‹ All hubs
Agentic Engineering

Loops, tools, and context — the craft of building software around models that act.

Agentic Engineering / Overview
Entry point

Overview

Building software where a language model does the work: assembling what it sees, designing what it can do, measuring whether it succeeded, and keeping the whole thing fast enough and cheap enough to ship.

What this hub is

For most of software history, you told the computer exactly what to do. Agentic engineering inverts that: you build an environment — context, tools, feedback — and a model decides what to do inside it. The code you write stops being the worker and becomes the workplace. That inversion sounds small and changes almost everything: how you debug (read transcripts, not stack traces), how you test (statistics, not assertions), and where the effort goes (interfaces and evidence, not algorithms).

This hub teaches the discipline that has grown up around that inversion. It is written for someone who can program but has never built an agent, and it optimizes for understanding per minute — trunk first, details only where they carry weight.

How to read it

The sections are numbered and the order matters — each one leans on the ones before it.

  1. The Map — the whole field on one page. Start here.
  2. Mental Models — the eight ideas everything else hangs off.
  3. Vocabulary — the working terms, tersely defined. Also a reference page you’ll come back to.
  4. Major Problems — where the field’s energy actually goes, and why.
  5. Topics 05–10 — the agent loop, context engineering, tool design, evals, multi-agent systems, and cost — in dependency order, each with a hands-on step.
  6. Be Creative — the capstone: you make something.
  7. Go Deeper — a short, opinionated reading list.

The site’s depth toggle gives you two paths. Tourist mode shows only each section’s summary — a fast, self-contained pass through the whole subject. Full mode shows the complete write-ups. Flip to Tourist when you want the shape, Full when you want the craft.

You will build something

This hub ends with you producing an artifact: a one-page proposal, a written critique of an existing agent system, or a small working prototype. Each topic’s “Apply it” step feeds that capstone — by the time you reach it, you’ll have a working mini-agent, an eval set, and opinions. Read with that destination in mind; reading to build sticks far better than reading to finish.

How long it takes

The Tourist path is about twenty minutes. Reading everything in Full mode is three to four hours. Doing the “Apply it” steps adds four to six hours at a keyboard, and the capstone runs from an hour (an idea) to a weekend (a prototype). You don’t have to do it all at once — the numbering means you can always find your way back.