AI Engineer Starter Handbook — AI skill for Claude Code

The AI Engineer's Starter Handbook

*A no-nonsense guide for developers who want to build real AI systems, not finish more tutorials.*

---

Who this is for

If you have skill level equivalent to a 2nd-year CS student and you want to get serious about building AI systems, this is for you.

You should be able to:

• Write Python without looking things up constantly
• Use git (add, commit, push, branch)
• Call a REST API and parse JSON
• Read a technical paper without giving up on the first page

You do NOT need to know ML, understand transformers, or have taken any AI course. This is an engineering handbook, not a research primer. If you can build a web app with Flask or FastAPI, you can build production AI systems. Same discipline, different primitives.

The mindset shift

Most AI tutorials teach you to use frameworks. That's backwards.

The frameworks churn every six months. What compounds is:

1. Understanding what's happening underneath
2. Evaluation literacy (how do you know it works?)
3. Production taste (what breaks in the real world?)

Here's the rule to internalize: **build the dumbest version first, then add abstractions only when you feel their pain.** Every shortcut you take in the beginning costs you later when you have to debug a black box.

That's why this handbook has you build with the raw API before touching LangChain or LangGraph. Don't skip it.

---

Section 1: LLM Mechanics and Application Primitives

*Time budget: about one week of focused work.*

What you're actually learning

You're learning to treat an LLM as an API. It's a stateless function that takes text, makes decisions, and returns text or structured data. That's it. The magic is in how you compose calls, structure inputs, and validate outputs.

Concepts to understand

**Tokens.** LLMs don't see characters or words. They see tokens: chunks of text from the model's vocabulary. This matters because:

• You pay per token (input + output separately)
• Context windows are