Build with AI.
Pick a project. Finish it in a weekend. Then pick a bigger one. This is where the real learning happens.
Why "make something" matters
You can read about AI forever and still not be good at it. The only way to actually develop the partner skill is to use it to build things you couldn't have built without it. Here are five starter projects, ordered roughly easiest → hardest.
Project 1 · Your personal tutor
Goal: Build a custom study tutor for a subject you're learning. Use AI to quiz you, explain weak spots, and adapt to your level.
Why it's a great starter: Pure prompting, no code, immediate payoff. Most people who try this stop using textbooks.
Project 2 · A useful tool for your own life
Goal: Pick a small annoying problem in your life, like a packing checklist, a recipe converter, or a study schedule generator, and build a working tool for it with AI.
Why it matters: Forces you to translate a vague human need into precise instructions. That translation skill is the AI partner skill.
Project 3 · A 30-minute deep-dive
Goal: Pick a complex topic you know nothing about. In 30 minutes, use AI to take you from zero to "I can explain this to a friend." Then explain it to a friend.
Why it matters: Practices the "just-in-time learning" move that AI uniquely enables.
Project 4 · Stress-test the AI
Goal: Pick a topic you already know well (a sport, a book, a hobby). Find where the AI gets it wrong. Document the failure modes.
Why it matters: Builds the most important defensive skill: knowing when to trust AI and when to verify.
Project 5 · Build your own (small) language model
Goal: Fine-tune an open-source language model on a laptop. Yes, really.
This sounds impossible. It's not. The author of this site did it on a MacBook Air in 4 minutes for $0. The trick is simple: don't train from scratch. Fine-tune an existing model from Meta or Hugging Face on a small dataset.
What you'll need: A Mac with Apple Silicon (or any computer with 16GB+ RAM), Apple MLX or Hugging Face TRL, a Hugging Face dataset, and an evening.
Why this matters: Once you've fine-tuned a model with your own hands, you'll never see AI as magic again. You'll understand the machine. That changes how you talk to it.
→ See Aarav's project for an example
What's next
v2 of this chapter will have full step-by-step walkthroughs for each project, plus sample prompts, plus a place to share what you built. If you finish any of these, send your work in: the gallery is open.