The Checkout Button Is Disappearing, and AI Agents Are Buying

For two decades the ‘add to cart’ button marked the decisive moment in any online purchase. That button is now fading, not because of a redesign, but because the hand pressing it is no longer always human. In June 2026 Mastercard launched Agent Pay for Machines, a service that lets AI agents complete microscale purchases at machine speed across a network spanning more than 200 countries and territories, according to the company's own release. Around the same time, a Checkout.com report found that 89% of merchants are actively preparing for this shift even though AI agents currently handle only 3% of transactions. That gap between readiness and reality is the real story this summer, more interesting than any single new gadget.

The story is not confined to the US or Europe. UAE consumers, according to the same survey, expressed markedly more trust in AI shopping agents than their counterparts in the United States or Britain. This piece breaks down four competing payment protocols, reads the growth numbers, and explains why the UAE reached the front of the line before most Western markets even found the door.

What makes this moment different from earlier waves of e-commerce hype is that the infrastructure arrived first this time. Payment networks, browsers and chat platforms all shipped competing standards within the same twelve months, before most shoppers had even tried buying something through an agent once. Usually a technology proves itself with consumers, then the plumbing catches up. Here the plumbing came first, and the consumer behavior is now racing to catch it.

In the space of one year, four rival frameworks emerged, each trying to become the common language between AI agents and merchants. Stripe and OpenAI launched the Agentic Commerce Protocol (ACP) on September 29, 2025, an open standard built on ‘Shared Payment Tokens’ scoped so tightly that the agent never sees a raw card number, according to Stripe's own newsroom post. It started with Etsy, then expanded to more than a million merchants on Shopify, including Glossier, Vuori and SKIMS.

Google took a different route. On September 16, 2025 it unveiled the Agent Payments Protocol (AP2), built around cryptographically signed digital contracts it calls Mandates, backed by more than 60 partners including PayPal, Mastercard and American Express, per the Google Cloud blog. By April 2026 Google added a ‘human not present’ mode letting an agent buy a limited-release ticket the instant it drops. Google then donated the whole protocol to the FIDO Alliance so no single company would own it.

Visa went for trust and fraud screening instead. Its Trusted Agent Protocol, introduced in October 2025, helps merchants tell a legitimate AI agent apart from a malicious bot, and by December 18, 2025 the company reported that 47% of US shoppers were already using at least one AI tool somewhere in their shopping journey. Mastercard, meanwhile, extended its own program to machine-to-machine transactions through more than 30 founding partners. The upshot is that any merchant wanting to sell to AI agents now faces four different standards rather than one, a situation that looks a lot like the early browser wars.

On June 10, 2026, Mastercard rolled out Agent Pay for Machines, a payment layer purpose built for transactions no human ever sees: a server calling an API and paying per call, or a sensor buying a fresh minute of weather data. ‘Machine payments can make it possible for services to be bought and sold among agents at fundamentally different scales than payments today,’ said Jorn Lambert, the company's chief product officer, ‘very high volumes, very small values, very fast and at extremely low latency.’

The service handles agent credentialing and verification, programmable spending limits set in advance, and settlement across cards, bank accounts and even stablecoins. Founding partners span traditional payment firms like Checkout.com and Adyen, blockchain platforms including Coinbase, Ripple and the Solana Foundation, and Stripe itself, despite Stripe competing head to head with Mastercard on a rival protocol. That detail says something: no company is betting on a single standard, they are all trying to have a seat at every table.

The real difference from a normal credit card is scale. A ten cent charge for a single API call makes no economic sense under ordinary transaction fees, but it makes complete sense once thousands of micro charges get bundled into one hourly settlement. That is the market Mastercard is chasing, not the shopper buying a pair of shoes, but the invisible economy of machines quietly negotiating and paying each other all day long.

Checkout.com's ‘Agentic Commerce 2026’ report, published June 9, 2026, surfaced a striking trust gap between the UAE and Western markets. Consumers across six markets were asked whether they felt comfortable letting an AI agent complete a purchase on their behalf. In the UAE, 79% said yes, clearly ahead of the rates the same survey recorded for the UK and the US. More telling still, 64% of UAE consumers said they trust an AI shopping agent more than they trust their own family members to pick out purchases, against just 27% in the US and 25% in the UK, as reported by Zawya citing the study.

The numbers go beyond general trust. 72% of UAE respondents are willing to use an AI agent to jump digital queues when buying event tickets, and 62% are open to letting an agent buy things without telling anyone. Caution returns, though, once sensitive financial data enters the picture: only 24% are comfortable sharing salary or bank balance information with an AI agent.

