Count how many times in your life you have bought dish soap and you quickly reach three digits, and count how many times you actually chose one and you arrive at roughly once. That was somewhere in your first flat, when the bottle with the green label ended up in the cart for reasons nobody can reconstruct today and was never questioned again. Every purchase after that was not a decision but a repurchase, triggered by an empty bottle next to the sink and carried out by a hand that knew the way to the shelf better than its owner did, and for every one of those repurchases the loyalty card dutifully credited its points.
So the loyalty program at the supermarket or the drugstore has for decades mostly paid for repetition, and that was a sensible deal as long as a human stood at the shelf whose habit could be tended with a few points so that it would not wander off to the competitor across the street. Every purchase is either a decision or a habit, and this inconspicuous distinction is currently sorting out which purchases go to an AI agent first.
On 6 October American Express published the Amex Business Playbook for Agentic Commerce, a guide for merchants, together with two surveys the advisory firm Teneo conducted in August 2026 among 502 decision-makers at US businesses and 2,005 US consumers who shop online regularly. The most striking figure sits in an unassuming paragraph about trust: 52 percent of the consumers surveyed feel comfortable letting an AI agent buy household essentials and groceries, but only 15 percent would leave electronics or appliances to one. Anna Marrs, who heads the global merchant business at American Express, adds “We believe agentic commerce could be the most significant change in shopping and purchasing behavior since the start of e-commerce”, and even if you do not share that assessment, it is plain that the change does not begin everywhere at once.
Nobody Lets an Agent Choose the Wedding Ring
The obvious reading of these numbers is economic, and American Express suggests it itself when it speaks of “everyday, low-cost items” on one side and “high-value items” on the other: the more expensive a mistake, the less control people give away, so the cheap things go to the agent first. This reading is plausible and probably not entirely wrong, but it makes a prediction that fails in everyday life. By its logic the twelve-euro bottle of wine you bring to your in-laws on Saturday should be an ideal job for the agent and the cat food that costs more than that wine every month a poor one, and anyone who has bought both suspects it is rather the other way round.
Nobody lets an agent pick their wedding ring, while nobody ever picked their dish soap in the first place, and between the two runs a line we at Engaginglab call the Delegation Line. It separates purchases that are still decisions from purchases that have long since become habits, and price fits this line so neatly only because expensive things are bought rarely and therefore hardly get the chance to become habits.
The brain draws this line out of thrift rather than laziness, because the neuroscientist Karl Friston describes it as an organ that constantly predicts what happens next and tries to keep the gap between expectation and outcome as small as possible (Nature Reviews Neuroscience, 2010). From this angle a habit is a behavior whose outcome is predicted so reliably that it no longer needs attention, so what gets handed over first is whatever carries no uncertainty, while people keep whatever could produce an expensive prediction error. The 15-euro present for a nine-year-old niece therefore belongs to the second kind more than the 20-euro detergent does, even though it is cheaper, because with the detergent you know what comes out, and with the niece you never do.
The Trendex data cannot settle whether this reading is right, because household essentials are cheap and habitual at the same time, and a survey about hypothetical behavior, commissioned by a payment network with its own stake in the subject, cannot pull the two apart. That separation would be the most useful question for the merchant advisory council American Express is launching as a pilot, namely whether the 52 percent were already rebuying their household essentials out of habit before, because that decides whether people are handing over a choice or a habit.
The agent first takes over the purchases that were never decisions, and those are exactly the purchases loyalty programs have paid for with points for decades.
What DoorDash Already Shows
DoorDash has already turned the Delegation Line into a product, because since 30 September the delivery company has been testing an agent in the US to which customers can send a text message that says nothing but “order my usual”, and the agent knows that this means, say, the Friday night order. According to company figures compiled by PYMNTS, more than 30 percent of grocery orders placed through the Ask DoorDash assistant are routine restocks, which can now be rebuilt with a single tap, and before checkout the assistant scans the grocer’s shelves for cheaper swaps.
That last clause describes what happens to a habit once it changes hands: the human who rebought the green dish soap out of loyalty to a long-forgotten decision was a reliable customer for brand and retailer, while the agent doing the same repurchase checks on every order whether the same result is available for less. Where the agent helps with the choice itself, and is therefore part of a real decision, nearly half of the restaurant orders placed through Ask DoorDash went, by DoorDash’s own numbers, to places the customer had never tried. Andy Fang, DoorDash’s co-founder, said at the company’s Dash Forward event who benefits from this shift: “Every other AI agent is meeting you for the first time. We’ve known you for years.” The habit does not vanish into thin air, it moves to whoever holds the customer’s memory, and that is less and less often the loyalty card.
