For years, adding a commerce channel meant a familiar project: decide the assortment, map the catalogue, configure checkout, connect orders and measure the result.
AI shopping is beginning to look remarkably similar — except some merchants are being enrolled before their operating model has caught up.
The important question is no longer whether your products can appear inside an AI assistant. It is who decides what that assistant can see, sell, collect and promise on your behalf.
Shopify made that question more immediate this month.
On September 8, Meta appeared as another AI channel inside Shopify’s Agentic Storefronts. Merchants can make products available to Meta and, where eligible, enable Shopify-powered direct checkout on Meta surfaces.
Meta is not an isolated integration.
Shopify currently documents Agentic Storefronts across ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, with channel-specific discovery and checkout behaviour.
That starts to look less like an AI feature.
It starts to look like channel management.
Agentic commerce is becoming a distribution problem
The first wave of agentic-commerce discussion focused on whether an AI could find the right product.
That still matters. Product data has to be structured well enough for another system to understand what the merchant sells.
But distribution introduces a different set of decisions.
Which channels receive the catalogue? Which products should appear? Can the channel merely recommend the product, or can it transact? Which customer information reaches the channel? Where does checkout happen? Who owns the order? Who handles returns? How is attribution measured?
These aren’t model questions.
They are commerce operating-model questions.
Shopify’s implementation makes the distinction unusually visible.
Under Shopify’s current default configuration, available AI channels can receive product information and eligible channels can use direct checkout. Merchants who disable the default “Allow Shopify to manage for me” setting can instead manage catalogue access, automatic enrolment into future channels and checkout behaviour themselves.
That is effectively a primitive control plane for agentic distribution.
Our view is that enterprise merchants should treat it with the same seriousness they already apply to marketplaces, affiliates, social commerce and retail media.
Discovery and transaction are different permissions
One detail is particularly important.
Not every AI channel operates the same way.
Shopify currently describes ChatGPT primarily as a discovery channel: shoppers can find products through ChatGPT, but complete the transaction using the merchant’s online-store checkout, including through an in-app browser.
Eligible merchants using Google AI Mode/Gemini, Microsoft Copilot or Meta can enable Shopify-powered direct checkout that completes inside the AI experience. Availability and eligibility vary by channel.
Those two models should not be treated as equivalent.
A discovery channel needs access to enough product information to recommend accurately.
A transaction channel also becomes involved in checkout and customer information.
Shopify says AI channels with direct checkout can receive order details including a customer’s name, email, phone number and physical address as necessary for the transaction. It also states that the AI channels do not gain access to the merchant’s general customer database or complete order history.
That boundary is useful.
It is also exactly the kind of boundary enterprises should document deliberately rather than discover during an incident review.
The useful architecture separates six controls
We think AI commerce becomes much easier to reason about if merchants stop treating “agentic storefront” as one permission.
There are at least six separate controls.
1. Catalogue access
Can this channel consume authoritative product information?
That sounds simple until the assortment includes restricted products, region-specific inventory, B2B-only items, discontinued SKUs or products whose descriptions make sense only in the context of the storefront.
Shopify allows merchants to control Shopify Catalog access for individual supported AI channels.
Enterprise architectures should preserve that distinction even when another platform implements the plumbing differently.
2. Channel enrolment
Should every newly supported AI destination automatically become an active commerce channel?
For a small merchant, automatic distribution may be useful.
For an enterprise brand operating across markets, legal entities and channel agreements, automatic enrolment can deserve a governance decision.
Shopify exposes this explicitly through its auto-enrolment control.
That is more consequential than the setting initially appears.
A new AI partner is not merely another place where a product card renders. It may eventually become another source of orders, customer interactions and operational exceptions.
3. Transaction authority
Can the AI channel recommend, hand off or complete the purchase?
These are three different levels of authority.
Commerce teams already understand this pattern elsewhere.
A price-comparison engine can refer traffic. A marketplace can originate an order. A call-centre agent may be allowed to modify an order within defined limits.
AI channels should be modelled with the same clarity.
“Available to AI” is too coarse a status.
4. Data disclosure
What information crosses the channel boundary before and after purchase?
Product titles and availability are one class of data. Customer identity, address and order information are another.
The minimum data required should follow the capability being granted.
Discovery does not require the same information as fulfilment.
This is less glamorous than an AI shopping demo, but it is what privacy, security and architecture teams will eventually ask.
5. Attribution and economics
Who gets credit for the order? What did the channel cost? What happened to conversion? Can the merchant distinguish AI-referred checkout from direct checkout?
