Agentic Commerce: What Store Owners Should Do Now

Agentic commerce deserves a place in your quarterly plan: eligible Shopify stores can already reach shoppers in ChatGPT with $0 in additional channel fees, while normal payment processing still applies. Start with accurate product data and a supported checkout path; fund custom integrations when the extra contribution can cover the work. Shopify's current ChatGPT channel requirements establish that starting point.

Verified against the makers' announcements and live documentation on 23 October 2026. The decision turns on where customers can discover your products, where they can pay, and whether the resulting orders add profitable demand.

What Is Agentic Commerce?

Agentic commerce means an AI agent finds, compares and pays for a product on a customer's behalf inside a chat or assistant, with the customer's authorisation. An agent is software that can take actions toward a goal. Some shopping channels currently handle only discovery and send the buyer to the merchant's website to finish.

A recommendation answers, “Which lamp suits this room?” A buying agent can also check the chosen variant, delivery options and total, then submit an authorised purchase. OpenAI's checkout specification describes that connection between the assistant and the merchant's order system. OpenAI's checkout documentation makes the merchant responsible for the transaction behind the interface.

Agentic Commerce Examples

Consider a homewares customer looking for a compact bedside lamp that can arrive before a house move. The assistant needs dimensions, stock, delivery options and the total price to make a useful selection. “Beautiful lighting for modern living” cannot answer those constraints.

There are different levels of help. An assistant might recommend a lamp and open your storefront. It might display a checkout inside the conversation and ask the customer to approve it. A delegated buying system might operate under permission the customer gave earlier. Those experiences require different integrations and controls; seeing your product in a chat proves only that it was discoverable.

For a founder, the immediate question is concrete: can the assistant identify the right product and carry a valid order into your business?

What Changed for Merchants

The change is that assistants have documented connections to catalogs, checkout systems and payment authorisation. Merchants can assess an integration against published requirements instead of buying an unspecified promise of “AI visibility.”

OpenAI introduced Instant Checkout and the Agentic Commerce Protocol on 29 September 2025. Stripe's same-day announcement describes a buyer-approved purchase inside ChatGPT and the merchant's own systems handling the order. Stripe's original ACP announcement establishes the launch. OpenAI's subsequent product-discovery announcement shifted the emphasis to merchant-owned checkout; the original launch is not today's onboarding promise.

Google announced Agent Payments Protocol on 16 September 2025, followed by Universal Commerce Protocol on 11 January 2026. The first addresses authority to pay; the second addresses the broader shopping transaction. Google developed UCP with Shopify and other retail partners. Google's AP2 announcement and UCP announcement describe different jobs.

The practical result is channel-specific access. A protocol being public does not mean your store has been admitted to every assistant that uses it.

Two Doors for Shopping Agents: Walmart and Amazon

Open your catalog deliberately; open purchasing only through a route whose permissions you can enforce. A customer wanting to use an agent and a retailer agreeing to serve that agent are separate decisions.

Since October 21, 2026, Walmart has offered a concrete example of sanctioned access: it announced Walmart Agent Gateway, an MCP server for shopping agents, and Agent Profiles, planned for November 2026. MCP, or Model Context Protocol, connects agents to external tools. The announcement describes a beta for companies and developers. Walmart's announcement separates that beta from the forthcoming customer controls.

The Gateway connects agents through Sparky, Walmart's shopping assistant, with current product and fulfilment data and a route from finding items to checkout. It can help identify a customer's usual items without handing the agent the customer's broader purchase history. Companies and developers can request access, but the beta is invitation-only, and features may vary by participant. This is controlled access to Walmart's shopping systems, not a checkout service that an independent merchant can install. Walmart's Gateway page explains the access boundary.

Agent Profiles will separate each agent's purchases inside the customer's account. First use through the Gateway creates the profile; the customer links it by signing in with Walmart, which keeps that sign-in private. The customer can require approval for every purchase or set a spending cap, and delete the profile at any time. Approval is the default, and Walmart can still require it within the cap. Agents cannot receive card details or change payment methods; Walmart's usual returns process applies. Walmart's control details describe the planned experience.

The technical boundary matters: Walmart's terms require human approval at checkout outside the Gateway; autonomous checkout is reserved for compliant Gateway transactions. Its scoped tokens bind access to a particular transaction context. A spending permission is not unrestricted account access. Walmart's agent-access terms define that distinction.

