Everyone's talking about AI adoption. Almost nobody has the real numbers. Help us change that — and get the full report 👉 Engineers | Leaders

Software Developer

6 courses · ~12 hours· 2 certifications available

phase 1 — Understand the mechanism

Learn how AI coding tools actually work under the hood.

[ ] Understanding Tool Calling

This is how Cursor, Copilot, and every AI coding tool works under the hood.

4 sections · 10 lessons
  • [ ] Foundations (2 lessons)
  • [ ] How Models Decide (2 lessons)
  • [ ] Advanced Patterns (3 lessons)
  • [ ] Security and Production (3 lessons)
[→ start course]

phase 2 — Build your own agent

Build a working coding agent from scratch.

[ ] Building Your Own Coding Agent

You'll write ~200 lines of Python that do what Cursor does. Then you'll understand Cursor.

4 sections · 10 lessons
  • [ ] Getting Started (2 lessons)
  • [ ] The Conversation Loop (2 lessons)
  • [ ] Adding Tools (3 lessons)
  • [ ] Making It Real (3 lessons)

certification: Certified AI Agent Builder

[→ start course]

phase 3 — Master context

Learn what separates hobby agents from production agents.

[ ] Context Engineering

This is what separates hobby agents from production agents.

8 sections · 25 lessons
  • [ ] Tokens and Inference (2 lessons)
  • [ ] The Real Size of Your Context Window (4 lessons)
  • [ ] Anatomy of the Messages Array (5 lessons)
  • [ ] Dynamic Allocation: Tool Calling (2 lessons)
  • [ ] The Ralph Wiggum Loop (3 lessons)
  • [ ] Sub-Agents: Managed Runtimes for AI (3 lessons)
  • [ ] Message Passing: The Erlang OTP of AI (3 lessons)
  • [ ] Context Management Strategies (3 lessons)

certification: Certified Context Engineer

[→ start course]

phase 4 — Go deeper [optional]

Optional deep dive into transformer architecture.

[ ] Understanding Transformers [preview]

Not required, but understanding attention helps you write better prompts and debug model behavior.

2 sections · 4 lessons
  • [ ] Attention Mechanisms (3 lessons)
  • [ ] Embeddings (1 lesson)
[→ start course]
[ ] Understanding Micrograd

Build the autograd engine and neural network that power every ML framework — from Value objects to gradient descent.

5 sections · 13 lessons
  • [ ] The Big Picture (2 lessons)
  • [ ] The Value Class (3 lessons)
  • [ ] Building a Neural Network (3 lessons)
  • [ ] Training (3 lessons)
  • [ ] The Full Picture (2 lessons)
[→ start course]
[ ] Understanding Makemore

From bigram counting to WaveNet — build character-level language models that generate names, learning the training techniques that preceded transformers.

5 sections · 15 lessons
  • [ ] Bigrams (4 lessons)
  • [ ] The MLP Language Model (3 lessons)
  • [ ] Activations and Batch Normalization (3 lessons)
  • [ ] Becoming a Backprop Ninja (3 lessons)
  • [ ] Building a WaveNet (2 lessons)
[→ start course]

key terms

Tool Calling·Agent·Context Engineering·System Prompt·Finite-State Machine·Agent Harness

related reading

→ Pi: the architecture of an AI coding agent

→ Beads: the architecture of a graph issue tracker for AI agents