Before you buy an AI agent, write down one process
Agents only work as well as the decision you taught them. For most small business owners the rule still lives in their head, which is why the most valuable first step with AI doesn't involve AI at all.
An AI agent can only follow the rules you can write down. Most owners can't write theirs, because the rules have lived in their heads for years, and I've watched a client's AI project stall for exactly that reason.
Someone on my feed recently told a 12-person company that it needed autonomous agents.
What they say: "AI agents will run your business."
What they mean: "You'll still do the work. You'll just do it while babysitting a demo."
I've built AI products for years, so here's what actually happens when a small business adopts an agent the way the hype sells it.
The agent moves the work, it doesn't remove it
The agent doesn't take the work off your plate. It moves it. Instead of doing the task, you're now checking whether the agent did it right, in the right order, with your customer's name spelled correctly, using data that's been wrong since last quarter.
That's a new job you didn't sign up for.
The quiet truth nobody posts, because it doesn't sell: an agent only works as well as the decision you taught it. If you couldn't write the rule down yourself, the agent can't follow it.
And most small business owners can't write it down, because the rule has lived in their head for eight years. When to offer a discount. Which enquiries are worth a call back. What "urgent" means for this customer but not that one. It's judgment, and nobody has ever had to explain it.
The smaller, better version
So "agents will run your business" really means something smaller, and that's the good news.
You don't need to document your whole company. You need one process: the one eating the most time or causing the most rework. Write that one down, step by step:
- What starts it (an email arrives, a form is submitted, a day of the week comes round).
- What information you look at.
- The decisions you make, and the rule behind each one.
- What you produce or send.
- How you know it was done right.
Once you've written that down, you've done the step most owners skip: you've pulled one rule out of your head and onto paper.
I learned this with my own AI agents. Before any of them drafted a single reply on X, I had to write down rules I'd only ever carried in my head: how long a reply can be, which topics it must never touch, which facts I'm allowed to cite, and never saying the same thing twice to the same person. That written list is what lets them work without me watching. When one rule was worded too narrowly, a fact I'd retired slipped through, and the fix was rewriting the rule, not the AI. The case study has the details.
I've also watched the opposite, with a client. A crypto investing academy asked me to build a system that spotted promising opportunities in social media chatter. The system worked: it collected tens of thousands of posts, scored them and sent alerts. What stalled was the part only they could supply. Their real edge was their own judgment about which opportunities were worth backing, and it had never been written down. They expected the AI to know it. It couldn't, because nobody had told it. The case study covers what I'd do differently.
Why it's worth doing even if you never automate
That step is valuable before any automation works. Every process you can name stops being something only you know. You can hand it to a new hire, a virtual assistant or an agent, and whichever one you choose, you can check the work against the steps you wrote.
A business that lives in one person's head can't grow, can't delegate, and can't survive that person leaving.
So the honest pitch for adopting agents is this: it doesn't start with teaching a machine to run things. It starts with writing down one process you already run. For most owners, that's the part they've been putting off.
It's less exciting, and it's far more likely to work.
The steps are only half of it. The other half is the judgment inside them, the approvals only your most experienced person knows how to make. Start by asking that person what would make them say no, and write their answer down word for word.
Originally posted on LinkedIn