This pattern has a structural explanation as much as a cultural one. The UAE has one of the highest smartphone penetration rates anywhere, mature digital payment rails, and a consumer base that has spent years fully on digital government and banking services. Once an AI agent becomes just one more layer on top of a payment system that was already digital, the leap of trust is far shorter than in markets still partly running on checks and cash.

On November 17, 2025, Salesforce took its Agentforce Commerce platform fully live, attaching a striking figure to the launch: traffic reaching retail sites from AI assistants grew 119% year over year in the first half of 2025, according to company data. What matters more than the growth rate itself is what it implies, a shopper who no longer starts a purchase from a search box or a store app but from a conversation with an assistant that goes looking on their behalf first.

Salesforce projected AI agents would influence 22% of global orders during Cyber Week, a striking figure given that this kind of influence barely existed two years earlier. To manage that shift the platform ships with three agent types, a shopper agent for discovery and recommendations, a buyer agent that handles checkout and payment, and a merchant agent that optimizes fulfillment and inventory on the seller's side.

The point of that three way split is that agentic commerce is not a single event happening at the payment moment. It is a full chain of automated decisions running from the first product search through packing and shipping. A merchant who thinks adding one AI compatible payment button is enough is missing the larger picture.

At Google I/O on May 19, 2026, Google unveiled Universal Cart, a single shopping hub that pulls in whatever a user sees across Search, the Gemini assistant, YouTube and even Gmail into one persistent cart that does not empty when the tab closes. The cart tracks price changes, checks whether products are compatible with each other, and sends alerts when something goes out of stock, while plugging into Google Wallet for loyalty programs. Launch partners include big names like Nike, Sephora, Target and Walmart, alongside Shopify merchants.

The more important technical piece is that Universal Cart is compatible with the Universal Commerce Protocol (UCP), Google's attempt to standardize how different platforms talk to AI agents instead of forcing every merchant to build a separate integration for each protocol. In April 2026, Google shipped AP2 version 0.2.0, adding ‘human not present’ payments, meaning an agent can buy a concert ticket or a limited release item the moment it becomes available, without waiting for the user's real time approval, acting only on spending conditions set in advance.

The bet Google is making is not just that people will buy from inside a chat window. It is that everyday search queries, YouTube views and even email become data feeding one unified purchase decision. The browser, in that sense, stops being a gateway to many stores and becomes the store itself.

On September 29, 2025, Stripe and OpenAI launched Instant Checkout inside ChatGPT, letting a user complete an entire purchase without leaving the chat window. The feature started with Etsy, then expanded to more than a million merchants on Shopify, including recognizable names like Glossier, Vuori, Spanx and SKIMS, according to Stripe's newsroom release.

The real innovation here is not the chat interface itself but the protection built behind it. The ACP protocol relies on what it calls a Shared Payment Token, a token scoped to one specific transaction, tied to a particular merchant and a particular cart total, so the AI agent, and even OpenAI itself, never sees the raw card number at any point. Stripe's Kevin Miller put it plainly: ‘Stripe has spent the last 15 years optimizing commerce for human buyers. Now, we are starting to do the same for agents.’

What sets this model apart from earlier payment integrations is that a merchant connects once and becomes sellable through any AI agent that supports the protocol, rather than building a separate integration for every platform. That is exactly why the four way protocol war matters. A merchant who picks the wrong standard today could find itself invisible to half the AI agents on the market within a single year.

The Gulf's lead is not just a matter of survey answers. Among the partners Visa picked to field test its Trusted Agent Protocol was Aldar, the Abu Dhabi based real estate and retail developer, alongside American partners like Ramp and Consumer Reports, according to Visa's press release dated December 18, 2025. Choosing Aldar specifically is telling. The company runs large shopping and property services operations, and its experience with AI agents will help set how an entire property retail sector in the region handles visitors who never book an appointment or browse a website themselves, but send an AI agent to search and compare on their behalf.

That choice is not an accident. Gulf payment infrastructure grew over the past decade built directly around the mobile phone, largely skipping the checks and paper transfer stage that still lingers in many Western markets. When an entire country builds its payment system digital from the ground up, adding an AI agent layer on top is a comparatively small engineering adjustment, not a full rebuild.

The practical takeaway for app owners and online retailers in the region is that the preparation window is narrower than it looks. Once 79% of consumers in one market are comfortable delegating a purchase decision to an AI agent, any online store that does not speak the ‘language’ of these agents, through at least one of the four protocols, risks becoming effectively invisible to a growing slice of customers, even if its website looks excellent to a human eye.

