E-commerce13 min read

Agentic commerce: what we learned in New York about AI that buys for your customer

Webfluencer
WebfluencerEditorial team
Agentic commerceAIBranding
Agentic commerce: what we learned in New York about AI that buys for your customer

On Tuesday, June 2, 2026, during NY Tech Week, we were in New York for Building & Entrepreneurship in the Agent Era, the evening Cloudflare and Shopify hosted together at Shopify's space in SoHo. The lineup featured Stephanie Cohen and Rita Kozlov of Cloudflare, Brennan Loh of Shopify and investor Nick Grossman of Union Square Ventures. Underneath the conversation sat one question, and five months later it has only grown sharper: if an AI agent does the shopping, who does brand loyalty belong to?

This article is our answer. It covers what agentic commerce is, what has happened since that evening at OpenAI, Google, Amazon, Walmart, Shopify and Cloudflare, what already works in the Netherlands and Europe, and what your brand should do this month. We checked everything on November 3, 2026. Unless we name a brand, an example is an illustrative scenario and not a client case.

What is agentic commerce?

Agentic commerce is online selling in which an AI agent searches, compares and sometimes checks out on a customer's behalf, within the limits the customer sets. The difference from a chatbot or AI search engine is in the verb: an assistant answers, an agent acts. It opens product pages, reads your return policy, fills a cart and, if the customer allows it, places the order.

In practice we see three modes, and each one gives your store a different role:

  1. 1 Advising. The agent searches and compares, the customer buys in your store. This is how shopping in ChatGPT works today: the purchase is completed on the store's own checkout.
  2. 2 Preparing. The agent fills the cart and hands the customer over to pay. Google's Universal Cart can transfer a cart to the store's site this way.
  3. 3 Checking out. The agent places the order itself, after the customer approves or within a limit set in advance. Shopify has allowed this since late September for agents working in the buyer's browser, and Walmart offers it with spending limits.

The third mode gets the most attention, but the first is where it happens today. In March, OpenAI pulled back from checkout inside ChatGPT precisely because people search for products there but rarely buy.

The question from SoHo: who owns the customer?

In an online store, a brand works through a hundred small moments: the homepage, the photography, the tone of a product description, the packaging opened at home. An agent skips most of them. It reads data, policies, prices and reviews, and gives the customer a summary. The storefront you designed with care, it sees as text.

So what is left of brand loyalty? In our view, three things, and all three become more important:

  • Being named. In an instruction like "reorder that rain jacket from last year, same brand", the brand is the query. A brand that is named does not have to compete.
  • Becoming the default. After a good first purchase, an agent can remember what the customer was happy with, and the repeat purchase goes to the same store without comparison. That makes the first delivery, and everything the customer experiences with it, more important than ever.
  • Being the best described option. When the customer names no brand, the store wins whose product the agent is most certain will fit, arrive on time and be returnable. That is not a matter of fine words but of complete, verifiable facts.

The conclusion runs against what you often hear. Agentic commerce does not make brand building less important, it makes it more important. Unbranded queries turn into a battle over specs and price, and that battle rarely leaves a margin. A brand people name survives the agent. A brand that relies on the impulse of a beautiful homepage loses its stage.

What has happened since June

That evening in SoHo did not stand alone. The market had already shifted that spring, and the five months after moved fast. The key steps:

  • March 2026: OpenAI scales back Instant Checkout in ChatGPT. Stores go back to their own checkout and ChatGPT focuses on search and product discovery.
  • April 30, 2026: Mastercard and Rabobank complete the first payment by an AI agent in the Netherlands. After a single instruction, an assistant booked a coffee tasting through Mastercard Agent Pay. The agent never saw the card details, and the payment was confirmed with a Mastercard Payment Passkey.
  • July 1, 2026: Cloudflare reports that automated agents and bots drive more than half of all web requests, and lets site owners choose separately between search, training and agent use.
  • September 2026: Amazon blocks Muse, Meta's shopping agent. According to Amazon, Muse did not identify itself as an agent when buying and could browse customer accounts and order history. In the same weeks, Shopify partnered with Muse instead.
  • September 28, 2026: Shopify opens checkout to agents working in the buyer's browser, through WebMCP. An agent can change the address and delivery option and place the order, but only after the buyer approves.
  • October 15, 2026: Google launches Universal Cart in the United States, in Google Search and the Gemini app: one cart across stores, with checkout powered by the Universal Commerce Protocol (UCP). The store remains the seller of record and handles confirmation and support.
  • October 21, 2026: Walmart opens an Agent Gateway for third-party agents. Customers approve every purchase or set a spending limit, agents get no payment details, and agents that break the rules are blocked.
  • October 28, 2026: in Shopify Editions Fall '26, any agent or app gets one connection to Shopify Catalog, cart, checkout and Shop Pay through Shop.dev. According to Shopify, every Shopify store is agent-ready by default. What that means for your own store is in our article on Shopify Editions Fall '26.