Amazon illustrates the other door: refusing third-party purchasing agents. TechCrunch reported Amazon blocking third-party shopping agents on October 21. Amazon's own statement about Perplexity says external purchasing agents must operate openly and respect a retailer's decision to participate; it requested Amazon's removal from Comet. Amazon's own shopping features do not imply permission for outside agents to use its checkout.

On October 22, Axios reported that major US banks were restricting third-party agents using customers' login credentials.

For your store, publish an agent policy that answers these questions before enabling purchases:

• Who may enter? Name approved integrations, require identifiable agents, and distinguish catalog browsing from account and checkout access.
• Who authorises the order? Require customer confirmation by default. If you permit delegated purchases, define the spending cap, its time period, eligible products and when fresh approval is required.
• What may the agent access? Use scoped credentials or payment tokens. State whether customer-password sign-in is prohibited and how permission is withdrawn.
• Who handles a mistake? Record the agent, customer permission, cart and final total. Name the owner of support, returns, refunds and disputed orders, and explain how access can be suspended.

A homewares store can accept catalog referrals now while keeping unfamiliar agents out of customer accounts. Enable a purchase integration when the platform can identify the agent, enforce the customer's permission and produce an order your staff can support. Publishing a policy is useful only when the integration enforces it.

Agentic Commerce Companies: Who Does What?

Start with the customer-facing channel, then identify the catalog, checkout and payment requirements behind it. These programmes occupy different parts of a purchase.

The maker-specific requirements follow. Treat the table as a map of responsibilities, not a list of interchangeable software subscriptions.

OpenAI Instant Checkout and ACP: Check Access Before Building

OpenAI now prioritises product discovery and merchant-owned checkout, while ACP remains an open integration standard. Its current merchant page says it is moving away from standalone Instant Checkout. Start with a product feed and your own checkout; a still-published checkout specification is not evidence that a new store can enable in-chat payment. OpenAI's getting-started guide limits direct feed onboarding to approved partners and provides the application route.

Agentic Checkout: The Merchant Requirements

For a direct integration, start with the catalog. OpenAI's stable file-upload schema requires 9 basic fields: item ID, title, description, product URL, brand, seller name, image URL, availability and price. Each purchasable variant needs its own record. Checkout must be enabled separately; setting a product's checkout flag does not grant access. Checkout also requires public seller privacy and terms links. OpenAI's product-feed specification defines those boundaries.

The onboarding guide recommends a daily full catalog with updates during the day, and requires products to comply with its prohibited-products policy. A fresh feed is particularly important for a small apparel brand: the agent must see the availability of the selected size, not the availability of the parent product. OpenAI's integration guide covers delivery methods and policy exclusions.

Where OpenAI separately enables a checkout integration, its published specification says your commerce system maintains the checkout, calculates shipping and tax, processes payment and sends order updates. It must handle authenticated requests, retries and integration checks. OpenAI allows the merchant's existing payment service provider, or PSP, with delegated-payment support where applicable. The checkout specification is the implementation contract.

The founder's acceptance test should be an order that reaches fulfilment correctly, including the selected variant, final total and customer details. A product appearing in a conversation is an earlier milestone.

Stripe Agentic Commerce: Payment Credentials Are One Part

Stripe co-developed ACP with OpenAI, but ACP does not require every merchant to replace its commerce backend or use one exclusive payment processor. Stripe describes the seller retaining the customer relationship and remaining merchant of record, meaning the business responsible for the sale. Stripe's ACP announcement explains that architecture.

Stripe's Shared Payment Tokens are in public preview. They pass scoped payment authority without exposing the underlying credentials to the agent. Current documentation binds the grant to the seller's Stripe profile with amount and expiry controls, and supports Stripe processing or a compatible third-party processor through its Resolve API. Availability covers the US, Canada and selected European countries. Stripe's current token documentation supplies the implementation requirements.

For a merchant using another PSP, ask it to confirm the supported token and settlement path before commissioning development. The documented seller setup still requires a Stripe account and profile. A token solves a credential handoff; your catalog, stock checks and order handling still need to work.

Google AP2 and UCP Solve Different Problems

Google's Agent Payments Protocol establishes permission to pay; Universal Commerce Protocol connects the wider shopping journey. Google's current FAQ explicitly places AP2 as a possible payment layer within UCP. Google's UCP FAQ makes that separation clear.

