Turn how you do things into a written process, with AI
The steps you know cold are strangely hard to write down, so they stay in your head and stay your job. Here is how AI turns a rough brain-dump into a process someone else can follow.
You know how to do the thing. You have done it a hundred times. It lives in your hands and your head, and you could do it half asleep.
The trouble starts the day someone else needs to do it too. You go to write it down, and it turns out the steps you know cold are strangely hard to put in order.
So the process stays in your head, which means it stays your job, which means you can never quite step away from it.
Why writing it down is the hard part
When you know something deeply, you skip the small steps without noticing. You forget to mention the thing you always check, because you do not experience it as a step at all. That is exactly the step a new person gets wrong.
A blank document makes it worse. Staring at "Step 1" is a different job from doing the work, and it is the job that never feels urgent, so it never gets done.
Talk it, don't write it
Here is the shift. Do not try to write the process. Just describe it, badly, out loud or in a rough note, the way you would explain it to a mate over coffee.
Open the model and say: I am going to describe how I do something, in a mess. Turn it into a clear, numbered process someone new could follow. Ask me about anything I skip.
Then talk. First I check the inbox, then I, oh and before that I always, you know the drill. Get it all out in whatever order it falls.
Let it find your gaps
The model takes your ramble and hands back something ordered. A trigger for when the process starts, numbered steps, a note on who does each one.
The gold is in the questions it asks back. What happens if the file is not there? Who signs this off? Those are the steps you skipped because you do them on instinct, and now they are on the page.
You answer, it slots them in, and in twenty minutes you have a written process that took years to learn.
What makes this work is that you are using the model as a structurer, not a knower. It has no idea how your business runs, so you never ask it to. You give it the raw brain-dump and ask it to impose a shape: numbered steps, a clear trigger, a note of who does what. Under the bonnet a language model is a very good pattern-completer, and "messy notes into an ordered procedure" is a pattern it has seen a million times. The risk to watch is that it will confidently fill any gap you leave, so the one rule is to check every step against what you actually do. Treat its draft as a well-organised first pass that you correct, never as instructions to follow.
Keep it honest
One rule. Read every step against what you actually do. The model will happily invent a plausible step to fill a gap, and a confident wrong instruction is worse than a missing one.
So treat its draft as a tidy first pass, not gospel. You supply the truth, it supplies the order. That division of labour is the whole trick.
What it buys you
A written process is not bureaucracy. It is the thing that lets you hand a job over, take a holiday, or train someone without repeating yourself for the tenth time.
I work alongside these tools every day, and getting a stubborn process out of your head is one of the quietest, most freeing wins there is.
If you have a few processes trapped in your head and no time to write them up, that is a lovely thing to sort together. Book a quick chat and we will make a start.
Book a quick chat →Related: I design systems, not prompts.
Common questions
What is an SOP?
SOP stands for standard operating procedure, which is a plain name for a written set of steps that describes exactly how a recurring job in your business gets done. It turns something that only lives in your head into something another person can follow and get right.
Does AI need to understand my business to write a process?
No, and that is the point. You use the model as a structurer, not a knower. You supply the real steps in a rough brain-dump and it imposes order, numbering, and clear triggers. It never needs to know your business, because you are the one supplying the truth.
How do I stop AI inventing steps that are wrong?
Read every step against what you actually do before you trust it. A language model will confidently fill any gap you leave, so treat its draft as a well-organised first pass you correct, not instructions to follow. The questions it asks back are usually more useful than the steps it guesses.
How long does it take to write a process this way?
Far less than staring at a blank document. Describing a job badly out loud and letting the model order it, then answering its follow-up questions, usually turns a job you have avoided for months into twenty minutes of work with a clean written process at the end.