The gap between knowing and doing
You've heard the advice. Document your processes. Build a system. Get things out of your head.
Maybe you've even seen a framework for how to do it. You understand the logic. And then you sit down to write and produce something like this:
"Record podcast episode."
That's a task label. It's not a procedure. And the difference between those two things is the difference between having an AI strategy and having something AI can work with.
Descriptions vs instructions
Most people, when they try to document what they do, write descriptions. They describe the task at a level that makes sense to them - because they already know how to do it. "Edit the newsletter." "Process the invoice." "Prepare for the client call."
These are useless to anyone who isn't you. They're useless to AI for the same reason. There isn't enough information to act on.
A procedure is different. A procedure is a set of instructions precise enough that someone who has never done the task before could follow them and get the same result you would. Not a similar result. The same result.
The first-hire test
Here's the standard to hold yourself to: if you hired someone tomorrow and handed them this document, could they complete the task without asking you a single question?
If the answer is no, if they'd need to come back and ask "which tool do I use?" or "where do I find that file?" or "what do I do when this happens?", the procedure isn't finished.
This is the same standard that applies to AI. An AI agent shouldn't need to ask you clarifying questions mid-task. It needs the steps laid out, the tools named, the decisions pre-made.
The first-hire test and the AI-ready test are the same test.
What a bad procedure looks like
Take podcast recording as an example. A common first attempt at documenting this:
"Prepare content. Set up equipment. Record episode. Send to producer."
Four steps. Looks tidy. Completely useless. Every one of those steps hides a dozen decisions and actions that only make sense because you've done them hundreds of times.
"Prepare content" … prepare how? From where? What counts as prepared? "Set up equipment" … which equipment, in what order, with what settings?
If you handed this to a new hire, they'd be back in your office within five minutes. If you handed it to an AI agent, it would either stall or guess. Neither is what you want.
What a good one looks like
Here's the same task, written as a procedure:
Open the content calendar. Find the next episode marked "ready to record." Note the title and subtitle.
Open the episode planning document. Review the talking points. Add, remove, or reorder based on anything that's changed since planning.
Open Descript. Create a new project using the episode title.
Position the camera. Check framing and lighting. Confirm the microphone input is selected.
Hit record. Pause for three seconds before speaking -- it gives a clean edit point.
Work through the talking points in order.
Stop recording. Save the project.
Export the audio file and upload it to the shared folder for the producer.
Every step names the tool. Every step specifies the action. No step requires you to know something that isn't written down.
A new hire could sit down with this and produce a podcast episode draft. Not a perfect on (they'd lack your delivery and your instincts) but a functional one. That's the bar.
Where AI shows up
Now look at that procedure again, specifically steps 1 and 2. Pulling the next episode from the calendar, grabbing the title and subtitle, cross-referencing a knowledge base, and assembling a structured set of talking points. That's repetitive, predictable, low-judgement work. The kind of thing an AI agent handles well.
You couldn't see that opportunity from "prepare content." You can see it clearly from the detailed procedure. What used to take 15 minutes of manual work each week is now handled automatically before you sit down to record.
That's the pattern. The procedure doesn't just document what you do -- it reveals what you shouldn't be doing. Every step that's repetitive, predictable, and doesn't need your judgement is a candidate for automation. The steps that need your taste, your relationships, or your expertise -- those are the ones worth protecting.
Write one this week
Pick a task you did today. Not the most complex one. Something you do regularly that takes 15 to 30 minutes.
Set a timer for 30 minutes and write the procedure. Step by step. Name the tools, name the files, name the decisions. Apply the first-hire test at every step: could someone follow this tomorrow without asking you anything?
Then look at your steps. Mark each one: does this require my judgement, or is it just execution? The execution steps are your AI backlog.
One procedure. Thirty minutes. That's not a strategy document or a transformation roadmap. It's a single, concrete asset that something other than you can act on. The next one gets easier, and the backlog builds from there.
🎧 Listen to the full episode: The 5P Framework: How to Build an AI Backlog for Your Solo Business
📄 Resources and templates: Free Resources | Lone Wolf Unleashed