Your next customer may never visit your homepage.
They may never admire the hero image, read the About page, open six product tabs, or appreciate the tasteful animation somebody spent two weeks discussing in Slack. Instead, they may tell an AI agent, “Find me a Canadian-made option that fits these dimensions, works with the equipment I already own, can arrive before Friday, and has a return policy written by someone who has met a human being.”
The agent will compare products, specifications, prices, availability, delivery terms, reviews, policies, and whatever evidence it can find that the merchant is real. It may assemble the cart and complete the purchase without the customer ever entering the carefully constructed marketing funnel.
If the agent cannot understand the product, confirm the current price, verify inventory, interpret the shipping terms, or complete checkout cleanly, it will probably move on. There will be no bounce in Analytics, no abandoned cart, no heatmap showing where interest died. The sale will simply happen somewhere else.
That is agentic commerce, and the infrastructure for it is already being built.
Google’s Universal Commerce Protocol is designed to turn interactions in AI Mode and Gemini into direct purchases. OpenAI’s Agentic Commerce Protocol allows ChatGPT to ingest structured catalogues, understand inventory, surface products, and support checkout. Stripe has launched an Agentic Commerce Suite, while Shopify now distributes merchant products into AI channels with synchronized pricing and inventory. Availability still varies. Google’s implementation remains in an early-access phase, and OpenAI’s Instant Checkout currently requires approval, so we have not reached the point where every toaster can autonomously buy another toaster. The direction is no longer theoretical, however. The pipes are going in now.
Mental availability, physical availability, and now machine availability
Byron Sharp and the Ehrenberg-Bass Institute describe brand growth through mental availability and physical availability. Mental availability means the buyer can easily think of the brand in a buying situation. Physical availability means the brand is easy to find and buy when that situation arrives.
We discussed this in Most Marketing Fails Because It Only Talks to People Who Are Ready Today. Coca-Cola does not need every advertisement to make you instantly thirsty. It wants Coke to be sitting comfortably in your memory when you eventually stand in front of a cooler.
Agentic commerce adds another practical layer, which I would call machine availability.
Machine availability means the customer’s agent can discover the brand, understand the offer, determine whether it fits the request, establish the current price, confirm availability, interpret the policies, and complete an authorized transaction.
A product can be mentally available to the buyer and physically available in the warehouse while remaining functionally invisible to an AI agent. That happens when the catalogue is vague, the variants are inconsistent, the inventory feed is stale, the return policy is buried in prose, and the checkout requires three human guesses plus a prayer.
The agent does not care that everyone inside the company knows what “standard size” means. It was not at the meeting.
AEO just got a credit card
Our argument about SEO and AEO has been fairly consistent: search works best when a business becomes visible, credible, specific, and useful enough to belong in the answer. SEO has become harder because cheating has become less useful, and answer engines need clear evidence before they can confidently explain or recommend a business.
Agentic commerce carries that argument one step further.
AEO helps the system understand and cite you. Agentic commerce asks whether the system can buy from you.
That expands the work well beyond keyword research and a few extra schema properties. The agent needs operational truth. It needs to know which product is being sold, how variants relate, whether the advertised price agrees with checkout, whether the item is in stock, which addresses are eligible, how quickly the order can arrive, and what happens if the customer wants to return it.
OpenAI’s product-feed specification requires structured catalogue data with identifiers, descriptions, pricing, inventory, media, and fulfilment options. Its documentation recommends regular full snapshots with updates during the day so product changes remain current. Google likewise keeps Merchant Center central to UCP and tells merchants to maintain complete product feeds, brand assets, return policies, and business contact information.
This is where the old divide between “marketing data” and “operations data” starts to collapse. The product description may come from marketing, the quantity comes from inventory, the delivery date comes from fulfilment, the current price comes from commerce logic, and the warranty comes from a policy somebody last reviewed during the Obama administration.
The agent sees the combined result.
If those systems disagree, the agent may see a company that cannot reliably answer a basic purchasing question. Frankly, some human customers have been seeing the same thing for years. They were simply too polite to express it as a failed API response.
Your catalogue is becoming part of the product
Most companies still treat the product feed as an unpleasant file that gets uploaded after the real website is finished.
Agentic commerce makes the catalogue part of the product itself.
A human shopper can compensate for incomplete data. They can inspect photographs, open another tab, infer what the merchant probably meant, call the store, or make an emotionally questionable purchase because the colour looked nice.
An agent is more likely to eliminate uncertain options.
In this environment, missing product data does more than weaken the page. It can remove the product from consideration.
