The library · 139 resources

Everything worth reading

Not a link dump. Every entry says what you get from it and when it is worth opening. Start with a path if you do not know where to begin, or use the filter and go hunting.

01 Reading paths

Four routes through the library, depending on where you are starting from.

I have never written code

For You have an idea and a laptop. Seven steps, about one week at forty minutes a day, ending with a live link you can send to someone.

I can code, but agents are new to me

For You already know git and a language. This path is about unlearning chat habits and building the verification loop that makes agents safe to trust.

I want to go deep on security

For You are about to put something real on the internet, or you are the person on your team who gets asked whether it is safe. Two evenings.

I want to run more than one agent

For One agent already works for you and you are wondering about five. Read the case against first, because most tasks do not need it.

02 The library

Essentials only

Start here

Eight links, in this order. They give you the definition, the one rule that keeps you safe, and the two guides every serious practitioner reads. One evening.

The story and the culture

One tweet, fifteen months, a Word of the Year and a change of mind in public. The arc is the lesson: vibes for throwaways, engineering for anything that matters.

The evidence: what the numbers say

Both sides, with sources. Measured effects sit between minus 19 percent and plus 25 percent; self-reported effects sit between 1.4x and 8x. That gap is the whole lesson.

Learn the fundamentals

Tokens, context, MCP, git and the instruction-file standards. This is the vocabulary layer — the words that make every other page on this site readable.

The agents, IDEs and app builders

Four surfaces, one ladder: browser builders, editor agents, terminal agents, cloud agents. Prices and free tiers change every quarter, so always open the page rather than trusting a tutorial.

Claude Code, end to end

The main track of this masterclass. Five ideas carry everything: the loop, context as the scarce resource, permission as trust, almost everything is reversible, and teach it once rather than every time.

Prompting, planning and context engineering

The one constraint under everything: attention is a finite budget and quality drops as the window fills. This shelf covers how to spend that budget, and the spec-driven toolkits built on top of it.

Testing, verification and quality

The agent stops when the work looks done. Without a runnable check, "looks done" is its only signal and you become the test suite. Everything on this shelf fixes that.

Security in the AI era

Two different problems: the app you shipped has no access control, and the agent you trusted has too much power. Both have known patterns, known fixes and a standard checklist.

What actually went wrong

A library of documented incidents, each with a root cause and a control you can apply tonight. Different vendors, one pattern: the agent has your authority and there is no undo.

Multi-agent, parallel and long-running work

The expensive end of the ladder. Read the case for it and the case against it before you spend three to ten times the tokens to get the same result.

Models, tokens and cost

Which model, at what effort, for how much. An agent re-sends its whole context every turn, so cost is mostly input — and a flat subscription is the predictable option for a student.

Deploy and infrastructure

The gap between a demo on localhost and a live link with auth and a database is where most beginners stop. Every service here has a real free tier — and a real limit worth reading first.

People, communities and what to do next

Nobody learns this from documentation alone. Follow a handful of people who publish their mistakes, join one room where builders ship, and pick a next course.