Agentic commerce is one of the most frequently named emerging priorities in recent ecommerce interviews, with remarkably few named tools behind it. Directors and VPs at global footwear brands, national retailers, large CPG manufacturers and mid-size DTC labels all put it in their top three goals for the next eighteen months. Asked what they are actually doing about it, they describe product-data cleanup, a question-and-answer widget on their product pages, and a conference they went to. That gap is not evasion. It is what the early phase of a category looks like from inside, and it is more instructive than the definition.
What is agentic commerce?
Agentic commerce is commerce in which an AI agent carries out part or all of a purchase on a person's behalf — finding candidate products, comparing them, and in the fuller version completing checkout without the person ever visiting the merchant's site.
The distinguishing test is whether the agent can act, not merely advise. An assistant that recommends three mattresses and hands you three links is AI-assisted shopping; the person still transacts. An agent that selects one and pays for it is agentic commerce, and it changes who owns the customer relationship at the moment of purchase.
That definition is clean. The word, as buyers actually use it, is not.
What is the difference between agentic AI and agentic commerce?
Agentic AI is the broader concept of software agents acting autonomously. Agentic commerce is specifically an AI agent acting in a purchasing process on a customer's behalf. Buyers also use “agentic” for on-site AI shopping assistants, which sit between the two.
Those three senses arrive in these interviews attached to three separate programmes, run by different teams, with different budgets and different time horizons. Pulling them apart matters because a stated agentic priority frequently turns out on inspection to be the first of them rather than the third.
Agentic AI, internally
Software agents automating the company's own workflows — content production, campaign operations, analytics, contact-centre handling. Nothing customer-facing, nothing to do with how people buy.
The most commonly described sense across these interviews, and not restricted to retail — airlines, banks, health insurers, software firms and hotel groups all describe it.
AI assistants on the brand's own site
Generative question-and-answer on product pages, conversational site search, on-site shopping assistants. Real, shipping, measurable — and entirely inside the brand's own property.
Mid-size DTC and apparel brands especially, often as a conversion project that predates the vocabulary.
Agents transacting for the customer
An agent on a surface the brand does not own — an assistant, a social platform, a marketplace — that can discover, compare and buy. This is agentic commerce proper, and the subject of this page.
Platform and payments teams are building it. Brands in these interviews are mostly preparing for it rather than running it.
The conflation matters commercially, not just semantically. A brand that reads a competitor's “agentic” announcement as sense three, when it was sense one, will misjudge how far behind it is.
Is anyone actually buying agentic commerce software?
Not as a category. Agentic commerce does not appear among the use cases buyers in these interviews are in market for, and no buyer describes an RFP for an agentic commerce platform, a shortlist of them, or a budget line named after one.
The contrast next door is the sharpest signal in the data. Answer engine optimization and generative engine optimization both exist as categories buyers actively shop, with named vendors on named shortlists. Agentic commerce, raised in conversation far more often, has neither.
This is not a scepticism story. The buyers naming it are not hedging — a director at a global footwear brand calls agentic commerce their team's single most critical focus, and describes continuously evaluating the market for it while making clear there is no active evaluation to be in. Both halves of that are true at once, and that combination is the defining condition of the moment: genuine urgency with nothing yet to sign.
What are brands piloting for agentic commerce?
Four kinds of work recur. The striking thing about the list is not what is on it but where the money comes from: the recurring work is generally funded from budget lines that predate the category.
Making the product data legible to a machine
Catalogue titles, descriptions and FAQs rewritten for machine reading; product information management evaluated or replaced; feed management tightened. A mid-size DTC brand runs an AI catalogue-copy tool on trial across a small range specifically to align descriptions with search and answer-engine best practice. A large US retailer is evaluating a product information platform to fix the master-data gaps its order system leaves.
Funded from — PIM, product feed management, digital shelfPutting a conversational surface on their own site
Generative question-and-answer on product detail pages, and site search rebuilt to handle conversational queries. One apparel brand ran a Q&A widget on its product pages for a year, A/B tested it to a slight conversion lift, and is now re-shopping the function — the tools serving it are young enough that buyers have already had one disappear from under them.
