Fifteen years of ecommerce optimisation assumed a reader with eyes. Lifestyle photography, persuasive copy, urgency banners, social proof. Agentic commerce discards all of it. AI agents do not read a product page before payments. They read a feed, and if the feed is thin they move on before payments are ever involved.
The protocol landscape has settled enough to make this concrete. Competing agentic commerce standards now define how catalogue, pricing and inventory data reach AI agents, as recent surveys of the ecosystem describe. What they share is an assumption merchants rarely meet: that structured product data already exists before AI agents reach the checkout.
Vladyslav Kolodistyi, who works on payments infrastructure at PayAdmit, sees the same gap on most agentic commerce projects.
"Everyone asks about payments readiness first," says Vladyslav Kolodistyi. "It is the wrong order. Payments come at the end of a decision AI agents already made, with no checkout involved. If your data is bad, agentic commerce never reaches your payments stack at all, and you will conclude the channel does not work."
Agentic payments start long before the payment
The buying sequence in agentic commerce runs discovery, evaluation, selection, then payments. Human shoppers forgive weak data because they infer from images and context. AI agents do not infer. An attribute that is missing is an attribute that does not exist, and a product missing a size, a material or a delivery window will lose comparisons it should have won.
That makes catalogue quality a payments issue by consequence. A merchant with excellent payments infrastructure and mediocre data will see very little agentic commerce checkout volume, and no amount of payments authorisation tuning will fix it.
Vladyslav Kolodistyi describes the audit he recommends. "Take one product and ask what an agent could determine about it from your feed alone, with no page render," he says. "Most merchants discover half their differentiators live in marketing copy AI agents will never parse. That is the agentic commerce readiness gap in one exercise."
Pricing transparency is the second exposure. AI agents compare on total delivered cost, including shipping and any surcharge that appears late. Fees revealed at the payments checkout do not just harm agentic commerce conversion; they invalidate the comparison the agent already made, and some protocols will treat the mismatch as a failure.
The scale argument justifies the effort. Agent-led shopping is projected to reach a substantial share of ecommerce spending, as eMarketer has reported, and merchants absent from that comparison set never reach the payments stage at all. They are not competing.
A readiness sequence Vladyslav Kolodistyi uses with clients:
- Structure the catalogue so every purchase-relevant attribute exists as a field, not as prose
- Publish accurate real-time inventory, because AI agents penalise failed availability harder than humans do
- Make total cost including delivery resolvable before the checkout, not after
- Confirm the payments stack will accept agent-initiated payments rather than declining them by default
- Decide how returns and disputes route when AI agents placed the order
Only step four is a payments project in the traditional sense, and it is the shortest. The others determine whether agentic commerce traffic ever reaches it.
Data completeness is only half of it. Consistency matters as much, because AI agents comparing across merchants will treat a missing field and a differently named field identically. A catalogue that describes capacity in three formats across three categories is legible to a human shopper and unusable in agentic commerce.
Vladyslav Kolodistyi ties this back to payments outcomes. "Inconsistent data does not produce a payments error," he says. "It produces absence. Your payments team will report zero agentic commerce checkout volume and conclude the channel is early, when the real answer is that AI agents could not evaluate the offer."
Freshness is the third axis before payments. AI agents act on the data at the moment of decision, and a price or availability that changed since the last feed refresh will produce a failed checkout. In agentic commerce, a failed checkout is not a retry opportunity; the agent usually buys elsewhere.
What an AI agent checkout demands from the merchant behind it
Once AI agents decide to buy, the payments requirements are less exotic than the discussion suggests. Payments run on existing rails, through existing acquirers, with existing settlement. What changes is context.
"The AI agent checkout is not a new product," says Vladyslav Kolodistyi. "It is your payments stack and your checkout answering a call from a client you cannot see. The work is making sure the agent identity survives into the authorisation decision, and that your payments rules know what to do with it."
Merchants also underestimate how differently agentic commerce demand behaves. Human payments follow daily and seasonal rhythms. AI agents transact when the conditions in a mandate are met, which can concentrate agentic commerce payments into narrow windows with no marketing trigger behind them. Capacity planning built on historical curves will misjudge it.
Fulfilment feels the same effect. Agentic commerce can produce order patterns no human buyer would generate, including many small purchases across categories that never previously co-occurred, and payments reconciliation has to cope with the resulting checkout noise.
Payments configuration deserves a direct check rather than an assumption. Vladyslav Kolodistyi suggests three questions for the provider: does the payments platform accept agent-initiated payments today, does it preserve agent identity into authorisation, and what is the current default when AI agents reach the checkout. Answers are often uncomfortable.
"Most payments providers have not published an agentic commerce position," he says. "That does not mean they are neutral. It means the default applies, and the default in payments is to decline what you cannot recognise."
Returns policy is the piece merchants postpone. When AI agents order on a customer's behalf and the customer rejects the outcome, the return is not a normal one, and the payments refund path may not carry enough context to reconcile cleanly. Deciding the rule in advance is cheaper than improvising it at volume.
Vladyslav Kolodistyi warns against one specific overreaction. "Do not build a separate agentic commerce storefront," he says. "You will maintain two catalogues, they will drift, and AI agents will see the stale one. Fix the primary data. Agentic commerce is a distribution channel for the product information you already have, not a new business."
There is a competitive window here that he thinks merchants underrate. Structured data work is unglamorous, slow and invisible to customers, which means most competitors will delay it. The merchants who complete it early will be the default option in agent comparisons for as long as that gap persists.
Payments readiness should proceed in parallel rather than after. Confirming that AI agents are not silently declined at the checkout, that agent identity reaches the payments risk engine, and that disputes have a defined route are all short pieces of work. None require choosing an agentic commerce protocol, and all of them are wasted if the catalogue is not ready.
Measurement should be built alongside the work. Agentic commerce payments volume will be small at first and will hide inside aggregate payments reporting, so merchants need a segment for it from day one. Without that, the readiness investment cannot be evaluated and will lose its funding at the first budget review.
Vladyslav Kolodistyi is blunt about the sequencing risk. "Teams spend a quarter on agentic commerce payments work, cannot show the volume separately, and the programme dies," he says. "Instrument first. AI agents are easy to count in payments data if you decided to count them before they arrived."
Merchandising logic has to change alongside the data. Bundles, tiered pricing and conditional discounts are legible to human shoppers reading a page and often invisible to AI agents reading a feed, so a merchant's best commercial offer may simply not participate in agentic commerce comparisons. Encoding promotions as structured fields is unglamorous payments work with a direct revenue effect.
Vladyslav Kolodistyi sees the same omission repeatedly. "Merchants publish a clean feed and keep the promotions in the page template," he says. "AI agents then compare your list price against a competitor's promotional price, lose the comparison, and no payments team ever learns why the agentic commerce volume did not arrive."
Channel conflict deserves a decision too. If agentic commerce payments carry higher platform fees than direct checkout traffic, a merchant may reasonably want different pricing or different inventory exposure per channel. That is a commercial payments policy, and it needs setting before AI agents scale rather than after.
The summary Vladyslav Kolodistyi offers is deliberately unromantic. Agentic commerce rewards merchants whose data is complete, whose pricing is honest before the checkout, and whose payments stack does not treat agentic commerce automation as hostile. Everything else in the current discussion is speculation about which protocol wins. Further commentary from Vladyslav Kolodistyi on agentic commerce and payments infrastructure is published through his LinkedIn profile.
Top comments (0)