The catalogue needs stable identifiers, precise attributes, accurate variant relationships, current prices, real availability, delivery information, return conditions, and enough context to distinguish one offer from another. Google already supports merchant structured data for prices, availability, shipping, product variants, and return policies, while its Merchant Center feeds provide another route for sharing product information. Structured data on the website remains valuable, but agentic commerce adds feeds, protocols, APIs, and live transaction systems around it.
This is also why uploading a catalogue once and congratulating ourselves will not be enough. The product feed becomes a machine-facing promise about what the business can actually sell.
If the website says one price, Merchant Center says another, the ERP holds a third, and checkout invents a fourth, you do not have omnichannel commerce. You have four competing accounts of reality.
That problem belongs to enterprise integration as much as marketing. Inventory, pricing, catalogue management, ecommerce, fulfilment, and accounting have to agree often enough that software can act on the information safely.
The website now has two jobs
The arrival of agentic commerce does not make websites irrelevant. It makes their architecture more interesting.
A serious commerce website increasingly serves two audiences.
The human audience needs design, photography, explanation, positioning, reviews, evidence, reassurance, and the feeling that the company can be trusted with money. The machine audience needs structured facts, stable identifiers, predictable interfaces, live inventory, explicit policies, and a checkout flow that can be operated without interpreting a clever visual metaphor.
The human wants to understand the brand.
The machine needs to understand the transaction.
Both matter.
We made a similar argument in Your Website Is Not a Brochure. It Is Your First Business System. A modern website is already connected to lead capture, sales, support, analytics, search, forms, and internal operations. Agentic commerce extends that system to software customers acting with delegated human authority.
It also strengthens the argument in Fiverr Can Build You a Website. It Probably Cannot Build You Credibility. A cheap template may display a product grid. The more demanding work involves product modelling, system integration, structured data, catalogue feeds, live inventory, security, checkout, analytics, and maintaining enough flexibility that the business can still operate the site after the developer disappears into the mist.
The agent will not appreciate the cinematic homepage if it cannot determine whether the blue 14 mm fitting is compatible with the buyer’s equipment.
Barbarian.
Inventory accuracy is becoming marketing
This may be the most important practical change.
In ordinary ecommerce, stale inventory is an operational problem that eventually becomes a customer-service problem. In agentic commerce, it also becomes a discovery and conversion problem.
Google’s UCP documentation places responsibility for real-time inventory checks on the merchant. If an item becomes unavailable during checkout, the merchant’s API must return an out-of-stock response so the user can be shown the failure. OpenAI’s checkout model similarly leaves validation, fulfilment options, tax calculation, risk analysis, payment processing, and order acceptance in the merchant’s own systems.
The agent can only be as reliable as the systems beneath it.
This is where a business running on six spreadsheets and one heroic employee may encounter some friction. We wrote about that architecture in When Your Business Becomes a Distributed Spreadsheet Problem. If sales, operations, ecommerce, and accounting each maintain their own version of inventory, agentic commerce will not reconcile the truth through sheer artificial intelligence. It will expose the disagreement faster.
That makes inventory accuracy, product governance, and system ownership part of the marketing stack. The data that used to live quietly behind the website now determines whether the business can appear in a recommendation and survive checkout.
Marketing people may find themselves discussing warehouse synchronization and idempotency.
Developers may have to acknowledge that product language and brand trust affect transaction completion.
Everyone will be fine after a small adjustment period and several meetings where nobody admits this was the obvious conclusion.
The agent is a privileged software customer
An AI shopping agent behaves less like a browser visitor and more like an external software client acting under delegated authority.
That requires identity, authorization, scoped permissions, payment controls, order records, error handling, fraud detection, reconciliation, refunds, and audit trails.
OpenAI’s Agentic Commerce Protocol keeps the merchant in control of the transaction. ChatGPT gathers buyer and fulfilment information, calls the merchant’s checkout endpoints, and presents the resulting state, while the merchant validates the order, determines shipping and tax, evaluates risk, charges through its payment provider, and accepts or declines the transaction. OpenAI’s delegated payment design uses a one-time request with a maximum amount and expiry rather than handing an agent a reusable credit card and wishing it well.
Google’s architecture uses UCP for the shopping and checkout lifecycle, while its Agent Payments Protocol can provide cryptographically signed cart and payment mandates. The cart mandate records what the merchant offered, while the payment mandate records what the buyer authorized. That creates evidence about what was purchased, at which price, and under whose authority.
This is where the phrase “AI shopping” becomes too cute for the actual engineering problem.
What happens when inventory changes after recommendation? Can an agent substitute another variation? How much may it spend? What if shipping changes the total? Who owns an incorrect purchase? How is a partial refund handled? What prevents a duplicated request from creating two orders? How does customer support distinguish an agent-mediated purchase from an ordinary checkout?