Three patterns behind the news

Behind the separate headlines are three patterns. They are more useful than the news itself, because they will still hold next year.

1. The purchase stays with the store

OpenAI sends buyers back to the store's checkout. At Google, the store remains the seller of record, and the customer can transfer the cart to your own site. Shopify only lets an agent order after the buyer approves. Discovery moves to the agent, but completion, the customer relationship and service stay with you. Your store becomes less of a shop window and more the place where the purchase is confirmed.

2. If you don't identify yourself, you don't get in

Amazon did not block Muse for being an agent, but for not identifying itself as one. Walmart blocks agents that break its rules. Identity becomes the price of entry. That is exactly what Cloudflare works on, and it explains why Cloudflare stood next to Shopify that evening.

3. Limits become a product

Walmart builds spending limits per agent, Mastercard has the payment confirmed with a passkey and Shopify asks for approval before the order. That matches what Dutch consumers want. In research by Appinio for Riverty among 1,000 Dutch consumers (December 2025), 67.9% find it important to be able to review or stop decisions made by an AI, and 55.6% say every purchase needs human confirmation. More than half would let an AI assistant spend at most 50 euros a month on its own.

Cloudflare's role: the web's doorman

Cloudflare sits as a network between many websites and their visitors. For a long time that came down to one question: human or bot? With agents, that question has become too blunt. An agent buying a jacket for a customer is technically a bot, but commercially a customer. Cloudflare puts it this way: "Block every bot and you block your next customer; let everything through and you're exposed."

The answer is called Web Bot Auth. An agent signs its requests cryptographically, so a website can check who is knocking. For this, Cloudflare introduced the signed agents category in August 2025, with OpenAI's ChatGPT agent among the first participants. Visa and Mastercard build on it with the Trusted Agent Protocol and Agent Pay. In those, an agent also states whether it is only browsing or wants to pay.

For a brand the lesson is simple: you decide which agents get in, and for what. OpenAI makes that distinction with three separate bots: GPTBot for training, OAI-SearchBot for search results in ChatGPT and ChatGPT-User for actions a user starts. Block OAI-SearchBot and you no longer appear in ChatGPT's search answers. Blocking every AI bot out of caution also shuts out buyers.

What an agent sees of your brand: four doors

An agent reaches your brand through four doors. If one is locked, it picks another store, and you will never notice.

Four cast-iron doors in SoHo at blue hour, one of them open with warm light and a trail of light particles going inside
An agent only walks through a door that is open.

1. Findable: your product data

An agent can only recommend what it can read. Attributes that live only in running text, such as fit, material or roast level, are half visible to an agent. Put them in fixed fields: in your product feed, in structured data on the product page and, on Shopify, in metafields. Then an agent can filter on exactly what the customer asks for.

2. Understandable: your policies as facts

Do you ship to Belgium? How long can a gift be returned? What does shipping cost under 50 euros? An agent is not looking for atmosphere here, but for certainty. So write your policies as short, verifiable facts and keep them current in one place. On Shopify, that is the Knowledge base under Sales channels > Agentic. On webfluencer.nl we also publish an llms.txt: a short, readable summary of the site for language models.

3. Accessible: your network and your robots.txt

Check which bots your site turns away, in your robots.txt and, if you use Cloudflare, in your dashboard. A defensible choice: block training through GPTBot and Google-Extended, but allow search bots and agents that customers deploy themselves. On Shopify you edit robots.txt through the robots.txt.liquid template in your theme.

4. Payable: paying without a card

This is where the Dutch market differs. The agent payment routes we found run through cards and wallets: Shop Pay, Google Pay, Visa and Mastercard. According to the Thuiswinkel Markt Monitor, 73% of online purchases in the Netherlands in the first half of 2026 were paid with iDEAL | Wero. We could not find a route for an agent to pay with iDEAL | Wero. For now, Dutch buyers complete their purchase on your own checkout, so that checkout has to be good.

What already works in the Netherlands

In short: agents already talk about your brand, but checking out inside the agent barely happens here yet.

  • Discovery: an AI assistant already discusses your brand, in Dutch too, whether you are prepared or not.
  • Preparing and checking out in the agent: Google's Universal Cart works only in the United States, direct checkout in Google AI Mode and Gemini applies to buyers in the United States, Canada and Australia, and Walmart is a US retailer.
  • Payment by an agent: the first Dutch payment by an AI agent, in April, was a pilot by Mastercard and Rabobank. It is not widely available yet.
  • Selling across borders: if you already ship to the United States, Canada or Australia, the agent channels there are a real market today.

