Give Your AI SOPs: Skills Behind Every Routine

You wouldn’t hire a new team member, explain the job once, verbally, and then act surprised when they do it their own way. Yet that’s exactly how most people run their AI: instructions typed into a prompt, results half right, then the whole thing rewritten from scratch next time. The fix is the same one that works for people. Write the procedure down. Give your AI SOPs.

In the AI world these written procedures are called skills: a document that tells the AI exactly how you want a job done, which it reads every time it does that job. Every routine you run should have one behind it. This is probably the single highest-leverage habit I’ve built into how I run my businesses on AI, and it’s the thing I find myself teaching in almost every one-on-one session with builders.

The rule: fix the skill, not the routine

Here is why this matters so much. Say your morning briefing routine keeps summarising emails in a way you don’t like. Without a skill, you tinker with the routine, and your fix lives in one place, tangled up with the schedule and the plumbing.

With a skill, the routine is just a trigger: “at 6am, run the email briefing skill.” All the actual judgement lives in the skill document: how to prioritise, what to ignore, how you like replies drafted, what tone to use. When something comes out wrong, you open the document, add a line, and every future run is fixed. You’re not programming. You’re editing an SOP, in plain English, the same way you’d update a site procedure after something went sideways.

If a routine does something wrong, the routine is never the problem. The missing sentence in the skill is the problem.

A worked example: the email inbox skill

The skill I share most often is my email one. It’s a single document that tells the AI how to handle my inbox, and it includes things like:

• Look at the full conversation history with each contact before drafting anything, so replies carry context instead of starting cold.
• Cross-reference my calendar, so if someone asks for a meeting the draft reply already works around what’s booked.
• Recommend a reply for everything, so my job is deciding, not composing.

With that skill sitting behind a morning routine, my workflow is this: the briefing surfaces every email and message that needs me, I hit the microphone and talk through the lot in one pass, the AI drafts every reply with full context, and I review and send. What used to be an evening of typing is a few minutes of talking.

The skill is what makes this work. Without it, the AI writes generic replies and you correct the same faults every single morning. With it, every correction you’ve ever made is baked in.

How to write your first skill

Don’t start from a blank page. Start from an SOP you already have, or the procedure that already exists in your head:

1. Pick one job the AI does for you regularly. Email triage, quote follow-ups, meeting summaries, matching invoices against your price list.
2. Write down how you’d brief a sharp new office hire on it. What to check first, what good output looks like, what to never do.
3. Save it as the skill and point the routine at it.
4. Every time the output is wrong, add one line to the document. That’s it. That’s the whole maintenance model.

Within a month the skill knows your preferences better than a six-month employee would, because unlike the employee, it never forgets a correction. And skills compound: they’re documents, so you can share them. Inside Future Builder we pass working skills between members, which means a builder can skip weeks of tuning and start from a skill that already works.

This principle is the same one that makes filing come before AI: the intelligence is only as good as the documents it stands on. Get the SOPs written, and the AI stops being a clever toy and starts being a team member you trained once, properly.

Want your first set of skills built around how your business actually runs? That’s what my AI consulting sessions are for.