Your AI Just Got a Wallet
Here’s a sentence that would’ve sounded absurd two years ago: Visa built a payment system specifically for AI agents.
On April 8, 2026, Visa launched Intelligent Commerce Connect — a platform that lets AI agents discover products, compare prices, and complete purchases on your behalf using real payment credentials. Not a demo. Not a concept video. A production system currently in pilot with partners including AWS, Highnote, and Mesh.
This happened the same week ChatGPT crossed 900 million weekly active users and already has an Instant Checkout feature baked into the chat interface. Google launched its Universal Commerce Protocol back in January with Shopify, Target, and Walmart on board. McKinsey is projecting the agentic commerce market could hit $1 trillion in the US alone by 2030.
The way money moves online is about to look very different, and most people haven’t caught on yet.
What Agentic Commerce Actually Means
Strip away the buzzwords and agentic commerce is straightforward: an AI agent shops, compares, negotiates, and buys things for you. Not “recommends products.” Not “shows you a carousel.” Actually completes the transaction.
Think about what you do when you need to buy, say, a new pair of running shoes. You open three or four tabs. You compare prices. You check reviews. You look for coupon codes. You wonder if you should wait for a sale. Then you fill in your shipping address for the hundredth time and hope the sizing chart is accurate.
An agentic commerce system collapses all of that into a conversation. You tell your AI agent “I need trail running shoes, wide fit, under $150, and I’d prefer something with good ankle support.” The agent searches across merchants, checks real-time inventory, compares prices, maybe even negotiates a discount through a loyalty program you forgot you had — and then checks out. You approve the purchase, and it’s done.
The key difference from what we had before: the AI isn’t just answering questions about products. It has the authority and the payment infrastructure to act on your behalf.
The Four Steps of an Agent Purchase
Every agentic commerce transaction follows roughly the same pattern, regardless of which protocol or platform powers it:
Intent capture. The agent figures out what you actually want. This goes beyond keyword matching — a good agent understands that “something for my mom’s birthday, she’s into gardening, budget around $80” is a shopping request with specific constraints.
Discovery and comparison. The agent queries multiple merchant catalogs simultaneously. This is where structured product data becomes critical — more on that later. The agent evaluates options against your stated preferences, past purchase history, and any constraints you’ve set.
Negotiation and selection. In some implementations, agents can apply coupon codes, stack loyalty rewards, or choose between shipping options to optimize for your preferences (fastest delivery vs. lowest cost). The agent presents you with a shortlist or a single recommendation with reasoning.
Purchase execution. The agent initiates payment through tokenized credentials. You never share your raw card number with the agent — it works through secure token vaults. You get a confirmation, and the merchant handles fulfillment just like any other order.
The whole thing can take under a minute for routine purchases. For more complex buying decisions, the agent might go back and forth with you a few times before pulling the trigger.
The Protocol War Nobody’s Talking About
This is where things get messy. Three different protocols are competing to become the standard for how AI agents talk to merchants and payment processors. And the winner will shape online commerce for the next decade.
OpenAI’s Agentic Commerce Protocol (ACP)
OpenAI and Stripe co-developed ACP, and it’s the most visible protocol right now because it powers ChatGPT’s Instant Checkout feature. When a ChatGPT user buys something directly in the chat interface, ACP handles the merchant-agent communication behind the scenes.
ACP is an open standard — the spec is on GitHub — but OpenAI clearly designed it with their ecosystem in mind. The protocol handles product discovery, cart management, payment initiation, and order tracking. Merchants integrate through Stripe, which makes onboarding relatively painless if you’re already a Stripe customer.
The catch: OpenAI charges merchants a 4% transaction fee on every Instant Checkout purchase, on top of Stripe’s standard processing fees (~2.9% + $0.30). That’s a meaningful cut, and there are already signs OpenAI is pivoting away from Instant Checkout toward letting merchants build their own checkout experiences through ChatGPT apps. The strategy is still in flux.
Google’s Universal Commerce Protocol (UCP)
Google announced UCP in January 2026 with a coalition that reads like a who’s who of retail: Shopify, Etsy, Wayfair, Target, Walmart. The protocol is open-source and designed to be agent-agnostic — meaning any AI agent can use it, not just Google’s.
UCP takes a different architectural approach than ACP. It defines “functional primitives” — standardized operations like catalog lookup, cart management, and identity linking — that any agent or merchant can implement. It’s also explicitly compatible with other protocols like Agent-to-Agent (A2A) and the Model Context Protocol (MCP), which makes it more of a connective layer than a walled garden.
Recent updates added a Catalog capability for real-time product details and Identity Linking so shoppers can use their existing loyalty accounts through AI agents. Retailers including Lowe’s, Poshmark, and Reebok are live at launch.
The Google advantage: they already have the merchant relationships through Google Shopping and the product data infrastructure through Merchant Center. The weakness: Google has a history of launching protocols and then losing interest.
Visa’s Intelligent Commerce Connect
Visa’s approach is the most pragmatic of the three. Rather than building a new protocol, they built a bridge. Intelligent Commerce Connect is protocol-agnostic — it supports ACP, UCP, the Trusted Agent Protocol, and the Machine Payments Protocol (MPP). Merchants integrate once through Visa’s Acceptance Platform and can accept agent-initiated payments regardless of which protocol the agent uses.
