The Method ·
Why AI automations break, and what to build first
Most AI automations are built on top of nothing written down, so they stop working the first time the business changes. There is an order that survives: context, then skills, then workflows.

Someone online connects an email inbox to a spreadsheet, connects the spreadsheet to a chat app and points an AI at all of it. By the end of the video every new enquiry is read, sorted, answered and filed, and nobody has touched a thing. It works. You watch it work.
So you copy it. You give up a Sunday, wire up your own version, send yourself a test email and it works for you too. For a couple of weeks you tell people you have automated your enquiries.
Then it stops, and not with an error message. A customer mentions they never got a reply. You go and look, and the thing has been quoting a price you stopped using, or filing everything into a folder nobody opens, or sending the same message twice to the same person. You poke at it for an hour, give up and go back to doing the job by hand.
None of that is unusual, and the reason it happened is not that you built the thing badly.
What actually broke
Every step in that automation contained a decision about your business. Which enquiries deserve a fast reply and which can wait. What you charge. Who gets copied in. What a good answer sounds like. What to do with the awkward request that fits none of the categories.
None of those decisions were written down anywhere. They were typed into the middle of a step as a setting, and each one was true on the afternoon you typed it.
Then your business did what businesses do. You added a product. You put your prices up. A supplier started sending PDFs instead of a spreadsheet. Someone left and the person who replaced them works differently. Every one of those is a normal Tuesday, and every one of them makes a setting buried inside your automation wrong.
The worst part is that you cannot find which one. The knowledge is scattered across settings inside several different tools, none of it reads like a sentence and none of it sits anywhere you would think to look. So you don't repair it. You rebuild it from scratch, and a few months later it breaks again for a different reason.
An automation cannot know anything about your business that your business has not written down.
Three ways this usually goes wrong
The broken automation is one of three shapes, and plenty of businesses are sitting inside one of them right now.
Chat with nothing kept. You and your team use AI in a chat window. Everyone types their own background in every time, so two people asking the same question get different answers, and whatever you taught it on Tuesday is gone by Wednesday. It feels like using AI. Nothing accumulates.
Someone else's downloaded template. A prompt pack or a ready-made workflow, bought or free. It was built around their prices, their customers and their process, and those are the exact things that make your business yours. So the output comes back close but wrong, and you cannot correct it, because you never saw the decisions sitting inside it. A skill built on someone else's business will never fit yours.
Automation with nothing written underneath. This is the one from the video. Several tools joined together, plenty of steps and no written record anywhere of how the business actually works. It runs beautifully until the business changes, which it will.
Notice what those three have in common. In none of them has anybody sat down and written the business out.
The order this has to be built in
There is an order, and it is not optional. Chat is level one, and it is where everyone starts. Above it sits context, then skills, then workflows, then the operating system. Each level only works when the one underneath it is real.
Context is your business written down so an AI can read it: what you sell, what you charge, who buys it, how you work and what you refuse to do. Most of it currently lives in the heads of three or four people and has never been typed out anywhere.
A skill is one repeated job written down once. The Monday report, the quote, the reply to a late payer, the follow-up after a proposal goes quiet. Written properly, it reads like the instructions you would hand a new employee.
A workflow is skills joined up and put on a schedule. The report collects the numbers, writes itself and is sitting in your inbox at 7am with the three things that moved flagged at the top. Nobody started it.
The operating system is all of it in one place your team logs into: what ran, what it produced and what needs a person to decide.
Skip a level and the level above it inherits the hole. A skill with no context underneath produces confident nonsense, in the wrong tone, at the wrong price, and it is confident enough that somebody sends it out. A workflow built on shaky skills breaks silently, and you hear about it from a customer weeks later. A dashboard sitting on top of broken workflows only shows you the broken work faster.
This is also why the video worked for the person who filmed it. Their business was in their head the whole time they were building, so every decision they typed into a step was the right decision for them. You copied the wiring. You could not copy the thinking, because the thinking was never written down. It was never written down for them either, which is why their version will break as well.
What to build first
The first thing you build should not be an automation. It should be a document.
Pick the job that annoys you most, or the one that costs you the most, whichever you feel more strongly about. Open a blank document and write down how that job gets done, as if you were teaching someone who starts on Monday: what you need in front of you, what a good one looks like with a real example pasted in underneath, what to avoid, and what to do when it goes wrong.
Write it in sentences rather than shorthand. Inside the broken automation, your rule about deposits was a number typed into a box. Written down properly it reads like something you would say out loud to a new hire, along the lines of: we take a deposit on anything we have to order in specially, we skip it for the handful of accounts we have invoiced for years, and anyone asking for credit terms comes to me before a price is quoted. A number in a box goes out of date silently. A sentence goes out of date in front of you, in a document you can read, and you correct it once in one place.
Then use it. Paste that document in front of your normal request before you ask for anything. The difference arrives immediately, because the answer comes back with your prices in it, in the way your business actually speaks, with the exception you always have to explain by hand already handled.
It costs nothing to start. There is no software to buy, no subscription and nobody to hire, because all you are doing is typing what you already know into a document. If you wrote one of these a week, you would have most of your business written down by the end of the quarter. The whole method is written out step by step in the AIOS Guide if you want the longer version.
More than 15 automated routines run our businesses every day, including the weekly reports, the meeting notes and the finance checks. Every one of them started as a document somebody wrote, and not one of them started as a connection between two apps.
A skill only earns a schedule after it has been written down and used enough times that you trust it. That order is the entire difference between something that survives your business changing and something that does not.
What you write down stays yours
Your context and your skills are plain text in ordinary documents, in accounts you already control. There is no special format and no platform holding them.
That matters more than it sounds. A better AI model will arrive next year, and another one the year after. When it does, you point it at the same documents and carry on working. Nothing has to be built again, because nothing was locked inside the old one.
The alternative is what a tangle of automations leaves you with. The way your business works exists only as settings inside one company's tool, which means you are renting your own process, and you get to build it all over again when that company changes its pricing or shuts down.
Where to start this week
If you are not sure which of the five levels your business is on, the AIOS Score asks ten plain questions about how you work today and gives you your level and the single next move worth making. It takes about two minutes and there is nothing technical in it.
If you would rather just start writing, everything on the resources page is free. There is a prompt that interviews you and turns your answers into your first context document, and a list of the ten documents worth writing down first, so you are not sitting in front of a blank page deciding where to begin.
Start with one document this week. The automation can wait, and it will work when you get to it.
