Agentic Commerce: What It Means for Your Store

Agentic commerce is buying and selling where an AI agent does the work on someone's behalf: researching products, comparing options, checking policies, and completing the purchase. IBM's January 2026 definition puts it plainly, agents acting for consumers or businesses to research, negotiate and complete purchases without direct human intervention. For a store owner, the practical version is smaller and more useful than the trillion dollar forecasts. A share of your traffic is becoming software, and software reads structured data, not layouts. Stores that are legible to machines get picked; stores that are not go missing in a channel they never optimized for. Here is how the plumbing works, which parts are real, and what to change first.

What is agentic commerce, exactly?

Agentic commerce is ecommerce where the buyer is software acting under instructions. A shopper states what they want and what the limits are, budget, delivery date, brand, and the agent handles discovery, comparison and checkout. Mastercard's explainer from September 2025 describes it as the agent closing the loop, finishing the task instead of assisting with it.

The line worth drawing is between advising and acting. A recommendation engine suggests. A chatbot answers. An agent decides and transacts, which means your counterparty in the sale is no longer a person reading a page. It is a program reading data.

A person weighs the photography, the headline, the reviews, the feel of the site. An agent weighs price, availability, delivery date, return window, and whether your data is consistent enough to trust. Persuasion and legibility are different problems, and they need different work.

How does an agent actually buy something?

In four moves: the shopper states intent and constraints, the agent reads structured product data from candidate stores, the agent picks and builds a cart, and payment runs through a flow designed to authorize on the shopper's behalf. Stripe's guide to agentic commerce covers the payment half, including the Model Context Protocol for giving agents machine readable access to inventory, prices and checkout logic, and virtual card numbers so the agent never touches real card details (accessed 1 October 2026).

Two details get missed in most explainers.

First, autonomy is tiered. Low risk and repeat purchases run end to end, while higher value or sensitive ones bounce back for a human approval, which is how IBM described the pattern in its January 2026 overview. The agent does the shopping, not usually the final yes.

Second, the agent does not see your site the way a shopper does. If the price is injected by a script that runs after the page loads, or the stock count lives in an app widget, the agent may read nothing at all and move on to a store it can parse. Being wrong is bad. Being unreadable is worse, because you never find out.

What is the difference between agentic AI and agentic commerce?

Agentic AI is the capability; agentic commerce is one use of it. Agentic AI describes any system that plans, calls tools and acts toward a goal without step by step prompting. Point that at buying and selling and you inherit constraints software people rarely enjoy: authentication, payment authorization, fraud checks, receipts, disputes, returns.

The distinction matters when someone tries to sell you agentic commerce as a product. For a merchant, there is nothing called agentic commerce to buy. There is a store that agents can read, or a store they cannot.

Is agentic commerce the future, or is it a forecast?

Both, and the split is worth saying out loud. IBM cites estimates of 3 to 5 trillion dollars in agentic commerce globally by 2030, and a 2026 IBM Institute for Business Value study found 45% of consumers already use AI for part of the buying journey. Salesforce cites Gartner's projection that 33% of enterprise software applications will include agentic AI by 2028, up from under 1% (accessed 1 October 2026).

The reality check sits on the same results page. One of the top ten Google results for this query is a forum thread asking whether anyone has had success with agentic commerce yet (collected 1 October 2026). Large forecasts, thin public evidence of merchants closing real revenue through agents.

One more number from our own research, because it sets a useful ceiling. When we priced this term, it drew 5,400 Google searches a month against 121 searches a month inside AI assistants (DataForSEO, US, 1 October 2026). Assistant volume on this topic is about two percent of search volume. Anyone telling you the web is finished is ahead of their evidence.

Our read: discovery lands first and buying trails it. People already open an assistant to research a purchase, and that behavior does not wait for a payment standard. Build to be read and quoted correctly, and the checkout side arrives as a platform upgrade rather than a project.

Which companies are using agentic commerce?

Platforms, payment networks and large marketplaces, more than individual merchants. Amazon and Yahoo appear among adopters, Alibaba, Tencent and ByteDance have shipped AI shopping apps, Alibaba's Accio Work agent researches demand and generates listings for exporters, and Alipay's AI Pay processes payments made by autonomous agents in China (Wikipedia, accessed 1 October 2026).

That list is lopsided on purpose. The companies with a customer relationship at the assistant layer move first, because they own the surface where the shopper states intent. Merchants are the supply side of this market, which is a weaker position than it sounds and a simpler job than it sounds: be the store that answers correctly when an agent asks.

Which protocols matter, and which can you ignore?

