Coming soon Free sample first

Agentic Engineering for the Rest of Us

Claude Code, Codex, OpenCode and Pi - and the habits that carry between them, so no single AI company can take your workflow away.

On 9 June 2026 I got the best coding model I'd ever used, included in a plan I already paid for. By Friday night it was illegal for Anthropic to serve it to me. Eighteen days later it came back.

This book is what I did about that. It covers the four AI coding tools worth learning, what each is actually for, and how to set up your work so that switching between them takes an afternoon.

Book cover: Agentic Engineering for the Rest of Us by Harry Munro
Don't marry the model.

01 The premise

In June, my best tool was taken away for eighteen days.

Most people who use an AI coding agent use one. One tool, one company, one model, and a way of working that only exists inside it.

That works until the one company changes its mind, or someone changes it for them. In June a US export-control letter switched off a frontier model for everyone, mid-session, with no warning. Google shut Gemini CLI down for consumers the same month. My last book had said "Gemini CLI's free tier makes it the obvious starting point if budget is tight." It lasted barely six months.

So this book puts the durable stuff first: what a harness is, where your agent's memory actually lives (plain files), how to judge the work that comes back. The tools come second, as swappable detail. Anything that will date is fenced off in boxes marked "this detail will age", so you know which pages to distrust in a year.

02 What's inside

Four tools, one way of working.

Claude Code for collaborating, Codex for delegating, OpenCode for freedom, Pi for control. What each one is good at, and when I reach for which.

Collaborating

Claude Code

Delegating

Codex

Freedom

OpenCode

Control

Pi

My real split, if you want it: roughly 75% Claude Code, 15% Codex, 5% OpenCode, the rest Pi. An opinion, not a comparison chart.

  1. 01

    The ten-minute proof

    Write one AGENTS.md and one skill, then watch two different tools obey the same instructions in the same project.

  2. 02

    The Portability Audit and the Fire Drill

    A checklist that measures how locked in you are, and a one-day exercise where you swap providers on a real project and write down what breaks.

  3. 03

    Process before power

    Specs and plans an agent can't misread. It's a thinking skill, not a coding skill, and it's the place where people who don't code for a living can beat people who do.

  4. 04

    The money, honestly

    A budget ladder from a free tier up to Max plans, and exactly what each rung costs you in capability. Every expensive chapter shows the cheaper path.

  5. 05

    From one good conversation to a software factory

    Loops, subagents, multi-agent workflows and a per-project chief of staff, introduced one rung at a time, only when the rung earns its cost.

  6. 06

    Built to age well

    Versions, prices and plan names live in boxes marked "this detail will age", so you know which pages to distrust in a year.

Written in British English by an engineer who uses all four.

03 Who it's for

For people who build things with agents but don't write software for a living.

  • You already use an AI coding agent, and you've wondered what happens if it disappears or doubles in price
  • You're a product manager, analyst, engineer, researcher or founder, and you're comfortable in a terminal (or willing to be - Chapter 3 is honest about the alternatives and what they cost you)
  • You want opinions from someone who works in these tools every day, not a feature comparison chart

Not for you if

you're after a browser app-builder walkthrough. The book tells you to skip Lovable, Bolt and Replit, and explains why.

New to all of this?

Start with Claude Code for the Rest of Us. It's free and covers the terminal, the install and your first builds. This book picks up where it stops.

04 The small print

Questions, answered

Is the book out yet?

Not yet. The free sample (the first three chapters) comes first. Leave your email and you'll get it the day it's ready. I haven't put a date here because I'd rather not make one up.

What will you send me?

The sample, the day it's ready, and perhaps a short note or two about the ideas in the book before that. After that, one email when the book's available. Nothing else unless you ask.

Won't a book about AI tools be out of date in six months?

Parts of it, yes. I've built it for that. Versions, prices and plan names live in boxes and end-of-chapter Field Guides stamped "as of mid-2026", so the rest of the book doesn't rot around them. The ideas are what you're buying: harness versus model, memory as plain files, portable setups, keeping the judge separate from the worker.

Do I need a $200-a-month plan?

No. It's the best experience for a few power chapters, and I say so. But every technique works on cheaper tiers, and the book shows the frugal route each time.

I'm in the UK or EU. Does it still work for me?

Almost entirely. The exception is Sakana's Fugu Ultra (one chapter), which isn't available in the UK, EU or EEA. I checked the UK myself.

Is this a new edition of Claude Code for the Rest of Us?

No, it's the next book. That one gets you set up and building in one tool. This one gets you working across four without being stuck with any of them. It points back to the first book for the basics instead of teaching them again.

Did AI write this book?

Semi-automated is the honest answer. The ideas were entirely mine; I used Claude to help settle the structure (whether to include project examples, how to sequence the chapters) and to build a skeleton, then repeated loops often orchestrated by Grok Bot to handle research, fact-checking and editorial review using skills I refined to keep the writing in my own voice. I wrote my own system for creating the manuscript in PDF and EPUB. The full end-to-end proofread and final edit were done by me, the cover design is all me.

Will there be a paperback or Kindle edition?

A paperback, yes. The email list gets the PDF and EPUB first, sold direct from this page. The paperback follows on Amazon about two weeks later. There's no Kindle edition, but the EPUB can go on a Kindle through Amazon's Send to Kindle. Nothing's for sale yet.

The form did not work

If the page reported an error, wait a moment and try again. Failing that, email [email protected] and I will add you by hand.

05 The author

Harry Munro

Harry Munro is a Chartered Engineer with more than a decade of industry experience spanning technology, simulation and AI, in companies large and small. He founded the School of Simulation in 2024 and consults actively across industry. A Londoner by origin and a Bermuda resident since 2023, he enjoys a rum-fuelled island life while helping people and companies around the world reach their technical goals; in his spare time he is a dad, and holds enthusiastic amateur status at skiing, sailing and squash. He also wrote Claude Code for the Rest of Us and The Ralph Loop.

Free sample

Get the first three chapters free.

Including the whole Eighteen Days story. One email when it's ready.