AP2 uses signed records called mandates. An Intent Mandate records the buyer's instructions or constraints; a Cart Mandate records the agreed purchase. Its design covers both present-time approval and purchases under earlier, bounded permission. Google also documents an A2A x402 extension for agent crypto payments. That specialised payment route is not a requirement to put an ordinary card-based store on Google. Google's AP2 documentation describes the mechanisms.

A merchant evaluating AP2 should ask its agent or payment partner how it receives and validates authorisation evidence. There is no product feed you can submit to AP2 to acquire shoppers.

Google UCP is the commerce integration to examine for Google AI Mode and Gemini. Google's guide now offers native checkout and a cart path that transfers the shopper to the merchant's website. Going live requires Google's approval; the native route includes a published integration profile, authenticated endpoints, a payment handler and order updates. Google's implementation overview is the starting point.

For a custom retail integration, the eligibility work has three parts:

• Catalog and policies: Merchant Center must be in good standing with products approved for free listings. Supply return policies and customer support details; map feed identifiers to the checkout's product identifiers.
• Product permission: Google's feed uses native_commerce(checkout_eligibility) to opt a product into checkout. Missing or false means it is not eligible. Product restrictions include subscriptions, personalised goods, used or refurbished items, and final-sale goods.
• Payment readiness: The Google Pay handler needs a compatible payment path. The gateway route sends the encrypted payment payload to your PSP. Production use requires the appropriate Google Pay & Wallet Console access and merchant ID.

These requirements come from Google's Merchant Center preparation guide and Google Pay handler documentation.

A made-to-order furniture business should check product restrictions before spending on native checkout. A cart handoff may fit its consultation or configuration process better. A stocked homewares retailer with straightforward shipping has a clearer candidate for a direct-purchase trial.

Shopify Agentic Storefronts: Check the Channel and Product

Shopify Agentic Storefronts gives eligible stores a managed route into supported AI channels. The feature is active by default for eligible Shopify stores, with controls under Sales channels > Agentic. The Agentic plan for external stores differs: direct checkout starts deactivated. Shopify says merchants keep the customer relationship and receive order attribution in the admin. Shopify's storefront overview explains the channel model.

For an existing Shopify online store, Catalog eligibility normally means Starter or higher, a public storefront, a product title, at least 1 image, a positive price and a published product URL. Permitted products and account standing also matter. Shopify documents an Agentic-plan exception for external stores. Catalog requirements provide the full conditions.

ChatGPT currently sends Shopify shoppers to the store's own checkout, inside an in-app browser or a new tab. Stores can be outside the United States but must sell to US customers. Catalog eligibility, the agentic supplemental terms and completed terms, privacy, and return/refund policies are required. Existing checkout payment methods remain available. Shopify's ChatGPT help page describes this referral flow.

Google direct checkout through Shopify has a different boundary: the store and customer must be US-based. Products must be in Merchant Center, with Catalog eligibility, required policies and supplemental terms satisfied. The Google & YouTube channel can sync products and shipping, or the merchant can use another Merchant Center feed route. Normal processing fees remain, with no additional selling fee listed for this channel. Shopify's Google channel requirements are the authority for this route.

Shopify also exposes agent discovery files, including /agents.md and /llms.txt, automatically. Those files describe the store; they do not replace Shopify Catalog's product feed. Catalog Mapping can connect custom product fields to the data distributed to agents. Shopify's product-discovery documentation distinguishes the two.

If you are deciding whether Shopify fits the business, use the Shopify review. If you are building a browser-based shopping agent, the separate Shopify WebMCP checkout guide covers that implementation path.

Card Networks: Verify the Agent and Payment Path

Visa and Mastercard are building trust and payment infrastructure, not a substitute for merchandising or channel onboarding. For a store owner, the useful next conversation is with the payment provider and the platform that will deliver the transaction.

Visa Agentic Commerce: Intelligent Commerce and Trusted Agent Protocol

Visa Intelligent Commerce is Visa's portfolio for agent-led transactions. Its public page explicitly describes deployment as ongoing, so the advertised portfolio should not be read as every merchant's available feature set. Visa's programme page states that qualification.

