Season 1 • Topic Idea 1
Day 29
By Friday, a small online shop has turned away three long-time customers asking for refunds, and two of them have said so in public reviews. Nobody broke a rule. That’s the strange part.
Here’s how it usually goes. A small online business hands its refund emails to an AI assistant. The owner gives it the refund policy, the same page that’s on the website: refunds within 30 days, receipt required, no exceptions. It’s a clear page. It reads well. It has sat on the site for years without a single complaint.
For those same years, the refund emails went to the senior support rep. She knew things the page didn’t say. A regular customer got the benefit of the doubt. A late package was the shipper’s fault, not the customer’s, so the 30-day clock started when the box arrived, not when the order went in. Anyone who sounded upset got a phone call instead of an email.
Nobody had written any of that down. Nobody had ever needed to. She just knew, and because she knew, the page looked complete.
On Monday the assistant starts answering. It reads the page and does exactly what the page says. A customer whose package arrived on day 29 asks for a refund a week later. The page counts from the order, so the assistant says no, politely, with a link to the policy. A regular who lost the receipt gets the same polite no. An upset customer gets a calm, correct, three-paragraph email, which makes things worse. When that customer writes back, angrier, the assistant answers the only way it can: by quoting the policy again.
On Wednesday the owner skims a few of the replies. They’re polite, accurate and on brand. Every one quotes the policy correctly. Nothing looks wrong, because nothing in them breaks the page.
By Friday the owner is reading reviews and deciding the AI is broken. It isn’t. It followed the policy perfectly. The policy was never the whole job. The job was the policy plus everything the rep did without thinking, and only the first half made the trip.
I’ve Handed Over the Same Page
I’ve done this. Not with refunds, but with the same move: I handed someone the written version of a job and expected them to do the real one. A new hire, a contractor, and lately an AI assistant. Every time, the written version looked complete to me, because I was the one quietly filling in the rest.
Nobody in that shop was careless. The rep was doing good work that nobody had ever asked her to describe. The owner did what any busy owner does: handed over the page that existed.
The gap was there before the AI arrived. A good person was covering it. The AI doesn’t cover gaps. It does what the page says, so the gap shows up all at once, usually in front of a customer.
In a Boston Consulting Group survey, executives put about 70 percent of their AI trouble on people and process, and about 10 percent on the AI itself. If you’ve had a week like that shop’s, the cause was most likely on the page. That’s the good news, in an odd way: a rule nobody wrote down is the one kind of problem you can fix with a pen.
Five Lines Before I Hand Anything Over
Now, before I hand a task to an AI, I write down what a good coworker would “just know.” Five questions, answered in plain sentences, pasted in with the instructions every time. I answer them with whoever does the task now. The person carrying the rules is the only one who can name them. The answers go in with the task, not in a separate document nobody opens. It takes a few minutes the first time and almost nothing after that, because the answers barely change.
The question that does the most work is the third one: which exceptions do people handle without thinking? For the refund emails, that one question would have caught the late-package rule and the regular-customer rule. The fifth question, when to stop and check, would have sent the upset customers to a person instead of a polite paragraph.
The price is real. A few minutes per task adds up when there are many tasks. The first week runs slower, because I read the first few results myself instead of trusting them. And the list has a limit I can’t get around: it only catches what someone can name. The rule nobody knows they follow still slips through, which is why reading those first results matters.
Some people will think the old way was better. The person who always just knew may feel the list turns their judgment into paperwork. It’s the opposite. The list is the first time anyone wrote down how much they were carrying.
The “Just Know” List
Copy this. Answer each line in a sentence or two, and paste the answers in with the instructions every time you hand the task over.
THE “JUST KNOW” LIST
Task: ______________________________
1. What does “done” look like, in one sentence?
2. What would you never do here, even if asked?
3. Which exceptions do people handle without thinking?
4. Who should it check with when it’s unclear?
5. What should make it stop and check with you?
The AI didn’t cause the problem in that shop. It exposed it. Every team has a version of this: a task that looks documented because someone reliable is quietly carrying the parts that aren’t. Nobody notices the carrying until that person takes a vacation, and the process turns out to have been a person all along.
Line worth keeping: The AI didn’t create the gap. It removed the person who was quietly covering it.
Try the list on one task you hand over this week. What was the exception you’d never written down?
Where To Go Next
Want the management view?
→ Subscribers can download the companion Leadership Brief: LB-001: What Your Team Just Knows: A Worksheet for Handing Work to AI [<brief_url>]
It turns the idea into a management tool. Inside:
• warning signs
• questions to ask
• economics
• Monday action
• worksheet
If you’re trying to understand where people are quietly compensating for missing process, start here.
Want the technical leadership view?
→ The companion Expression Layer post on LinkedIn The Expression Layer (LinkedIn)
Want the architecture view?
→ The companion Expression Layer article on Substack The Expression Layer
→ Free Subscribers can download the companion Leadership Brief: LB-001: What Your Team Just Knows: A Worksheet for Handing Work to AI.
Includes:
• Executive summary
• Questions to ask
• Monday action
• “Just Know” worksheet