I gave my AI a MacBook Air.
Not a metaphor. An old MacBook Air I hadn’t used in a long time, sitting in the closet with a dead battery and my daughter’s stickers on the lid. I plugged it in, wiped it, set it up as its own machine, and installed the AI stack on it. That was the beginning of me not treating AI like an app anymore.
Before that, I was doing what everyone else does. I’d open Claude on my MacBook Pro — the same laptop I use for design work, client calls, the tabs I don’t close, the twelve half-written docs I keep meaning to finish. I’d ask it to help me draft a proposal or reconcile a bank statement or write a caption, and it would help. Then I’d close the tab, or the machine would restart, and everything it knew about me evaporated.
The real problem wasn’t the memory loss. I could fix memory. The real problem was that I had my whole business — every client note, every password manager, every unfinished draft — living on the same machine that was running an AI I hadn’t fully thought through the security of yet.
That bothered me more than I wanted to admit.
What actually made me nervous
I’m not paranoid. I’m not a security engineer. But I run five businesses. There is a lot of information about other people on my primary laptop. Client tax IDs. Vendor contracts. Screenshots of conversations. Half a decade of business context sitting in one browser session.
And I was about to start letting an AI operate inside that environment full-time. Not just chat with it once and close the tab. Actually give it agency — file access, credential access, the ability to write and edit and send.
The moment I imagined it having read/write access to my Downloads folder while my daughter’s dance recital videos were also sitting there — same folder, same permissions — I closed the laptop and thought: this isn’t the way.
The AI doesn’t need my Downloads folder. The AI doesn’t need any of my personal context that isn’t strictly its job. What it needs is its own workspace, its own memory, and clear boundaries around what it can see. Sandboxing. Not because I don’t trust the model — but because clean environments make everything downstream cleaner.
So I gave it its own machine.
The MacBook Air experiment
The old Air was slow. It only had 8GB of RAM. The battery didn’t hold a charge and the fan sounded like a small helicopter. But for what I needed — a dedicated environment for running my AI stack, storing operator memory files, and holding the workflow scripts — it was fine.
I wiped it. Set up a fresh user account with no personal iCloud, no photo library, none of the tabs and passwords I have on my main machine. Installed the AI stack. Created folders for each operator — one for Ledger, one for the research operator, one for the content one. Each folder had its own memory file, its own configuration, its own scope of what it was allowed to touch.
I ran the whole setup off that MacBook Air for about six months.
Three things changed the moment I did that.
One: I stopped closing tabs at the end of the day. The AI stayed running. Its memory stayed intact. When I came back Monday morning, it remembered what we were working on Friday. Not because I did anything clever with prompts — because the machine never restarted. It had a persistent home.
Two: I got comfortable giving it more real work. When I knew it was operating in its own sandboxed environment, I stopped hesitating. I let it read the folder of client emails I’d exported. I let it draft with my brand voice files loaded in memory. I let it run overnight while I slept. That would have felt reckless on my main laptop. On its own machine, it was fine.
Three: I stopped losing an hour a day to context switching. Before, every time I wanted the AI to help me, I had to context-switch out of whatever else I was doing — close windows, quiet the notifications, focus. When the AI lives on a different machine sitting on the corner of my desk, I don’t switch contexts. I glance over. It’s already working.
Why the Air couldn’t scale
By month six, I was running four operators in parallel. The Air was choking. The fan was constant. Response times were dragging. And I was starting to want to run bigger models — the ones that need real memory to run well.
I bought a Mac Mini. The M4 with 32GB of RAM. Not the top spec, not the base. The one that felt like it would last me for years without me thinking about it again.
Set it up the same way. Fresh user account. No iCloud. Just the AI stack, the operator folders, and the memory files. Moved everything over in an afternoon.
The Mac Mini has been sitting in a docking station under my desk for eight months now. It doesn’t have a screen attached. It doesn’t have a keyboard I use. I ssh into it from my main laptop when I need to configure something. Everything else, the operators run themselves.
That machine has become the single most important piece of hardware in my business. It runs my bookkeeping operator, my content operator, my research operator, and my visual operator. It has never crashed. It has never lost memory. It has never let me down.
The MacBook Air got a second life
Here’s the part I didn’t expect.
My daughter had been watching me set this up for months. She asked me one day, half-joking, whether she could have “one of those” for her homework. She meant an AI she could talk to that would help her with her English essays.
I looked at the MacBook Air sitting in the closet again. The one I’d used to run my whole business on for six months.
I wiped it. Set it up with a fresh account for her. Installed a smaller AI stack — nothing near what I run, just enough for her to have a persistent assistant. Gave it its own memory file. Trained it on the writing patterns her teacher wanted, so it wouldn’t rewrite her voice, just help her structure her ideas.
She has been using it for three months. Her writing has gotten better. Her confidence has gotten better. And she has never once asked to use my laptop or the Mac Mini or anything else in the house. Her AI has its own desk too. Her own hardware.
My husband saw what she was doing and asked for one for his side project. That’s three AI environments running in our house now. Each one has its own hardware, its own memory, its own scope. None of them touch anyone else’s data.
That is what security means when you actually run AI in a real household or a real business. Not a subscription to a security tool. Not paranoia. Just clean environments with clear boundaries.
Why every operator gets its own desk now
The Mac Mini lesson turned into a principle.
Every operator I run — Ledger for bookkeeping, the content operator, the research one — has its own home. Its own folder. Its own memory. Its own scope of what it’s allowed to see. They talk to each other when they need to, through structured hand-offs I designed. But they don’t share environments.
That’s what makes them reliable. It’s also what makes them safe.
When I set up the AI COO Setup Kit, this is the piece that surprises people the most. They come in thinking they’re going to install a chatbot. They leave with an AI environment that lives in its own space, has its own memory, remembers what happened yesterday, and doesn’t touch anything it’s not supposed to touch.
The hardware matters. The environment matters. The boundaries matter. That’s what “own desk” actually means. And it’s the difference between AI as a party trick and AI as a member of your team.
If your AI doesn’t have its own desk, you’re still just borrowing a tool. Once it does, you’re running a business with a real team.
What to do next
If you’re running AI on the same laptop you use for everything else — that’s the first thing I’d change. Not because it’s dangerous. Because it’s noisy. Because context switching costs you an hour a day and you don’t realize it.
You don’t need a Mac Mini. You need any machine — even a five-year-old MacBook Air — that you can dedicate. Once the AI has a home, everything else in your setup gets simpler.
The AI COO Setup Kit walks through this in the first section. What hardware. How to sandbox. How to set up memory that persists. Then it goes through everything else — the operator files, the workflows, the daily habits — that turn a machine into a team member.
Give your AI a desk. Then let me know what changes.