The Cyber Week the agents actually moved

Skeptics of the agentic economy have one reliable move: everything is a projection. The consultant decks, the trillion-dollar forecasts, the surveys of intention: all of it describes a future that stays politely two years away. So this file does something different. It reads one week's receipts. Between November 25 and December 1, 2025 (Cyber Week), shoppers worldwide spent $336.6 billion online. By the count of the company that processes a large slice of those transactions, $67 billion of it was driven by AI and agents. That is not a projection. It happened, it was measured, and the measurement carries a date.

Section OneThe ledger, and who holds the pen

Start with the loudest numbers and the honest label they need. On December 5, 2025, Salesforce published its Cyber Week tally: global online sales of $336.6 billion, up 7 percent year over year, with the United States contributing $79.6 billion, up 5 percent. Inside that total, Salesforce counts $67 billion in sales driven by AI and agents, influencing 20 percent of all orders that week. Its own agent layer, Agentforce Commerce, powered 61 million orders in those seven days.

Salesforce, "Salesforce Data: AI and Agents Propel Cyber Week to Record $336.6B in Global Spend," December 5, 2025.

Read the verb carefully, because it is doing precise work. "Influenced 20 percent of orders" does not mean one in five purchases was completed by a robot. It means that somewhere on the path to one in five orders, a machine sat in the loop: a generated recommendation, an answered product question, an agent walking a customer to checkout. The $67 billion is the value of orders that passed through that loop, not the value of orders no human touched. That is a weaker claim than the headline suggests and still a remarkable one: a fifth of the biggest shopping week ever recorded had a machine somewhere in the decision path.

Now the label. These are Salesforce's own figures, from Salesforce's own Shopping Index, a model built on the activity of more than 1.5 billion shoppers across 89 countries flowing through its commerce systems. The company sells agent software. The figures have not been independently audited. That does not make them worthless; it makes them a vendor's ledger. A ledger is still a record of transactions rather than a forecast of them. But the pen belongs to a party with an interest, and any honest reading starts by saying so.

A ledger is not a forecast. Even a vendor's ledger.

Section TwoThe retailers who showed up armed

Inside the same report sits the claim that matters most to anyone who runs a store. Salesforce says retailers using its dedicated agent layer, Agentforce 360, grew Cyber Week sales 32 percent faster than their peers. And it names them: Pandora, Shark Ninja, Funko. Not anonymized "leading retailers": companies with reputations, on the record, attached to a checkable number.

Salesforce, "Salesforce Data: AI and Agents Propel Cyber Week to Record $336.6B in Global Spend," December 5, 2025.

The caveats are the same pen and one more. The 32 percent comes from the vendor whose product it flatters. And the comparison carries self-selection: brands that deployed shopping agents early almost certainly differ from their peers in more ways than the agents (budget, digital maturity, appetite for risk). What the number establishes is not that agents caused 32 points of growth. It establishes that in the biggest shopping week ever recorded, the retailers who treated machine-shoppers as real customers were the ones the vendor could parade, by name, without any of them objecting.

Even so, the skeptic's position gets uncomfortable here. Strip every generous assumption away and one fact remains: during the one week that decides whole retail years, the stores that had prepared a surface a machine could transact with grew, and the named examples were willing to stand behind the comparison publicly. Whatever the true causal share of the agents, "do nothing and wait" was the strategy on the losing side of the split.

Section ThreeThe direction check

Here is where the receipts stop being one company's story. Adobe Analytics, which tracks over one trillion visits to US retail sites and sells no agent platform, measured traffic arriving at US retail websites from AI sources (chatbots, assistants, AI-powered search) for November 1 through December 31, 2025: up 693 percent against the same period a year earlier. November alone was up 769 percent; December, 673 percent.

Digital Commerce 360, "Generative AI online holiday shopping traffic 2025," January 13, 2026, reporting Adobe Analytics data.

One discipline matters when quoting Adobe: the company has published a family of eye-catching percentages over the past two years, each belonging to a different measurement window, and blending them produces nonsense. The 693 percent belongs to one window only: US retail sites, November 1 to December 31, 2025, year over year. Every figure in this article carries its own window for the same reason.

The visitors are not just arriving; they are buying. Similarweb's third annual Global Ecommerce Report estimates that visitors referred by ChatGPT converted at 11.4 percent, against 5.3 percent for organic search, better than double the rate. And Bain reports that AI already accounts for up to 25 percent of referral traffic for some retailers, though Bain's own next clause notes that is still less than 1 percent of those retailers' total traffic.

Similarweb, "The Global State of Ecommerce 2025" (3rd Annual Global Ecommerce Report), Sept 2025; Bain & Company, "Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey."

The doubled conversion rate is worth a moment of mechanics, and here we are interpreting rather than measuring. A visitor who arrives from a search results page is often still comparing; the click is the start of the funnel. A visitor who arrives from an AI conversation has usually finished comparing inside the conversation: asked the questions, weighed the alternatives, and clicked through to the one answer that survived. If that reading is right, the 11.4 percent is not a curiosity of early adopters. It is what the end of a funnel looks like when the funnel has moved somewhere you cannot see it.

Three panels that sell no agents point the same way the vendor's ledger does.

Section FourThe consumer's own testimony

The last receipt comes from the shoppers themselves. In January 2026, IBM and the National Retail Federation published a study of more than 18,000 consumers across 23 countries. 45 percent already use AI to assist buying decisions: 41 percent for product research, 33 percent to interpret reviews, 31 percent to hunt for deals.

IBM Newsroom, "IBM–NRF Study: Brands and Retailers Navigate a New Reality as AI Shapes Consumer Decisions Before Shopping Begins," January 7, 2026.

The same study carries the honest counterweight, and we will not bury it: 72 percent of those consumers still prefer physical stores. Both facts are true at once, and the second explains the first's real meaning. The machine has not replaced the shop. It has moved in upstream of the shop, into the research, the review-reading, the deal-hunting that decides which shortlist a buyer walks in with. By the time 72 percent walk through a physical door, a growing share of them have already let a machine decide which doors were worth walking through.

45 percent already ask the machine. 72 percent still walk into the store. Both are true. The machine goes first.

In ClosingThe question is whether it can read you

Put the receipts in one line. In a single, dated week, a vendor's ledger counted $67 billion of AI-driven sales and one in five orders influenced by machines. An independent panel watching a trillion visits saw AI-referred retail traffic up 693 percent over the holiday season. Another found those visitors converting at double the organic rate. And 45 percent of 18,000 consumers told a fourth study they already shop this way. The biggest numbers are vendors measuring their own platforms (we have said so at every step), but every independent needle points the same direction.

For a small business, the conclusion is short. The machine-shopper is no longer hypothetical; it spent countable money in a week that already happened. The remaining question is not whether it arrives, but what it finds when it goes looking: whether your prices, your availability, your reviews, and what you actually do are legible to a machine that reads fast and skips anything it cannot parse. Machines can only act on what they can clearly read. The customers of Cyber Week 2025 sent theirs ahead to do the reading.

"Legible" is not a mystical property. It means prices stated in plain text on a page, not locked inside an image or a PDF. Availability and opening hours that a parser can find. Reviews that exist somewhere a machine indexes. A site that answers who you are, what you sell, and where you are in sentences rather than atmosphere. None of this is expensive. All of it decides whether the shopper who moved $67 billion in one week can put you on the shortlist, or skips to the competitor it could parse.

If you want to know what the machines currently see when they look at you, you can measure it at /signal-index/, or write to us at /contact/.