The Agent Runs the Switch-Off Test on Every Purchase
The industry has an answer to this worry that sounds reassuring at first: BigCommerce argues that loyalty programs become more valuable once agents can read them, because member prices, points and benefits then flow directly into the agent’s purchase decision, and sums it up in two sentences: “Suddenly, loyalty isn’t just a retention tool. It becomes a decision-making input.” According to the same report, Walmart has announced plans to let Walmart+ and Sam’s Club members sign in to their accounts inside AI shopping environments.
The sentence is true, and that is precisely the problem, because a benefit read by an optimizer becomes a row in a comparison table in which three percent in points at one retailer competes with four percent off at another, and what wins in that table is a price, not a bond. For reward systems of every kind we at Engaginglab use a simple check, the Switch-Off Test: remove the reward, in thought or in practice, and watch whether the behavior stays. In agentic commerce the agent runs this test itself on every single purchase without anyone asking it to, and any loyalty that hung only on the points vanishes the moment another retailer offers one cent more.
A habit is a purchase whose outcome the customer predicts so confidently that it no longer needs attention, and those purchases go to an agent first. A decision is a purchase where a mistake would cost something, in money, in reputation or in self-image, and those purchases the customer keeps.
Retail loyalty programs have so far paid almost exclusively for habit, while their value in agentic commerce is decided by the few purchases that are still decisions.
Who Loses Their Moment First
The Delegation Line runs through every assortment, but it hits industries in a particular order, first grocery and drugstore retail, whose loyalty programs rest almost entirely on repurchases, later the home improvement store, and only at the very end the retailer selling kitchens or televisions, who will probably not notice anything for a long time.
We have called the moment in which a person hands authorship of a purchase to an agent the Zero Moment of Agency, and in that moment it is decided whether a brand appears in the instructions as a fixed parameter or is, for the agent, only one candidate among many, sorted by price, availability and rating. American Express devotes a whole chapter of the playbook to this question, “Stand Out When AI Narrows the Choices”, but with dish soap the decisive moment is so inconspicuous that nobody notices it, because the instruction simply reads “the usual”. The agent then has to interpret for itself whether the usual means a brand or merely a product category it may replace with something cheaper at the next offer, and that interpretation only goes the brand’s way if the customer ever had a reason of their own to want that particular brand.
The Diaper and the Week Before the Birth
How upside down the budget often is shows in a drugstore case, assembled from patterns that are typical there. A young family buys diapers in their child’s first year at a rhythm few other categories reach, and the loyalty program credits points for every pack, for the eightieth just as for the first. The actual decision, which brand and which drugstore, was made in the weeks before the birth, when the parents stood at the shelf for the first time without a clue, and in exactly those weeks the program did nothing that set it apart from any other program. From the second pack on the purchase is habit, and that habit is the first thing to go to the agent.
The birth of a child is one of the moments we call the Off-Autopilot Moment, along with a move, an expiring contract, a job change and the first bad experience with a brand, moments in which a person drops out of habit and decides again which relationships to keep. In agentic commerce the agent takes over the autopilot and the human is left with the Off-Autopilot Moment. For a loyalty program this shifts the work in three places: away from the reward for the eightieth repurchase, towards the few weeks in which a household chooses for itself again, and finally into the sentence the customer dictates to their agent when handing over the shopping, a sentence in which a drugstore only appears as a fixed parameter if it did something in the week before the birth that the parents remember.
The Harbor Pilot’s Job
A harbor pilot does not steer the ship across the ocean, which the autopilot handles more reliably than any human on the open sea, but boards precisely where the water turns shallow and the approach confusing, and in those few hours on the bridge it is decided whether the ship reaches port safely. Loyalty programs are getting a related job in agentic commerce, because the agent will soon sail the open sea of repurchases, and the people responsible for loyalty become pilots for the approaches, for the move, the birth and the expiring contract, and for the moment a customer tells their agent for the first time what to buy for them from now on.
The dish soap with the green label will go on being bought, probably more reliably than before, only by someone who owes the loyalty card nothing. What remains open is when your customers last decided for themselves to choose you, and who was with them in that moment.