For all the talk about agentic commerce, the actual numbers remain modest so far. Only 3% of transactions currently involve an AI agent in any form, according to the Checkout.com report. Expectations, though, are moving on a completely different clock. 33% of surveyed consumers expect at least 10% of their own purchases to be AI driven within just one year from now. The gap between 3% today and more than 10% expected soon is exactly what is unsettling retail.

What is more interesting is that merchants themselves see this gap coming. 72% admit that consumers will adopt agent led shopping faster than their own stores can get ready, even though 89% say they are actively preparing. In other words, the awareness is there, but execution speed is not keeping pace with it.

There is also a smaller but telling gap in financial trust. The average consumer is comfortable letting an AI agent spend roughly 177 pounds on a single purchase without extra approval, while merchants estimate that ceiling closer to 200 pounds, meaning merchants are expecting more trust from customers than customers are actually extending. Any store that sets its agent spending limits based on merchant assumptions rather than real consumer behavior risks running into unexpected pushback from its own customers.

The 89% figure deserves a second look too. Preparing does not mean the same thing at every company. For a large retailer with an existing API team, it might mean a working integration with one or two protocols already in testing. For a smaller merchant, it might mean nothing more than a slide in an internal strategy deck. Checkout.com's own analysts flagged this distinction, noting that stated readiness tends to run well ahead of technical readiness whenever a new payment rail appears, and agentic commerce is unlikely to be the exception.

Trust in agentic commerce is not unconditional, and the Checkout.com report maps out exactly where its limits sit. Asked what safeguards they consider essential before letting an AI agent spend on their behalf, consumers put three requests at the top of the list, all within a few points of each other, a clear spending cap (30%), the ability to revoke the agent's authority instantly at any moment (29%), and an easy way to cancel or reverse a purchase after it has gone through (28%). How close those three numbers sit to each other suggests consumers do not treat any one of them as optional. They want all three as a baseline.

On the other end of the spectrum, real skepticism persists. 24% of respondents globally say they will never delegate a purchase decision to AI at all, and 27% go further, saying they trust no organization, not a bank, not a tech platform, not a merchant, to safely run an AI agent on their behalf.

What this means is that the future of agentic commerce will not be built purely on protocol speed or model intelligence. It will be built on details that look small, a clear cancel button, an instant notification for every transaction, a spending cap that is easy to adjust from inside the app. Any platform that skips these small details, however technically advanced, will lose a meaningful slice of users who were already hesitant to begin with.

For app owners and online retailers in the region, the first move is not to lock in one protocol permanently. All four standards are still competing and the market has not settled on a winner. A more practical priority is making sure the product data itself is structured and machine readable, current prices, accurate inventory, clear sizing and specification details, because any AI agent, whatever protocol it runs on, needs that data before anything else.

The second move is reviewing the checkout experience from a genuinely different angle, not ‘is this easy for a human’ but ‘can an automated agent complete this without human intervention if it needs to.’ That does not mean abandoning the traditional interface, most buyers today are still human, but it does mean building a parallel path an agent can walk confidently.

The third move, especially urgent in a Gulf context, is revisiting the safety policies covered earlier, a spending cap, instant revocation, a clear way to reverse a purchase. A store that offers these safeguards clearly will look more trustworthy to an AI agent that is, by design, built to favor merchants who reduce risk for the user. Falling behind on this preparation will not necessarily cost a retailer human customers right away, but it will gradually make that retailer invisible to a slice of the market that grows every month.

One more practical note worth flagging: liability is still unsettled. If an agent buys the wrong size or the wrong quantity, existing consumer protection rules were written with a human clicking the button in mind, not a delegated agent acting on a standing instruction. Regional regulators have not yet published clear guidance on where responsibility sits when a mandate is technically valid but the outcome is not what the shopper wanted. Merchants who document their agent interactions carefully now will be in a far stronger position once that guidance eventually arrives.

Mastercard Launches Agent Pay for Machines to Unlock Super-Fast, Always-On Payments

Consumer demand for AI shopping is forming fast but trust for agentic commerce is still catching up, Checkout.com

UAE leads in adoption of AI shopping, Zawya

Salesforce Unveils Agentforce Commerce Capabilities as AI Assistants Grow Retail Traffic Volumes by 119%

Announcing Agent Payments Protocol (AP2), Google Cloud Blog

Google donates Agent Payments Protocol to FIDO Alliance

Google launches Universal Cart and updates AP2 at I/O 2026, The Next Web

Stripe powers Instant Checkout in ChatGPT and releases Agentic Commerce Protocol codeveloped with OpenAI

Visa and Partners Complete Secure AI Transactions, Setting the Stage for Mainstream Adoption in 2026

Visa Introduces Trusted Agent Protocol: An Ecosystem-Led Framework for AI Commerce