Visa's Trusted Agent Protocol lets a merchant or infrastructure provider verify signatures from approved agents. Its getting-started guidance centres on validating the request using the agent's public key. Visa's merchant documentation describes that work.

Ask your platform or site-protection provider whether it performs this verification. A bot identifying itself as a shopping assistant is not sufficient evidence to relax checkout controls.

Mastercard Agent Pay

Mastercard Agent Pay describes registered agents, tokenised payments and customer controls over what an agent can purchase. Those are payment-control claims made by Mastercard's own announcement.

Mastercard's September 30, 2026 announcement adds trust and intelligence services to Agent Pay. Its first probability score for identifying agent-initiated transactions is rolling out for testing in the US. That is a testing status, not universal merchant availability.

Ask your PSP or acquirer, the provider connecting your business to card acceptance, which Agent Pay flows it supports, how it identifies them and what merchant terms apply. The public overview does not establish a universal store-onboarding checklist. Neither network programme supplies the product feed that makes a lamp discoverable in ChatGPT or Google.

What Changes for Founders, Marketers and Builders

Product evidence becomes a distribution asset: the assistant needs enough reliable information to match your offer to the buyer's constraints. Your brand still needs demand, but better creative cannot repair a missing size, ambiguous delivery promise or stale stock record.

For founders and operators: assign ownership of catalog accuracy, channel eligibility and profitable orders. A homewares founder should know who fixes a mismatched delivery promise and who handles the customer's return. The assistant's involvement does not remove that operational work.

For marketers: turn shopper questions into product facts. Replace unsupported superlatives with dimensions, materials, compatibility, care instructions and clear use cases. Keep price, availability and delivery information consistent across the page and feed. Shopify specifically recommends detailed product information, identifiers, variants and current policies to improve matching. Shopify's catalog optimisation guidance supports this approach.

Reviews should contribute evidence about the product customers receive. Keep product reviews distinct from seller-service reviews, retain their context, and fix recurring complaints rather than manufacturing praise. OpenAI's feed specification separates optional product review aggregates from store review aggregates, which require additional setup. A review count is not a guaranteed ranking lever. OpenAI's review-data specification provides the field-level rules.

For builders: preserve product identity, authoritative totals and a reliable order record across the handoff. Start with the commerce platform's existing integration when it meets the requirement. A custom route earns its cost when it serves a supported buying flow that the managed route cannot handle.

Allocate the first preparation budget to product data, feed maintenance and measurable checkout completion. Keep paid acquisition as a separate experiment with its own acquisition-cost target. The ChatGPT ads guide covers that buying decision; ad spend does not substitute for proving your catalog and checkout are eligible.

The Numbers: Budget Against Incremental Contribution

Judge the project by contribution from extra demand or genuinely avoided acquisition cost. A channel-attributed sale can be an existing customer taking a different route.

For a US homewares store, build the case from your own order and cost records:

• Start with average order revenue, then subtract product cost, fulfilment, expected returns and payment processing to find contribution before acquisition.
• Record catalog preparation hours at your actual staff or contractor rate, plus the upkeep budget for the quarter.
• Confirm any channel fee before calculating contribution. Shopify's current ChatGPT referral route adds no selling fee; another integration can have different commercial terms.

These are alternative cases; do not add both benefits to the same sale. For a custom build, include quoted development and maintenance costs in the same calculation. A managed catalog trial is the sensible first move when the contribution needed to recover a custom build is unsupported by demand.

For an established store, compare contribution, new-customer mix, returns and repeat purchasing alongside channel orders. Where volume supports it, use a staged rollout or holdout to estimate incrementality. With sparse data, report the uncertainty and keep the spend small.

What's Overhyped About Agentic Commerce?

The overstatement is treating published infrastructure as guaranteed demand or universal autonomous buying. Four distinctions protect the budget:

Open access to a protocol does not confer channel admission. OpenAI requires approval for direct feed onboarding, and Google requires approval for its UCP integrations.

A valid feed does not guarantee a recommendation. It gives an assistant usable product information. The buyer's needs and competing offers still matter.

A payment mandate does not mean every assistant buys unattended. Confirm the actual approval flow of the experience you are enabling.

A recorded channel order does not prove additional revenue. An attribution label cannot tell you what that customer would otherwise have done.