Shopify exposes sales, orders, sessions and online-store conversion reporting for Agentic Storefronts, although its documentation notes that some analytics are not yet available for headless stores.
There is another operational wrinkle: Shopify states that third-party analytics pixels do not fire when direct checkout completes on a Meta surface because the checkout did not occur on the merchant’s online store.
That is exactly the kind of small implementation detail that becomes a large attribution meeting later.
Commerce teams should decide their measurement model before comparing AI-channel performance with conventional storefront traffic.
6. Post-purchase responsibility
The transaction may move.
Responsibility largely does not.
For Meta orders, Shopify says the merchant remains merchant of record and continues to manage fulfilment, returns and customer service.
That makes OMS and post-purchase integration central to the channel design.
If an AI channel creates an order, the warehouse still needs to fulfil it. Fraud controls still need to evaluate it. Customer service still needs to understand it. Returns still need a policy. A marketplace seller may still need a sub-order (as managed in Emporio).
The exciting new interface does not remove the boring old invariants of commerce.
This becomes harder outside a single-platform stack
Shopify has an advantage here because it can coordinate catalogue, channel settings, checkout and orders within its own operating model.
Enterprise commerce is often less tidy.
The product may originate in a PIM. Price may come from ERP. Availability may come from OMS. Reviews may sit in a separate platform. Checkout may be commerce-platform native. Orders may route through another orchestration layer. Marketplace offers may introduce seller-specific inventory and fulfilment. Customer identity may live in CRM or CDP.
In that environment, the merchant needs to know which system remains authoritative for each promise an AI channel makes.
That is where “connect the LLM to our APIs” stops being an agentic commerce architecture.
A useful rule: AI can distribute the promise, but the system of record must own it
This is the principle we would use in an enterprise design review.
An AI experience can communicate that a product costs ₹4,999. The pricing service should own whether ₹4,999 is valid.
An AI experience can tell a shopper an item arrives tomorrow. The inventory and fulfilment systems should own whether that promise is possible.
An AI experience can initiate an order. Checkout and payment systems should own transaction integrity.
An AI experience can explain a return. The order system and return policy should own what is actually permitted.
The distinction becomes even more important when an AI interface is operated by somebody other than the merchant.
The channel may change. The source of truth should not.
Marketplace teams have seen this movie before
There is a useful precedent for commerce leaders.
Marketplaces forced brands to answer many of these same questions.
Which assortment goes to Amazon or another marketplace? Who owns the catalogue? Which price wins? Where does inventory live? Who receives the customer relationship? Where does the order enter? Who fulfils it? Who owns returns? How do we reconcile settlement?
AI commerce will not reproduce the marketplace model exactly, but the operating discipline is familiar.
The difference is speed.
Adding a marketplace has historically been a deliberate programme.
AI channels can appear as a setting in an admin console.
That makes governance more important, not less.
What should a commerce leader ask this quarter?
Not “Do we have an AI shopping strategy?”
Start with something more concrete.
Which AI channels can access our catalogue today? Do not assume the answer.
Which can complete a transaction versus only refer one? Document the difference.
Who approves a new channel? Commerce? Marketing? Architecture? Security? Legal? Someone should own the decision.
Which system owns every promise exposed through the channel? Catalogue, price, promotion, inventory, delivery, returns and subscription rules should each have an authoritative source.
How will we identify and measure these orders? Do this before the first executive dashboard asks whether agentic commerce converts.
Can we switch a channel off without breaking the rest of commerce? That is a useful test of architecture quality.
The next channel strategy meeting may include machines
Last week’s agentic-commerce discussion was largely about whether commerce infrastructure is ready for AI shoppers.
The next question is more operational.
When AI platforms become places where products are discovered and transactions begin or complete, merchants need the same discipline they use for every other meaningful sales channel.
Assortment. Permissions. Commercial rules. Data boundaries. Attribution. Fulfilment. Returns. Ownership.
The interface happens to be conversational.
The commerce responsibilities are remarkably familiar.
And that may be the most useful way to think about agentic commerce right now: not as a magical replacement for ecommerce, but as a rapidly expanding set of channels that need an explicit operating model, similar to the governance boundaries we discussed around Salesforce AIforce.
This article reflects Tailoredd’s architectural interpretation of the current direction of agentic commerce. Channel availability, eligibility and capabilities change quickly; implementation decisions should be validated against each platform’s current documentation.
