"I don't trust AI. I'm on the hook if it gets something wrong."
I hear this from owners who already use AI. They're right. It will get something wrong.
So will your next hire, and that is a training problem we can fix.
Nobody gives a new hire the customer list and the company card on day one. They check in. They ask questions. You read their work until they've earned the room to do it on their own. AI should start the same way, and that's how I run it in my own business and for the clients I build for.
Start on a short leash. Every action or tool the AI can use is configured with zero trust and slowly builds trust (and more autonomy) over time.
Keep the work visible, always. One place you look every day, where you can see what the AI is doing and has done.
Give it the handbook. Your prices, your policies, your standards. The more context AI has, the better judgment it will return, and if it isn't written down, it doesn't get said.
It knows when to escalate. When something needs you, it flags it and waits for your decision. Low risk work goes first, and the check-ins come down over time.
Keep the receipts. Log everything it does, so when something looks off you can see what happened and when.
At Yardigans, a yard care company I build for, that's an approvals page the crew checks from their phones and a Slack log of everything the agent does. On my own development team of AI agents, it's three sign offs from me before anything ships, and then the AI reviews my approval one more time.
This week, pick the one thing you'd never let a new hire do without checking it first. That's your first approval.
Set up this way, nothing goes out with your name on it that you didn't mean to say.
If you want a second set of eyes on how to set yours up, message me.
What's the one thing you'd never let it do on its own?