OpenAI’s production checklist includes tests for address handling, recalculated shipping totals, tokenized payments, order events, out-of-stock errors, payment declines, signed webhooks, TLS, and compliance. The boring details have arrived, as they always do when a demo becomes software.
This work belongs to custom software development, business analysis, and managed IT and security alongside ecommerce design. The AI is only one participant in a much larger transaction system.
It will get most of the keynote slides anyway.
Agentic commerce is a supply-chain problem wearing a marketing hat
The phrase “agentic commerce” makes this sound like a new shopping interface.
The deeper change reaches further into the business.
An agent can recommend a product only if it can understand the catalogue. It can promise delivery only if fulfilment data is trustworthy. It can complete checkout only if pricing, tax, payment, fraud, inventory, and order systems respond correctly. It can support the customer only if order status, returns, and merchant policies are available after the purchase.
Google’s current UCP roadmap includes multi-item carts, account linking, loyalty capabilities, tracking, and returns. Shopify says products in its Agentic Storefronts remain synchronized across AI surfaces with real-time pricing and inventory. Stripe’s technical guidance tells merchants to think about agent discovery, catalogue legibility, nonhuman traffic, internal organizational alignment, and AI governance.
So yes, this is marketing.
It is also catalogue governance, supply chain, integration, security, customer service, and software architecture.
The new shelf space may live inside an AI agent, but the product still has to leave the warehouse.
Search visibility now needs transactional legibility
A business can be visible to an AI model without being transact-able.
The model may understand that the company exists. It may even cite or recommend the product. The transaction can still fail because the agent cannot establish a current price, identify the correct variation, calculate delivery, or interpret the return conditions.
I think of this as transactional legibility.
A transactionally legible business publishes enough precise, current, structured information that an authorized agent can move from interest to purchase without inventing missing facts.
That does not replace SEO or AEO. It sits on top of them.
Search engine optimization helps the business become discoverable and understandable. Structured data helps search systems interpret products and offers. Brand content and reviews build trust. Catalogue feeds keep commercial information current. APIs and commerce protocols allow the agent to act. Internal systems make sure the order is real.
This is also where crawler access and infrastructure policy matter. Stripe’s agentic-commerce guidance advises businesses to verify that their robots.txt and firewall settings do not accidentally block legitimate AI discovery while still protecting expensive or sensitive dynamic routes. Opening everything to every bot would be unwise. Blocking every agent would be a fascinating way to achieve complete machine privacy and zero machine sales.
The work requires deliberate choices about which content, catalogue data, and transaction interfaces should be public, which agents should be trusted, and which parts of the system should remain inaccessible.
The agent needs a front door.
It does not need the keys to the boiler room.
Which businesses should care first?
Retailers and ecommerce merchants with large or frequently changing catalogues should pay attention immediately. Product variation, stock status, shipping, returns, and price synchronization will directly affect whether agents can recommend and sell their products.
Manufacturers, distributors, and industrial suppliers may have an even more interesting opportunity. Their products often require precise compatibility, dimensions, certifications, materials, model numbers, or technical constraints. A well-structured catalogue can help an agent navigate complexity that would exhaust an ordinary shopper. A vague catalogue can make an excellent product effectively disappear.
Restaurants and food-ordering businesses should also watch the space. Google is already extending UCP toward pickup and delivery ordering. Menus, modifiers, availability, service areas, timing, and fulfilment all have to become understandable to software.
Service businesses are less mature in the current protocols, but the same direction is visible. Booking, qualification, service geography, price ranges, appointment availability, and eligibility rules can all become agent-readable. Schema.org already treats services as products in its broad model, although a machine-readable description does not automatically create a safe autonomous booking process.
My expectation is that products come first because they fit cleaner transaction models. Services will follow where the work can be sufficiently structured. A haircut appointment is easier to express than a custom ERP migration. Somewhere between those two, the agent will wisely decide it needs a human.
Shopify helps, but it cannot repair the business underneath
Shopify is moving aggressively. Eligible merchants can manage Agentic Storefronts from the Shopify admin, with products distributed to supported AI channels through Shopify Catalog. Shopify says pricing and inventory remain synchronized, and it now offers an Agentic plan for brands on legacy or custom platforms that want access to AI channels without moving their entire online store.
That can remove a significant amount of integration work.
It does not repair bad source data.
If product names are inconsistent, variants are modelled badly, inventory is unreliable, and return policies were written during an argument in 2018, syndicating the catalogue more widely mainly distributes the confusion with impressive efficiency.
The same principle applies to WooCommerce, BigCommerce, Wix, custom ecommerce systems, and internal product databases. Platforms will increasingly provide agentic channels. Businesses still need clean data, sensible processes, and systems capable of keeping commercial claims synchronized.