Act now if you sell eligible, standard products and can use a managed route. Wait on custom checkout spending if your product needs unsupported configuration, your provider cannot confirm payment compatibility, or your catalog is unreliable. A wholesale-only operation outside these consumer channels can keep its existing checkout priorities while watching for a relevant supported route.

A Five-Step Checklist for This Quarter

Finish the quarter with a verified buying path and a contribution-based decision. Give each step an owner and a review date.

1. 1

   1. Confirm store, channel and product eligibility

   List the channels you want, the merchant and buyer countries they support, and the products you can sell through each. In Shopify, open Sales channels > Agentic and inspect Catalog eligibility. Record whether each route ends at your store or inside the assistant. Follow the maker's full requirements for exclusions and required policies.

2. 2

   2. Audit a small product set

   Choose products customers already understand and buy, with dependable fulfilment. Check every variant's identifier, title, image, price, inventory, dimensions, delivery information and return policy. Compare the feed with the product page.

3. 3

   3. Connect the supported feed and payment route

   Use your platform's channel settings first. For Shopify's Google route, confirm Merchant Center product and shipping data. For direct OpenAI feed onboarding or Google UCP integration, complete the appropriate approval process. Document which agents may browse, access accounts or submit orders, and how you enforce that permission. Ask the PSP to confirm token support, charges, refunds and transaction identification. Record the commercial terms before approving custom work.

4. 4

   4. Rehearse the complete customer order

   Once you have channel access, follow a supported buying path with an eligible product. Check the variant, shipping, total, buyer approval, order record, receipt and refund handling. Test unavailable inventory too. Reconcile the resulting order in the merchant system; do not rely exclusively on browser analytics.

5. 5

   5. Set the budget from contribution, then review

   Track preparation hours, ongoing work, channel orders, new customers, refunds and contribution. State how much demand you believe is additional and why. Expand when the evidence covers the next cost; keep a small managed presence when it does not. Pause unsupported checkout work while continuing useful catalog maintenance.

Monday's move: the store owner and catalog lead open channel eligibility together, select the initial products and fix the first mismatch between the storefront and feed. That produces a concrete improvement before the business commissions an integration.

Agentic Commerce FAQ

What does agentic commerce mean?

An authorised AI agent helps find, compare and purchase goods for a customer. Check which actions the specific channel supports: a recommendation, a store-checkout referral and an in-assistant purchase are different experiences.

What companies are using agentic commerce?

OpenAI, Google and Shopify document shopping channels or commerce integrations. Stripe provides payment infrastructure, while Visa and Mastercard describe agent identity and payment programmes. Their roles and merchant requirements differ; participation in one does not admit a store to all the others.

Does Amazon use agentic commerce?

Yes. Amazon documents Buy for Me, which can purchase selected products from other brands' websites on a customer's behalf. Its maker page describes a beta for a subset of US app customers. Amazon's Buy for Me explanation gives the scope. This is Amazon's own agent; its policy toward outside purchasing agents is a separate access decision.

Is agentic commerce part of AI?

Yes. It applies action-taking AI to shopping and purchasing. The model's recommendation still needs an authorised path into the merchant's catalog, checkout and payment systems.

Is Alexa considered agentic AI?

Some shopping features are agentic. Amazon describes Alexa+ comparing products, placing orders and making an automatic purchase when a customer has explicitly configured it. A voice interface alone does not establish those capabilities. Amazon's Alexa+ shopping page describes the supported experience.

What are the applications of AI in e-commerce?

Applications include product discovery, comparison, merchandising, customer support and authorised buying. For an independent store, accurate answers about products and a reliable order handoff are useful starting points.

What are 10 advantages of e-commerce?

Potential advantages include wider reach, extended shopping hours, searchable catalogs, easier comparison, remote ordering, digital payments, order tracking, repeat purchasing, customer communication and measurable transactions. These are general ecommerce capabilities, not guaranteed gains from adding an AI agent.

What are the top 5 e-commerce platforms?

There is no universal top five for every store. For an agentic-commerce decision, compare catalog quality, supported assistant channels, payment compatibility, checkout restrictions and total operating cost. Shopify is one documented route; choosing a platform does not automatically make every product eligible.

Who is Amazon's #1 competitor?

There is no single answer across countries and product categories. For your store, the useful comparison is the competing offer for the same item, delivery deadline and total cost, rather than a company-wide ranking.

For practical decisions on AI and business growth, subscribe to the newsletter.