Can an agent buy from you?

Where it usually breaks

A small shop sells hand-poured candles, and an AI agent is asked to find a lavender candle under fifteen pounds that ships this week. The shop comes up. The description is accurate, the price is right, the AI recommends it confidently. Then it tries to complete the purchase, checking real stock in the right scent, confirming the price hasn't changed since the page was last read, and placing a real order, and there's nowhere for it to do any of that. The website was built for someone to read and then email the shop, not for something to check and then act.

That's the test worth applying honestly: not "would an AI mention us", but "could an AI buy from us, right now, without a person stepping in halfway through." Most small shops have never asked themselves this question, because until recently there was no version of a customer that needed the answer to be yes.

Three places, in order of how often they trip a shop up. Stock: a product page says "in stock" when the real number, three left in that scent, only lives in a spreadsheet nobody's updated since Tuesday. Price: the page shows last month's price because a sale ended and nobody updated the listing, so an AI checking now gets a fact that's already wrong. And confirmation: even when stock and price are both correct, there's often no way for anything other than a person filling in a checkout form to place the order and get a real confirmation back.

None of these are exotic problems. They're the same problems a busy shop owner already juggles by hand, checking the spreadsheet before promising a customer something's available, updating a price tag after a sale ends. An AI trying to buy on a customer's behalf runs into exactly the same gaps a careless member of staff would, except it has no way to notice and correct for them the way a person can.

A person closes this gap without noticing. An AI has no way to.

A customer emailing to ask about stock gets a reply from someone who checks the actual shelf, not the website, so the mismatch between the page and reality never becomes visible to anyone. A person is quietly doing the job of closing that gap every time they answer the phone. An AI acting on a customer's behalf has no shelf to check and no reason to doubt the page, so it repeats whatever the page says as fact, confidently, and the mismatch a person would have quietly caught goes straight through to a customer who trusted the answer.

The fix here is often writing the fact down for the first time, not just wiring two systems together

Most small shops would fail this test today. Not because anything about the business is wrong, and not as a verdict on the shop: it's a description of a gap that's genuinely new, one most shops haven't had reason to think about until an AI tried and failed at exactly the point a person would have caught it without even noticing they had.

For a hotel with a booking engine, closing a gap like this is mostly a connection job: the true rate and the true availability already live somewhere, a system the hotel already runs, and the work is linking it to something an AI can check directly. A shop this size often doesn't have that system to connect in the first place. The true answer, whether three jars are really left in that scent, whether the price genuinely is still fifteen pounds, sometimes lives nowhere but a notebook by the till or whoever last looked at the shelf. That's a different kind of job, usually smaller but more basic: not wiring two systems together, but writing a fact down properly, current and checkable, before there's anything to connect at all.

Selfe's job either way is the same: connecting a shop's real stock, prices and checkout, however they're currently kept, to something an AI agent can check and complete a purchase through directly, rather than leaving it to read a page and hope the page is still true. This hub's own piece on why being found isn't the same as being bought from covers the fuller version of the same gap.

Does this mean rebuilding our online shop?

No. For a shop that already tracks stock and price somewhere, it means connecting what exists to something an AI can check directly. For one that doesn't yet, it means writing those facts down properly for the first time, not rebuilding the shop itself.

How would we even know if we'd pass this test?

Ask the same question an AI would ask, about real stock in a specific size or scent, and see whether anything other than a person reading the page could answer it correctly today.