PANDAROSE has built WooCommerce sites with custom PHP integrations and larger custom software platforms. That experience matters here because agentic commerce lives precisely where website development, catalogue data, operational software, and integrations meet.
This is not a plugin problem wearing a futuristic hat.
What an Agentic Commerce Readiness Review should examine
Most businesses do not need to implement UCP or ACP tomorrow. The current programmes are still developing, availability varies, and the standards will continue to evolve.
They do need to understand whether their existing ecommerce system is capable of participating when the channel becomes relevant.
A useful review should examine the product catalogue, variant structure, identifiers, structured data, Merchant Center configuration, inventory source, pricing logic, shipping rules, returns, customer-support information, checkout architecture, payment controls, analytics, crawler policy, and the systems responsible for keeping all of it current.
The more revealing questions are usually operational.
Which system owns price? Which system owns inventory? How quickly do changes propagate? Can the company identify a product consistently across the ERP, website, warehouse, and feed? Does checkout recalculate tax and delivery from current information? Are policies written clearly enough for a machine to apply without becoming creative? Can the business trace an agent-mediated order through support, refund, and reconciliation?
This is business analysis applied to a new sales channel. The protocol comes later. The first job is determining whether the business has a coherent commercial truth to expose.
If the answer currently lives in three spreadsheets and Brenda’s memory, the agentic-commerce roadmap has kindly identified an earlier project.
Where PANDAROSE fits
Agentic commerce lands directly in the overlap of Build. Support. Market.
Website development provides the human storefront, product experience, structured content, and commerce platform. SEO and AEO help search systems and buyers discover and understand the business. Custom software and enterprise integration connect catalogue, inventory, pricing, fulfilment, and internal systems. Managed IT protects the identities, infrastructure, integrations, and operational environment beneath the transaction.
This is why the topic is more interesting than another “AI will change shopping” prediction.
We understand the human storefront and the machine interface. We understand that a beautiful website cannot rescue stale inventory, while a perfect feed cannot build human trust by itself. We understand that the agent needs access to enough information to transact without receiving enough access to become an incident report.
Most importantly, we understand that this work has to survive production.
If your business sells online and you want to understand whether AI agents can discover your products, interpret your catalogue, trust your data, and complete a transaction safely, talk to PANDAROSE. We can review the website, catalogue, structured data, inventory, checkout, integrations, security, and internal systems before a new sales channel exposes every disagreement at once.
Your next customer may never visit your website.
Their agent will still decide whether you make the shortlist.
Frequently asked questions about agentic commerce
What is agentic commerce?
Agentic commerce uses AI agents to help shoppers discover, compare, select, and purchase products or services. Depending on the platform and merchant integration, the agent may move from recommendation into checkout without requiring the customer to leave the AI conversation.
Is agentic commerce available now?
Yes, although availability remains uneven. Shopify offers Agentic Storefronts and a global Agentic plan, Stripe is deploying its Agentic Commerce Suite, Google’s UCP programme is in early access, and OpenAI’s ACP is open for development while Instant Checkout remains limited to approved partners.
Does my business need UCP or ACP immediately?
Probably not immediately. Most businesses should first clean up product data, inventory, pricing, structured data, shipping, return policies, and internal integrations. Protocol implementation is difficult to justify when the underlying systems still disagree about what is being sold.
Will agentic commerce replace websites?
No. Websites remain important for human trust, brand experience, product research, customer service, and ordinary ecommerce. Agentic commerce adds machine-facing discovery and transaction surfaces alongside the human website.
How does agentic commerce affect SEO and AEO?
SEO and AEO help agents discover and understand a business. Agentic commerce adds the need for current catalogue data, machine-readable policies, live inventory, and compatible checkout flows. Visibility becomes more valuable when the agent can also act.
What is machine availability?
Machine availability is a practical extension of mental and physical availability. It describes whether an AI agent can discover, understand, evaluate, and transact with a business using reliable structured information and supported interfaces.
Is agentic commerce secure?
It can be, provided the architecture uses scoped authorization, validated checkout states, secure payment tokens, signed messages, fraud controls, audit trails, and clear human authority. The agent should receive only the access needed for the transaction.
References and further reading
- Google: Universal Commerce Protocol
- Google: UCP frequently asked questions
- Google: Secure agent commerce with AP2 and UCP
- OpenAI: Agentic Commerce Protocol
- OpenAI: Agentic Commerce key concepts
- OpenAI: Agentic commerce production checklist
- Stripe: How to prepare for agentic commerce
- Shopify: Agentic commerce for merchants
- Google: Product structured data
- Ehrenberg-Bass Institute: Measuring mental and physical availability