Funded from — site search, CRO, customer serviceScoring the platform on whether it is ready
The quietest signal, and the most telling. Agentic readiness has started appearing as a reason buyers give for the score they put on their commerce platform — one beauty brand gives agentic readiness as one of the two reasons for the score it puts on Shopify. Payment gateway changes are being weighed the same way. No new category is created, but the criteria inside an existing one quietly change.
Funded from — nothing; it is a renewal criterionWatching
Conference attendance, vendor conversations held deliberately outside a sales cycle, and monitoring what the platforms ship. Several buyers name Shoptalk specifically as where they went to understand agentic commerce — though at least one is blunt that such events are poor for evaluating vendors and useful mainly for peer conversation.
Funded from — travelRead together, these describe preparation that is defensible whether or not agents arrive on schedule. Cleaner product data, better on-site answers and a platform that will not have to be replaced are worth having regardless. That is not an accident of caution; it is the kind of investment buyers can justify before agentic commerce becomes a standalone purchasing category.
Who is building agentic commerce infrastructure?
Platform and payments companies are building the transaction infrastructure for agentic commerce; brands are mostly preparing their existing stacks to participate.
The parts that are genuinely under construction are being built by those teams, not by brands — and the interviews with those teams read completely differently from the brand ones. Where a retailer describes preparation, a platform describes a shipping roadmap: native checkout inside its own surface, built with existing payment partners, with the merchant remaining the merchant of record and continuing to own fulfilment.
That arrangement is where the commercial tension sits. It leaves the brand holding the liability, the returns and the shipping, while discovery, comparison and increasingly the transaction itself happen somewhere the brand does not own and cannot instrument. Payment providers describe recruiting merchants onto agentic platforms as an active commercial programme — suggesting that merchant participation, rather than direct brand ownership of the surface, is one model taking shape.
Interoperability protocols — the standards that would let an agent transact across merchants without bespoke integration — are the piece most often assumed and least often named. Only a small number of buyers raise them at all, and those who do are the ones building infrastructure rather than the ones waiting for it.
What is holding back agentic commerce?
Three obstacles come up unprompted.
Nobody knows whether consumers will do this. A VP of ecommerce at a large US retailer names it directly: it is unclear how quickly people will adopt AI platforms as places to research and shop, and the pace of the technology is running ahead of the evidence about the behaviour. Investment sized for fast adoption is the part with real downside.
Nobody is measuring it. Buyers have started measuring the discovery half — tools exist for tracking how a brand ranks in AI answers, and the more advanced teams use them. Not one of these interviews describes measuring agent-driven activity itself: what an agent saw, whether it considered you, why it went elsewhere. A programme with no numbers attached is hard to fund past its first year.
The tooling that does exist was built for someone else's stack. A director at a global footwear brand describes the practical version of this: the vendors moving fastest on AI-driven commerce built for a single platform ecosystem first, and integrating them into a global enterprise stack is its own project. Being early is not much use if the early tools assume a shape your business does not have.
How does agentic commerce relate to AEO and GEO?
They are adjacent stages of the same customer journey, and some of the preparation overlaps. Answer engine optimization is about whether an AI system can find and accurately represent a product; agentic commerce adds the ability to transact. Structured product data supports both — but agentic commerce introduces further requirements around checkout, payments, fulfilment and interoperability.
The overlap is what makes the readiness work affordable, and it explains why agentic preparation keeps showing up inside product-data and search budgets rather than a line of its own: the structured product information an agent needs to evaluate a product is largely the information an answer engine needs to describe it.
But the overlap stops at the point of purchase, and everything past that point is new. Being findable does not make you buyable. A brand can be well represented in every AI answer in its category and still be unable to complete an agent-initiated transaction, because that depends on the platform, the gateway, the merchant-of-record arrangement and a protocol layer that has not settled. AEO readiness helps make a product legible to an agent; it does not make the product transactable.
The difference is also who bears the risk. Being absent from an answer costs a brand consideration. Being unbuyable by an agent costs it the transaction.
Want this read against your own stack?
Get my read →Should you invest in agentic commerce now?
For most brands, these interviews support investing in agentic readiness rather than in a dedicated agentic-commerce product.
The buyers furthest along are not spending on an agentic commerce product, because buyers in these interviews are not yet treating agentic commerce as a standalone software category. What they are doing splits cleanly into work that pays off either way and work that only pays off if agents arrive quickly — and they are doing the first and deferring the second.
