Student path¶
This path is for university students, bootcamp learners, and anyone who can program a little and has used ChatGPT but not a coding agent. It takes one semester with the course syllabus, or about 15 hours of reading plus labs if you're self-paced.
What you'll be able to do at the end¶
- Explain what a language model is and isn't, well enough to predict where it will fail.
- Use a coding agent to explore, plan, build, test, and review a small project, while understanding every line you submit.
- Tell a good model claim from marketing, and test a model on your own tasks.
- Recognize prompt-injection risks and limit what an agent can access or change.
- Use AI in a way that makes you a better programmer, not a dependent one.
How to read each module¶
Every module has three depth levels. On this path, read Foundations and In practice, and skim Going deeper unless a topic grabs you or the syllabus says to read it. Do the short "Try it" exercises to practice what you've read. Read every TL;DR and Bryan's take too.
The path¶
Read in this order. It's not quite numeric. You start using tools early, then come back for the theory once you have something to hang it on.
| Step | Module | Focus on | Lab |
|---|---|---|---|
| 1 | 00 Orientation | "What we mean by vibe coding," the six durable ideas, ground rules | |
| 2 | 01 What is an LLM? | Next-token prediction, tokens, "What an LLM is not," hallucination | Lab 1 |
| 3 | 06 Harnesses, tools, and MCP | What a harness is, the short history, "A sane baseline setup"; read module 11's student checklist before running an agent | Lab 2 |
| 4 | 07 Context engineering | Prompting basics, what fills the context window, context limits | |
| 5 | 08 Agentic workflows | The agent loop, explore → plan → implement → verify → commit, tests as the contract, failure modes | Lab 3 |
| 6 | 15 Learning, careers, and responsibility | What the evidence says about learning, "How to learn with AI," academic integrity | |
| 7 | 10 AI code review | How AI reviewers work, how humans should review AI-written PRs | Lab 4 |
| 8 | 11 Security and prompt injection | Instructions and data share one channel, the lethal trifecta, student checklist | Lab 5 |
| 9 | 03 The model landscape | How to read model names and tiers, "Best model for the job" | |
| 10 | 05 How to tell if a model is good | What a benchmark is, the benchmark-claim checklist, build your own eval | Lab 6 |
| 11 | 09 Multiple agents | Why and why not, building vs reviewing, cross-model review | Lab 7 |
| 12 | 02 Training and tuning | The pipeline at a glance, distillation, "Models that tune themselves" | |
| 13 | 04 Open vs closed weights | The openness spectrum, licenses, the hardware math | |
| 14 | 12 Costs | Three ways to pay, why agentic coding eats tokens | |
| 15 | 13 Open-source ecosystem | A map of the stack, etiquette, "How students can contribute" | Lab 8 |
| 16 | 14 Keeping up | What stays stable, a curated information diet, a sustainable routine |
If you only have a weekend¶
Read the TL;DRs of every module, then read these four in full: 00, 01, 08, and 15. Then do Lab 3.
Habits to build from day one¶
- Spend ten minutes on a problem before you ask an AI. You'll ask a better question, and you'll learn more from the answer (module 15).
- Ask for hints, not answers, when the goal is learning. Ask for answers when the goal is shipping. Know which one you're doing.
- Read every line you submit. If you can't explain it, you don't get to submit it.
- Make the agent prove it. "Run the tests and show me the output" beats "does it work?" (module 08).
- Keep a "surprised me" log. Every time a model does something you didn't expect, write one line. It becomes your personal eval set (module 05) and your study guide.
- Follow your course's AI policy and say what you used. Make disclosure a professional habit.
Where to go next¶
- The engineer path once you have an internship or job.
- The papers list, starting with the short list of eighteen.
- Contribute to an open-source project with AI help, carefully (module 13).