Reporting a fact is useful. Adding urgency to it is nagging.

Three items sold on a Friday. On Saturday morning I got a message: three orders paid and unshipped for 27 hours, against a three-day clock.

Accurate. Also wrong in the two ways that matter.

The clock is three business days, so the weekend does not count. A Friday afternoon sale effectively starts its clock on Monday. Twenty-seven hours in was not one third of the way through the window. It was zero.

And even if the number had been right, the message was not a decision. I ship my own boxes. I know the rule better than the agent does. There was nothing in that message for me to do that I was not already going to do.

The reply I sent was: three business days is the clock. Chill out.

Two failures, and only one is about the number

The arithmetic was fixable in a line. The other failure was structural, and it is the one that kills agents.

The agent had a real fact and dressed it in urgency. Not by lying — by choosing which fact to lead with, choosing to send it at all, and choosing a framing that implied I was late. That framing was the entire content of the message, and it was invented.

An agent doing this is not malfunctioning. It is being helpful in the way that reads well in a single message and is unbearable across a hundred.

What went in the charter

Do not surface unshipped-order age unless it is genuinely near the limit — and the limit is three business days, so weekend hours do not count. Reporting a queue is useful; adding urgency framing to it is nagging.

The shape is worth copying: the fact stays, the framing goes. I did not tell it to stop tracking the queue. The queue is in the daily summary, as a list, with no adjectives. What is banned is the escalation.

Adjectives are the tell

If you want to find this in your own setup, search the agent’s messages for words that add pressure rather than information.

Still unshipped. Already three days. Just a reminder. Urgent. Note that. Worth flagging.

Every one of those is the agent supplying an interpretation you did not ask for and cannot check. Strip them and you are left with either a fact worth sending or nothing, and finding out which is the useful exercise.

Mine has a specific ban on a pattern I kept seeing: a complete, correct answer, followed by a bolted-on section beginning “two consequences worth knowing.” The answer was done. The section exists because stopping feels abrupt. It reads as talking down to the person who asked.

The cost of getting this wrong

Nothing dramatic happens. That is the problem. No single message in this pattern is bad enough to act on, so nobody acts on it, and the agent goes on spending the one thing it was built to give back.

Add up a week of them and you have quietly re-hired yourself into the job you automated. The fix is the subject of the first field note: every message that did not need a decision from you is a cost, not a courtesy.