How to read this book — and a disclaimer
This short playbook covers Build a Course in an Hour: how a complete course gets built fast with three tools — Cowork to structure it, the Minisite skill to build the pages, and Midjourney to illustrate it — using Build Your Own AI Team as the worked example. Five short chapters, read front to back in well under an hour.
This is educational only — not financial, tax or legal advice. AI product features, pricing and connectors change quickly; treat specifics as current at June 2026 and verify before relying on them. For regulated work, AI prepares and checks while a human reviews and lodges.
Contents
What's Inside
The Idea
Chapter One
Build Your Own AI Team
You are reading a course that was built live in about an hour — structured with Cowork, built with the Minisite skill, and illustrated with Midjourney. The chapters that follow show exactly how. But first, the subject we chose to build it about: building your own AI team.
For a non-technical founder, an ‘AI team’ is the set of assistants you can now assemble around yourself — Claude plus its ecosystem (Cowork, Claude for Chrome, Claude for Microsoft 365, Skills) and hundreds of MCP connectors that let it read your email, your calendar, your books, in plain English. Agents are a spectrum: prompting (you ask, it answers), workflows (fixed if-this-then-that), and agents (you give a goal and it observes, thinks and acts in a loop). The connective tissue is MCP — ‘USB-C for AI’ — a common standard that lets one model plug into many tools.
The durable rule, straight from Rupert’s QuickBooks course, still governs everything: automate the reads, supervise the writes. Let AI handle low-risk reading and drafting; keep a human approving anything that sends, pays, publishes or lodges.
The idea that matters
You don’t need to code. The skill is to give Claude a bounded job, connect the right tools, and supervise the result — not to write one clever prompt.
Try this
Before going further, write down three jobs in your work you’d most like to hand off. By the end you’ll know how to build a small ‘team’ to help with them — under supervision.
The Build
Chapter Two
Step 1 — Structure It with Cowork
The first move in building a course is structure, and it’s the perfect job for Cowork — Claude’s agentic desktop mode. You point it at the research, give a plain-English goal (‘turn this into a course like my others’), and it does the multi-step work: reads the material, proposes modules and lessons, asks a couple of clarifying questions where it’s genuinely unsure, and produces a phased build plan you approve before anything is built.
The shift is from ‘write me a prompt’ to ‘do this whole job, and check with me at the forks.’ You steer; it executes. Cowork shows its reasoning, lets you redirect mid-task, and asks permission before anything destructive.
Key insight
Give Cowork the outcome, not a prompt. Let it scope the work, ask its questions, and hand you a phased plan — then approve and let it build.
Try this
Take any pile of notes or research and ask Cowork to propose a structure for it — a course, a guide, a deck. Notice the clarifying questions it asks; those are the decisions that matter.
Chapter Three
Step 2 — Build the Pages with the Minisite Skill
With a structure agreed, the pages get built — and this is where a Skill earns its keep. A Skill is a small reusable package (a SKILL.md plus optional templates and scripts) that teaches Claude to do a specific task your way, every time. The Minisite skill encodes this site’s design system — the navigation, footer, colours, fonts, lesson layout and SEO — so every new landing page, dashboard and lesson comes out consistent without hand-building each one.
Skills and MCP are complementary: MCP gives the tools; Skills give the repeatable method. The same Skill runs across Claude.ai, Cowork and the API, and can be shared with a team — build the method once, reuse it forever. (Only install Skills from sources you trust; a malicious skill can run code.)
Key insight
A Skill turns ‘explain the house style again’ into ‘use the house style.’ It’s how one good design system gets applied perfectly across dozens of pages, fast.
Try this
Think of one task you explain to people (or to AI) the same way every time. That’s a candidate for a Skill — write the steps down once, and you’ve started building one.
Chapter Four
Step 3 — Illustrate with Midjourney
A course needs to look the part, and Midjourney turns a sentence into polished images. But here is the part most people miss: you don’t write those Midjourney prompts yourself. You ask Claude (or ChatGPT) to write them — describe what you need in plain English, and it hands back a numbered set of ready-to-paste prompts. You copy each one straight into Midjourney and generate. Every image in this course was made that way; none was typed into Midjourney by hand.
The trick is to tell Claude the things people forget. Ask it to keep one consistent style across every prompt — and once you’ve found a look you love, to reuse that image’s style-reference (--sref) code so the whole set matches. Tell it where each image will go so it sets the right aspect ratio (16:9 heroes, 3:2 in-lesson, 2:3 a book cover, 1:1 cards). And ask it to make the people genuinely diverse — varying ethnicity, gender, age and body type, with competence across all — so your imagery is inclusive and free of stereotype by default.
Key insight
Don’t type prompts into Midjourney. Ask Claude to write them — specifying one consistent style, the aspect ratio for each slot, and diverse representation — then copy-paste the results into Midjourney.
Try this
Ask Claude to write five Midjourney prompts to illustrate something you’re working on — one consistent style, the right aspect ratios, and diverse people. Copy one into Midjourney and see how close it lands.
Ship It
Chapter Five
Ship It — and Do It Yourself
The last step is shipping: wiring the new pages into the site — the course catalogue, the sitemap, the navigation — so people can actually find them. A course no one can reach isn’t finished.
Then the honest part: nothing AI-built is set-and-forget. Models change, connectors come and go, facts date; anything you build needs a review point and an owner. Put it together and you have a repeatable loop — Cowork to structure, a Skill to build, Midjourney to illustrate — that you can point at your own content, products and, ultimately, your own AI team. Start read-only, advance as you learn to trust the drafts, and keep the rule that makes it all safe.
Lodgement, payments and money stay human
For regulated work — bookkeeping, pay runs, BAS, super — AI prepares and checks while a human reviews and lodges, with a registered agent in the loop. ‘The AI did it’ is not a defence to the ATO. Educational only, not financial advice.
Try this
Pick one real job — a newsletter, a set of product pages, a weekly brief — and run the loop once: structure it with Cowork, build it with a Skill or template, illustrate it, ship it. Then book a time to review it.
You've reached the end
For the interactive version — the five lessons with illustrations — head to the Build a Course in an Hour dashboard. Remember: this is educational only, it was reviewed in June 2026, and it’s a living book — check back for the latest edition or grab a fresh PDF.