Work that pays off either way: getting product data structured and accurate, putting a real conversational answer on your own product pages, and treating platform and gateway renewals as decisions with an agentic dimension rather than routine ones. Work worth deferring: anything sized for a volume of agent-driven traffic nobody can currently observe, and anything that assumes a protocol has settled when the buyers building on those protocols are still a small minority.
The honest position, and the one most of these interviews describe without quite saying it, is that agentic commerce is currently a real strategic problem and not yet a purchasing decision. Those are different things, and treating the first as though it were the second is the mistake available right now.
Common questions
What is agentic commerce?
Agentic commerce is commerce in which an AI agent carries out part or all of a purchase on a person's behalf — finding candidate products, comparing them, and in the fuller version completing checkout without the person visiting the merchant's site. It is distinct from AI-assisted shopping, where an assistant helps but the person still transacts. The distinguishing test is whether the agent can act, not merely advise.
What is the difference between agentic commerce and agentic AI?
They are different problems that share a word. Agentic AI is internal: software agents automating a company's own workflows, content production, analytics or contact-centre handling. Agentic commerce is external: an agent acting for the customer, outside the brand's control. In these interviews the internal sense is by far the more common one, which is why a stated agentic priority often turns out on inspection to be a process-automation project rather than a commerce one.
Are brands actually buying agentic commerce software?
Not as a category. Agentic commerce does not appear among the use cases buyers in these interviews are in market for, and no buyer describes running an RFP for an agentic commerce platform or naming a budget line for one. This is the sharpest contrast in the data: answer engine optimization and generative engine optimization both exist as categories buyers actively shop, with vendors on shortlists. Agentic commerce appears repeatedly as a strategic priority even though buyers in these interviews are not yet treating it as a standalone software category.
What are brands actually piloting for agentic commerce?
Four kinds of work recur, generally funded from budget lines that already existed. Product data: catalogue titles, descriptions and FAQs, product information management, and feed management. On-site conversational surfaces: generative question-and-answer widgets on product pages, and conversational site search. Platform readiness: choosing or scoring a commerce platform and payment gateway on whether they will support agent-driven checkout. And watching: conference attendance and vendor conversations without an active evaluation. None of these is an agentic commerce purchase.
Where does checkout happen in agentic commerce?
In the versions described in these interviews, checkout increasingly happens inside a platform or assistant surface the brand does not own, while the merchant remains merchant of record and continues to own fulfilment. The commercial question those teams are working through is whether that split holds — the brand keeps the liability and the shipping, while discovery, comparison and increasingly the transaction happen somewhere it does not control. Emerging interoperability protocols are named by only a small number of buyers so far.
How do brands measure agentic commerce?
They largely do not yet. Buyers describe measuring AI discovery and visibility — how a brand ranks and is represented in AI answers — but no buyer in these interviews describes measuring agent-driven product consideration or transactions themselves. That makes attribution and ROI difficult before the channel has established observable volume.
How is agentic commerce related to AEO and GEO?
AEO/GEO and agentic commerce are adjacent parts of the same shift. AEO/GEO focuses on whether AI systems can find and accurately represent a brand or product; agentic commerce adds the ability for an agent to transact. Structured product data supports both, but agentic commerce also requires checkout, payments, fulfilment and platform interoperability. The overlap is why readiness work is usually funded from existing product-data and search budgets; the difference is why being findable does not make a brand buyable.
Should you invest in agentic commerce now?
The buyers furthest along are not spending on an agentic commerce product, because buyers in these interviews are not yet treating agentic commerce as a standalone software category. They are doing product-data work that pays off whether or not agents arrive on schedule, and they are treating platform and gateway decisions as decisions with an agentic dimension. The two open risks are that consumer adoption is genuinely unknown, and that no buyer here describes measuring agent-driven activity at all. Work that only pays off if agents arrive quickly is the part worth deferring.
Get this research made for your stack
What agentic commerce should mean for you depends on your category, your platform, and how much of your product data you actually control. Do a 15-minute interview and get the version of this that applies to you: what peers in your category are piloting, what they funded it out of, and what they decided to wait on.
Get my personalized read — 15-min interviewNo password · your interview is anonymized before it ever informs a page like this one.