The LLM Patchbook

A Patchbook guide: short, simple chapters that serve as a conceptual upgrade for software engineers.

What you will learn: the component parts of an AI agent; how to build classifiers, summarizers, and translators; how to build AI agents; how to build your own coding harness; how to fit LLMs into your architecture; how to reason about trade-offs with LLMs; how to reason about LLM compliance, privacy, and regulations.

What you will lose: some of the mystery surrounding AI agents.

Prerequisites: basic programming, including loops and API calls. A few years of experience helps but is not required.

Nice to have: you may have worked with ChatGPT, Claude, or some other chatbot. You may have worked with Claude Code, AWS Kiro, or some other coding AI.

You do not need to know: math, data science, or machine learning.

In development. Chapters appear below as they are written. The list above is the full planned roadmap; only the chapters listed below exist so far.

• Chapter 0: It sounds more complicated than it actually isAI agents are a handful of primitives, not magic.
• Chapter 1: An agent is a for loopChatGPT, Claude Code, and other agents are just while loops.
• Chapter 2: How to send a chat request to an LLMTalking to an LLM is just a POST request with a sender and a message.
• Chapter 3: Common uses of single API callsSingle API calls are not full agents, but they're already very useful.
• Chapter 3a: SummarizationA zero-shot summarizer. No machine learning required, just an API call.
• Chapter 3b: ClassificationA classifier that used to require ML expertise. Now just an API call.
• Chapter 3c: TransformationTranslation, simplification, and data conversion. Sequence-to-sequence learning, now just an API call.