Most people are trying to pick the right AI tool. That is the wrong question.
The bigger move is to run your whole education business on an AI operating system — a chief of staff, departments, employees, and playbooks that live in a simple folder of files and do the work you used to do by hand. This tutorial walks through a live upgrade of that system and the ideas that make it worth setting up.
What an AI operating system actually is
It is not a fancy install. It is a plugin you drop into Claude Desktop that creates a folder of plain markdown files. Inside that folder you get a chief of staff (I call mine Dean), four departments, a starter team of around 23 AI employees, and a set of playbooks. No terminal, no code, no GitHub. You answer a couple of questions and you have an AI-native business tailored to education.
Because it is just files and folders, it is portable. The same system runs in Claude, in Codex, even in other agent platforms. You own all the data, and it moves with you.
Upgrading without losing anything
The upgrade is a single command: /update-workspace. It reads all the work you have done, plans the changes, folds in the new features, and keeps every customization. Nothing gets deleted. A safety backup is taken first. After it runs, a “campus doctor” checks that the whole system is healthy — stale files flagged, teams accounted for, memory intact.
The point of eating your own dog food: the version that ships to buyers is the version running the real business.
The big new idea: model routing to save money
AI has gotten more expensive, and it is hard to know where to spend. So the system now routes work to the cheapest capable model. Think of it like a real office: you do not ask a senior executive to do simple data entry.
When you give the chief of staff a task, they decide. A simple search or data cleanup goes to a cheap, fast model like Haiku — or even a free local model on your own machine. The heavy thinking stays with the premium model. That routing is built in now, so routine steps stop burning premium tokens.
Playbooks: the part that gets me fired up
A playbook is a written workflow for a multi-step job. Picture asking a real office to “run a social media campaign.” That is not one person. Someone writes the copy, someone understands the call to action and sequence, someone loads it into the CRM and schedules it, someone tracks it.
In the operating system, those are different AI employees who hand work off to each other. A playbook captures who does which step, how they know the job was requested, and how they confirm it was done well. Once it exists, you just say “I have a webinar on this date — get it ready,” and the chief of staff runs the playbook without you spelling out every step.
The wiki: a second brain that compounds
Every important output gets added to a wiki index — a memory layer for the business. So the system does not only save outputs into Claude’s memory; it builds a separate, searchable index of what you have learned and made. The practical payoff: you can be vague and it still knows what you mean, because the context is already there. You stop repeating yourself.
The campus map: seeing who is working
A dashboard opens as a web page and shows the whole business at a glance — agent runs, deliverables produced, hours saved, which departments are live, which scheduled tasks are running or idle, the team directory, and the wiki. You open it at the start of a session and refresh it whenever you want.
Why this matters now
Within the next few months, the likely shift is that you will simply talk to your chief of staff and ask for outcomes. To be ready you need three things: a clear understanding of your business, an operating system with employees who know that business, and a way to distribute what you make. A tool helps you do a task. An operating system runs the business so you can get back to teaching.
