Add an approval step before anything goes out
Design the pause: who approves, what they see, and what happens if nobody does.
Agents & automation
Where a person approves, and how to design the pause so it actually happens.
5 min readPlain English
In one sentence
Human in the loop means a person checks or approves an AI’s work at the points that matter, before anything is sent, paid, published or deleted.
Think of it like…
In a busy restaurant, cooks plate the food, but the head chef looks at every dish before it goes out to a table. It works because the chef really looks, every time, and can send a plate back.
A good approval step is designed, not bolted on. It answers four questions: where the pause goes, who approves, what they see, and what happens if nobody does.
If nobody has changed a draft in weeks, either the AI is excellent or nobody is reading. Spot-check to find out which.
Good at
Bad at
Customer replies. An online shop’s assistant drafts answers; a person reads each one and clicks Send.
Supplier invoices. At a bakery, AI reads and files the week’s invoices, and the owner approves the batch on Friday before anything is paid.
Quotes. A plumbing firm’s assistant drafts a quote from the job notes; the owner checks the price and the promised date before it goes out.
Myth vs reality
Myth: Adding a human check makes AI safe.
Reality: Only if the check really happens. Make reviewing quick, give it to the right person, and never let anything risky go ahead without them.
Your team approves every AI-drafted reply, but nobody has edited one in a month. What should you do first?
Design the pause: who approves, what they see, and what happens if nobody does.
Forward receipts to one inbox; AI pulls out the details and you approve the batch weekly.
What to write down so everyone uses AI the same careful way.