn how AI context works, why Context Governance matters, and how to use ContextDigger as the discipline layer under your existing AI tools.

Track 1 - AI Basics for Developers

Start here if you’ve used ChatGPT, Claude, Gemini, Copilot, or Cursor but don’t yet have a strong mental model of tokens, context windows, and attention.

AI Coding Basics

How AI assistants actually work: tokens, attention, context windows, and why “more context” isn’t always better.

Scroll to the deep dive →

Context Glossary

Industry terminology for Context Aperture, Attention Budget, Focus Discipline, Discovery Tax, and more.

Scan the glossary →

Track 2 - Context Governance 101

Once you understand context at the model level, learn how to deliberately govern what the model is allowed to see in your repo.

Core Concepts

Deep dive into Context Aperture, Attention Budgets, and what happens when Aperture is exceeded.

Read the concepts →

Governance in Practice

See how ContextDigger enforces file/line budgets and suggests governed sub-areas instead of silently overloading context.

See governance docs →

Track 3 - Hands-on Labs

Put Context Governance into practice with concrete scenarios. Use the CLI to build governed bundles, then compare AI behavior with and without them in tools like ChatGPT, Claude, Gemini, Copilot, Cursor, Aider, Cody, or Codeium.

Lab: Checkout API Rate Limiting

Add rate limiting to a checkout API in a repo with multiple checkout flows. Compare: ungoverned AI vs governed bundles vs future MCP integration.

Run the lab →

More Real-World Use Cases

Explore additional scenarios: onboarding to a new repo, legacy bug fixes, cross cutting refactors, and more, each powered by governed context bundles.

Browse use cases →

Track 4 - Using ContextDigger with Your AI Tools

Learn how to go from a repo with no governance to a setup where every assistant you use reads governed context bundles, whether you prefer chat (ChatGPT, Claude, Gemini), IDE assistants (Copilot, Cursor, Windsurf), or terminal tools (Aider).

Install & Initialize

Install the CLI, run contextdigger init, and generate your first governed context bundle with contextdigger dig.

Follow getting started →

Step-by-step Tutorial

Walk through a guided workflow that combines ContextDigger with your AI assistant to solve a real task.

Open the tutorial →

Pick a Track and Start Governing Context

If you only do one thing today, complete the AI Basics track and the Checkout API lab. You’ll never look at “just give it the whole repo” the same way again.

Start with AI Basics → Jump into the Checkout Lab →

Deep Dive - AI Coding Basics

This section gives you a concrete mental model of what AI coding assistants do with your code: how they tokenize, where attention goes, and why context windows and governance matter.

What AI "Sees" When You Send Code

Key Insight: AI doesn't "read" code like you do. It converts everything to numbers (tokens), processes them with math (attention), and predicts what comes next (generation).

Step 1: Tokenization

When you give an AI assistant a Python file, it gets broken into tokens (roughly 4 characters each):

Your Python code:
def calculate_total(items):
return sum(item.price for item in items)
How AI sees it (tokens):
["def", " calculate", "_total", "(", "items", "):", "\n", " return", " sum", "(", "item", ".", "price", " for", " item", " in", " items", ")"]
→ 18 tokens
Why Tokens Matter
  • • Cost: AI tools bill per token
  • • Limits: Context windows are token-based (8K, 100K, 200K)
  • • Speed: More tokens = slower processing
Token Examples
  • • "function" = 1 token
  • • "calculateTotal" = 3 tokens (calculate, Total)
  • • " " (4 spaces) = 1 token
  • • 1 line of code is roughly 15 to 30 tokens

Step 2: Attention Mechanism

Once tokenized, the AI uses attention to understand relationships between tokens:

How Attention Works

Imagine the AI is reading: user.email = validate_email(input_email)

The AI looks at each token and asks: “Which other tokens are related to this one?”

  • → email pays attention to validate_email (same concept)
  • → user pays attention to where user is created
  • → input_email pays attention to earlier validation logic

Once you see AI through this lens of tokens, attention, and bounded windows, it becomes obvious why governed context matters so much. The rest of ContextDigger (aperture, budgets, refusal, contracts, provenance) is built on top of this mental model.

Want the formal definitions? Read the related sections in Core Concepts: Context Aperture, Attention Budget, and Focus & Refusal.