The Vending Machine Problem

On what breaks when your customer is no longer a person.

---

Sometime in the last eighteen months, the customer changed.

Not the demographic. Not the channel. The customer itself — the thing on the other end of the transaction — started to be something other than a human being. A growing share of web traffic that looks like browsing is actually an agent fetching, comparing, and increasingly, deciding. Google's latest Search rollout puts a buyer agent in the hands of anyone with a phone. ChatGPT does checkout. Perplexity does checkout. The pattern is not coming; it has arrived, and the only real question is what percentage and how fast.

Most of the commerce internet was not built for this. It was built for a person with eyeballs, a cursor, and a slow afternoon. Fixed prices on a product page. A promotion banner timed to a holiday. A checkout flow optimized for cart abandonment among humans who get distracted by a text message. None of those assumptions hold when the customer is a piece of software that has read the entire site in under a second, has perfect memory of every competitor's price, and has no emotional response to a banner that says LIMITED TIME.

This is the vending machine problem. A vending machine is a beautiful piece of retail when the customer is a tired commuter who wants a Coke. It is a catastrophe when the customer is another machine that can instantly query every vending machine in the city, compute the cheapest Coke per ounce, and route itself there with no friction. The vending machine has no way to respond. It cannot lower its price for this one buyer. It cannot bundle. It cannot ask what the buyer actually wants. It just sits there with its fixed prices and its fixed inventory and waits to lose.

That is the position most websites are now in.

---

The instinctive merchant response is to harden the walls — block the bots, rate-limit the scrapers, force a login. This is the wrong move for the same reason it was the wrong move when search engines arrived in the late nineties. The traffic is not the enemy; the traffic is the new shape of demand. Blocking it is choosing not to be in the market.

The second instinctive response is to lower prices, on the theory that agents are infinitely price-sensitive and the cheapest seller wins. This is also wrong, and more dangerously so, because it is half-true. Agents are price-sensitive, but they are not *only* price-sensitive. A good buyer agent — and the ones being deployed now are getting good quickly — is optimizing for its user's actual situation. Does this need to arrive by Friday? Then speed matters more than price. Does the user value warranties? Then the seller offering one wins even at a premium. Has the user just bought a complementary product elsewhere? Then a bundle they didn't think to ask for can win the transaction outright.

The merchant who races to the bottom on price is solving the wrong problem. They are answering a question the agent never asked.

---

What agents actually want, if you watch them carefully, is a counterparty that can *respond*. Not a price tag, but a position. Not a promotion, but a proposal. They want to be able to say: my buyer cares about this, will you move on that — and get an answer that is intelligent rather than scripted.

This is, of course, just commerce as it existed for most of human history. Before fixed-price retail (an invention from the 1870s, more recent than people realize), every transaction was a negotiation. The shopkeeper read the customer, the customer read the shopkeeper, and they arrived at a price that reflected what both sides actually valued. Fixed pricing was an efficiency hack for the industrial age, when scaling a store meant you couldn't have your best negotiator behind every counter. The web inherited this hack and made it worse, because at least a human shopkeeper could glance up and read the room.

Agents can read the room. They are, in a real sense, the return of the shopkeeper's eye — except now it sits on the customer's side, and it never blinks, and it scales to every transaction simultaneously. The bazaar, restored at planetary scale, except half of the participants have not realized the bazaar is back.

---

There are three things every merchant should understand about this shift, and they are not the things most merchants are currently focused on.

The first is that the unit of competition is no longer the page. It is the offer. A page is a static thing built for a browsing human; an offer is a dynamic thing shaped to a specific buyer's situation in a specific moment. The merchants who continue to optimize pages will be optimizing artifacts the agents barely look at. The merchants who learn to shape offers will be the ones still standing in five years.

The second is that the data that matters has moved. For two decades, the most valuable signal in commerce was clickstream — what humans did on your site. In an agent-mediated world, the most valuable signal is what offers were accepted, what offers were rejected, what offers were countered, and what the pattern of those responses says about real-time demand. This data is being generated right now, on every site agents touch, and almost no merchant is capturing it.

The third is that this is a network problem, not a site problem. A single merchant tuning their own responses in isolation is bringing a vending machine to a market that now runs on conversation. The merchants who win will be the ones plugged into something larger than themselves — something that sees the patterns across thousands of transactions and knows, before any individual seller could, which offers actually clear.

---

The buyer side of agentic commerce is being built at full speed by trillion-dollar incumbents. It is well-funded, well-staffed, and going to be ubiquitous within the year. The merchant side of that same equation is almost entirely unaddressed. Most merchants do not yet have a clear picture of how their traffic is changing, let alone a strategy for it. The board conversation is still about conversion rate optimization on a human funnel that is quietly being replaced. The marketing budget is still being spent on ads that agents do not see, click, or care about.

This is the most interesting commerce problem of the next decade, and almost nobody is treating it like one. The merchants who do — who stop defending the vending machine and start building the shopkeeper — will keep their margins when the dust settles. The ones who do not will be vending machines, in a world that no longer has tired commuters.

The shift has already started. The question is only who notices in time.

---