You can ignore the question, as long as you ask your platform and your payment processor to answer it for you. Four standards are worth being able to name, all per Wikipedia's roundup, accessed 1 October 2026:

• Agentic Commerce Protocol (ACP). Developed by OpenAI and Stripe. Defines how agents read product information, create and manage carts, start checkout, delegate payment credentials and receive order status.
• Universal Commerce Protocol (UCP). Announced by Google in January 2026 as an open web standard for interoperability between agents and retail systems.
• Agent Payments Protocol (AP2). Announced by Google in September 2025. Payment method agnostic, covering cards, bank transfers and stablecoins, using cryptographically signed mandates to record what the shopper authorized.
• Agentic Commerce Trust Protocol. Unveiled by Alipay and adopted first by Qwen, connected to Taobao Instant Commerce and AI Pay.

Standards fights get settled by whoever holds distribution, which here means the assistants, the checkout providers and the marketplaces. Betting your roadmap on one of them is a way to spend a quarter on something your processor ships for free. Every one of these protocols needs the same thing underneath: product data that is structured, current and true. That part is yours, and it is useful whether or not any of these names survive.

What do agents need from your store?

Seven things, roughly in the order they break:

1. Structured product data. Schema.org markup and consistent attributes: price, brand, color, dimensions, availability. Salesforce's 2026 overview makes the same point from the enterprise side, that discovery systems rely on machine readable data rather than layout.
2. Server rendered pages. If the content only exists after client side JavaScript runs, assume some readers get a blank page. That has always been an SEO problem, and it is a bigger one when the reader is a crawler with a budget.
3. Prices and stock that are true at read time. An agent that gets quoted one price and charged another does not file a support ticket. It stops picking you.
4. Policies in plain text. Returns, shipping windows, warranty, contact. Not in an image, not in a PDF, not behind an accordion that loads on click.
5. One variant model. Sizes, bundles and subscriptions described the same way everywhere, so a size 8 is the same object on your product page, your landing page and your feed.
6. Speed. Agents making dozens of comparisons give up on slow responses, and so do people. Page speed is one of the few conversion levers that compounds on every visit.
7. Consistency across channels. The product in your ad, your email and your landing page should be the same record, not three copies that drift apart after the first sale.

Read that list again and notice what it is. It is the SEO checklist with better manners. The work that makes a page rank is the work that makes a page quotable by an assistant, which is the work that makes a store readable by a buying agent. One job, three payoffs. Anyone selling you a separate agentic commerce program before you have done this is selling you the roof before the foundation.

Does agentic commerce kill your landing pages?

No, and the use cases say why. The purchases suited to agents are the predictable ones: replenishment, subscription management, business to business procurement, inventory reordering, price monitoring. Those are transactions where the decision was made a long time ago and the clicking is a chore.

Considered purchases still involve a person looking at something and wanting it. Paid traffic still lands on a page, and that page still decides whether the money you spent turns into revenue. Our customers' numbers have not moved toward the agent; they have moved toward more pages, faster, built with our page builder. Huron saw more than a 50% conversion lift year to date across paid channels. Woxer's conversion rate went up 208% in 90 days with cost per acquisition down $10. Simple Modern's Harvest Collection page converted 51% better than its Shopify counterpart. VaynerCommerce and POSSIBLE put revenue per visitor up 10.25% and conversion rate up 9.91% in 30 days (all verified August 2026).

Loop Earplugs is the clearest version of the pattern: more than 40 landing pages tested, roughly one a week, without touching core store code. That is what winning looks like in a channel where the reader is human. Nothing about agents makes it worth less. It adds a second audience reading the same store through a different door.

Where should you start?

Six steps, ordered by payoff per hour:

1. See what a machine sees. Fetch one product page and one landing page with JavaScript disabled. Whatever is missing is missing for agents and crawlers too.
2. Fix the structured data. Product, offer, price, availability, reviews. Make it match the page, because mismatches get you ignored rather than corrected.
3. Make price and stock honest at the source. One record feeding every surface, so a sale changes in one place.
4. Write the policies as text. Returns, shipping, warranty. Short, dated, on a crawlable page.
5. Ask, do not build. Get a straight answer from your platform and processor about which agent checkout standards they support and when. Put the engineering you were going to spend on a protocol into your catalog instead.
6. Keep shipping pages. The volume of pages you can launch and update is still the thing that moves revenue, and it is the capability that makes you ready for a new channel instead of blocked by it.

How we build for this

We build a computer for commerce, so pages are one app among several rather than the whole product, and the data underneath is shared by design. Our sites are server rendered because search decides whether a store gets found. We hold opinionated data models for products, orders and customers, so a generated ad uses real photos and real prices, and a landing page shows the stock you actually have. When a sale changes, you change it once. Around 91,000 brands have built with us (our own figure, accessed 1 October 2026).

None of that was built for agents. It turns out to be what agents need, because being readable by a machine was never a new requirement. It was the SEO requirement, with a stricter reader.

The forecasts will keep getting larger. The protocols will keep getting announced, and most of them will be settled above your head. What stays in your hands is a store that tells the truth in a format anything can read, and the ability to put a new page in front of whoever shows up. Shoppers, crawlers, agents. Same door, different readers, and the store that answers clearly is the one that gets picked.

See what it builds on your catalog