You ask your assistant to order the groceries you buy every week. Halfway through the list, it has a question. The usual 500 g bag of pasta is sold out. Two 250 g bags of the same pasta are available for the same total price. Should it buy those instead?
Yes. You go back to making breakfast.
A minute later, another question. The lactose-free milk is sold out. Should it buy ordinary milk?
This time you’re glad it asked. Your sister is staying for the weekend. The lactose-free milk was for her.
Both questions came from the same shopping list. Both concerned a substitution. One interrupted you to settle a bit of arithmetic. The other needed something the assistant couldn’t learn from the supermarket shelf: why you wanted that particular bottle of milk.
You’d let the assistant buy the pasta. You want it to ask about the milk. That’s the Authority Boundary: the line between what it may decide for you and what it must bring back to you.
What “Do The Shopping” Means
Imagine three households using the same assistant at the same supermarket. All three give it the same list.
| What the household tells it | What happens at checkout |
|---|---|
| “Fill the basket. I’ll check it before you pay.” | The shopping stops until someone is free to review the whole order. |
| “Buy what’s on the list. A different pack size is fine if it’s the same product, the same total quantity and no more expensive. Ask about other swaps.” | The pasta goes through. The milk comes back as a question. |
| “Choose substitutes within our budget. Keep anything marked lactose-free, nut-free or gluten-free that way. If you can’t confirm that, leave it out and tell me.” | The assistant can choose another brand of lactose-free milk. If none qualifies, it can finish the order without milk. |
The third household hasn’t bought a cleverer assistant.
In each case, the assistant can find a substitute. What changes is who gets to choose it. If it asks, you spend a moment at breakfast deciding what to buy. If it chooses well, you get to finish your breakfast. If it chooses badly, you may spend half an hour going out for milk.
The third household has also decided what happens when there’s no suitable milk: send the rest of the shopping. That might suit a family with a corner shop downstairs. It might be a terrible arrangement for someone whose weekly delivery is their only practical way to get groceries.
There’s no prize for picking the widest boundary. The useful instruction is the one whose consequences you’re prepared to live with.
A Good Question Earns Its Interruption
“Ask me if you’re unsure” sounds sensible until you try to finish breakfast. The assistant could be uncertain about the colour of a packet, entirely confident that ordinary milk is a reasonable substitute, and wrong about which of those things matters to you.
A better place to start is with the last few questions it asked. Which ones did you answer the same way every time? Which made you put down your coffee and think?
The pasta decision is a good candidate for a standing rule. You can say what must stay the same, and the assistant can check the packet size and price before paying. You’ve made one decision that can serve you again next week. The milk decision depends on who’s coming to stay. You may prefer to keep it, especially if explaining every possible guest and dietary preference would take longer than answering the occasional question.
This is also a useful test for whether to automate something in the first place. If you order groceries every week, teaching the assistant your pack-size rule could be worth the trouble. If you’re choosing a birthday cake for your daughter once a year, you might happily let it find three bakeries and still want to pick the cake yourself. Choosing is part of what you wanted to do.
Count What Comes Back To You
Suppose the assistant stops asking about substitutions altogether. Your phone is quieter. Have you saved any time?
You’ll know more when the bags arrive. If you put everything away and get on with your evening, the arrangement probably worked. If you read every label because you no longer know what it might have chosen, some of the work has moved from breakfast to the kitchen counter. If you then have to go out for milk, the ten seconds you saved that morning were a poor bargain.
That tension shows up in a consumer survey by Ping Identity. 47% of respondents worried more about having to keep checking an assistant’s work than about the assistant making a mistake. That distinction is familiar enough: “Please stop asking me about the pasta” is a long way from “Buy whatever you think is best.”
The same choice comes up when booking an evening out. You might let an assistant reserve your usual table for two at seven. Now suppose the only table available requires a non-refundable deposit. The restaurant, the time and the people are all the same. What changed is what it would cost you to change your mind. That’s a good reason for the assistant to ask again.
It also suggests why one service might be worth more than another, even if both use the same AI. Would you pay a little extra for a grocery service that reliably honours your substitution rules and lets you correct an order before it leaves the shop? Perhaps. The test is whether you would actually trust it enough to stop checking the basket. If you’d still inspect every item, the promise of fewer interruptions is worth rather less.
Next week, give the shopping assistant one new instruction: “A different pack size is fine if it’s the same pasta, the same total amount and costs no more. If you can’t get lactose-free milk, ask me.”
That first morning gave you two interruptions to learn from. Next time, the pasta can go straight into the basket. And when your phone lights up while you’re making breakfast, it has a question worth putting the coffee down for.
Leave A Comment
A place for reactions, disagreements, corrections and useful additions to this entry.