The United States shows where this is heading. According to Adobe, traffic from AI assistants to US retail sites grew 62% year over year in July 2026, and those visitors converted 60% better than other traffic. Adobe gives no share of total traffic: this is a channel that is growing fast, not one that is already large.

Three scenarios

To make it concrete, three scenarios, all three illustrative.

The coffee roaster with a subscription

A customer tells their assistant: order a kilo of filter beans every month, light roast, from Ethiopia or Kenya, 35 euros at most. Without a brand name, the agent compares stores that show origin, roast level, weight and price per kilo as fixed attributes and offer a subscription. A roaster that only mentions this in an atmospheric product description falls outside the selection.

The real gain comes the month after. If the first bag went down well, the instruction becomes: order again from the same roaster. From that moment, the roaster no longer competes on price. Packaging, taste and a handwritten card still do their job, just at a different moment: not at the first choice, but at the second.

The Sinterklaas gift with a hard deadline

On November 28 someone asks: find a drawing set under 40 euros for an eight-year-old, delivered by December 4 at the latest. Sinterklaas, the Dutch gift-giving holiday, falls on December 5. The agent needs three facts: is it in stock, when will it arrive if I order today, and can it still be exchanged after December 5? A store with a clear Sinterklaas order deadline and an extended return window in one fixed place beats a store that only puts this in a homepage banner. A banner is for people; an agent looks for it in your policies and product data.

The fashion brand and the fit

A customer asks: a rain jacket in size M that fits like my last one, with free returns. If they name the brand of the last jacket, the agent goes there first. If not, the store wins that gives size charts in centimeters, states whether a style runs small or large and names its return policy without conditions. Reviews about fit help, because an agent can weigh them.

What to do this month

  1. 1 Ask for yourself. Put two questions to ChatGPT, Gemini, Copilot and Perplexity: which brand do they recommend in your category and why, and what is your brand's return policy? Note what is right, what is missing and what is wrong. That is your baseline.
  2. 2 Put attributes in fields. Size, material, origin, compatibility: anything a customer would name as a filter belongs in your feed and structured data, not only in the description.
  3. 3 Write your policies as facts. Delivery times per country, return windows, shipping costs and service, short and in one place. On Shopify, that is what fills the Knowledge base.
  4. 4 Choose which bots you allow. Block training if you want to, but let in search bots and agents that customers deploy themselves. Check both your robots.txt and your firewall.
  5. 5 Measure AI traffic separately. Create a channel in your analytics for referrers like chatgpt.com, gemini.google.com, perplexity.ai and copilot.microsoft.com, so you can see what it brings in.
  6. 6 Make your checkout strong. As long as Dutch buyers pay on your own site, that is where the sale is decided: fast, with iDEAL | Wero at the top and no surprises in the shipping costs.
  7. 7 Invest in your name. The best position in agentic commerce is being named. That takes a brand people remember, with a story that still stands in three sentences. That is where our brand positioning and branding work comes in.

Frequently asked questions

What is agentic commerce?

Agentic commerce is online selling in which an AI agent searches, compares and sometimes checks out on a customer's behalf, within limits the customer sets, such as a budget or a required confirmation.

How is it different from AI search?

With AI search, an assistant answers and the customer buys. With agentic commerce, the agent also takes steps: filling the cart, entering details and, after approval, placing the order.

Can an AI agent already check out for me in the Netherlands?

Barely. The first Dutch payment by an AI agent was a pilot by Mastercard and Rabobank on April 30, 2026. Checkout in Google's Universal Cart is limited to the United States, and direct checkout in Google AI Mode to buyers in the United States, Canada and Australia. In the Netherlands, a purchase is almost always completed on the store's own site.

What are UCP and ACP?

The Universal Commerce Protocol (UCP) is an open standard that Shopify and Google developed together and announced in January 2026; Google's Universal Cart checks out on it. The Agentic Commerce Protocol (ACP) was developed by OpenAI with Stripe as the basis for Instant Checkout in ChatGPT, which started in September 2025. Both describe how an agent finds products, builds a cart and checks out.

Should I block AI bots?

Not all of them. You can block training and still stay findable for search bots and for agents customers deploy themselves. Block everything and you also disappear from the answers in which customers make their choice.

Is my Shopify store ready for agents?

According to Shopify, every store is agent-ready by default. Whether an agent actually recommends your products depends on your product data, your policies and your brand. The settings per area are in our article on Shopify Editions Fall '26.

In closing

The evening in SoHo ended with a question. Our answer, five months later: the customer belongs to the brand they name, and the agent follows that name as long as the store behind it delivers. Agentic commerce changes how people buy, not why they choose a brand.

Curious how your brand and store look to an agent? We build brands and Shopify stores that people choose by name, from Amsterdam and New York. Get in touch for a conversation.

Webfluencer
WebfluencerEditorial team
Deel