This is a smart play. While OpenAI and Google fight over which protocol wins, Visa is positioning itself as the payment layer that works with all of them. They handle tokenization, spend controls, authentication, and — critically — they support both Visa and non-Visa cards. That last detail is easy to miss but important: Visa isn’t trying to lock this down to their own network.
The real insight in Visa’s strategy is that they understand the payment infrastructure matters more than the protocol. Whoever controls the token vault and the authentication layer has leverage regardless of which protocol emerges as the standard.
What This Means If You Run a Business
If you sell anything online, agentic commerce isn’t optional to think about anymore. It’s not here in full force yet, but the infrastructure is being laid right now, and businesses that prepare early will have an advantage.
Your Product Data Needs to Be Agent-Readable
AI agents don’t browse your website like humans do. They query structured data feeds. If your product catalog isn’t available in a machine-readable format with accurate descriptions, real-time pricing, and inventory status, agents simply won’t find your products. It’s search engine optimization all over again, except the “search engine” is an AI agent with no patience for ambiguity.
This means investing in structured data — detailed product schemas, accurate categorization, and keeping your feeds current. If you’re on Shopify, you’re in decent shape since they’re deeply integrated with both ACP and UCP. If you’re on a custom platform, you’ll need to build out these feeds yourself or use a service that generates them.
The New SEO Is Agent Optimization
When an AI agent is shopping for “trail running shoes under $150 with good ankle support,” it’s not clicking through ten blue links. It’s querying product catalogs directly and ranking results based on how well they match the user’s stated intent.
The implications: keyword stuffing your product descriptions is even more useless than it already was. What matters is accurate, detailed product attributes — dimensions, materials, use cases, compatibility — that let an agent match your product to a specific customer need. Think of it as moving from “marketing to humans” to “being discoverable by algorithms.” Which, honestly, is what SEO always was. The difference is that the algorithm now understands natural language.
Margins Get Squeezed (Again)
OpenAI’s 4% fee on Instant Checkout is just the beginning. As more platforms enable agent purchasing, there will be another hand in the cookie jar between you and your customer. Combined with existing payment processing fees, marketplace fees, and advertising costs, margins on commodity products will get even thinner.
The counter-argument: agent commerce could reduce customer acquisition costs significantly. If an AI agent recommends your product because it genuinely matches what the customer needs, you didn’t have to pay for that click. The question is whether the platform fees offset the savings on advertising. Too early to tell.
The Privacy Problem That Nobody Has Solved
For agentic commerce to work at scale, AI agents need access to your payment credentials, your purchase history, your sizing information, your delivery preferences, and your budget constraints. That’s a lot of sensitive data flowing through systems that didn’t exist six months ago.
Visa’s approach uses tokenization — the agent never sees your raw card number, just a token that represents it. That’s the same technology behind Apple Pay and Google Pay, and it’s well-proven. But tokenization only solves the payment piece. Your shopping preferences, purchase patterns, and personal constraints live somewhere in the agent’s context, and the privacy policies around that data are still being written.
OpenAI’s privacy policy for Instant Checkout states they don’t use purchase data to train models, but they do store transaction information for customer support and compliance purposes. Google hasn’t published detailed privacy documentation for UCP yet. And the smaller players in the agentic commerce space? Their privacy practices range from “reasonable” to “whatever our lawyers could write up in a week.”
If you’re a consumer, the practical advice right now is to treat agent-initiated purchases the same way you’d treat any digital wallet: use it for routine, low-stakes purchases where convenience outweighs the privacy trade-off. Keep high-value or sensitive purchases manual until the trust frameworks mature.
When This Actually Hits Mainstream
The infrastructure is being built now, but full mainstream adoption is probably 18-24 months away. Here’s a rough timeline based on what’s already been announced and the typical pace of payment industry rollouts:
Right now (Q2 2026): Visa’s Intelligent Commerce Connect is in pilot. ChatGPT Instant Checkout works for select merchants in the US. Google UCP is live with a handful of retailers. Early adopters are experimenting.
Late 2026: Expect broader merchant rollouts as Shopify, Stripe, and the major payment processors build out their integrations. Agent-initiated purchases will work reliably for common product categories like electronics, fashion, and home goods.
2027: The protocol landscape likely consolidates. Either ACP and UCP converge into something interoperable, or one wins enough market share to become the de facto standard. Visa’s bridge approach means merchants who integrated through Intelligent Commerce Connect are hedged either way.
2028-2030: McKinsey’s $1 trillion US market projection assumes broad consumer adoption by this period. Agent-initiated purchases become as normal as one-click ordering is today. The checkout button doesn’t disappear, but it becomes a fallback rather than the default.
The Bigger Picture
Agentic commerce isn’t just about making shopping more convenient. It reshapes how brands reach consumers by inserting an AI intermediary that optimizes for the buyer’s interests, not the seller’s conversion funnel.
Think about what that means for advertising. If consumers increasingly delegate purchase decisions to AI agents, then the agent becomes the audience, not the person. Brand awareness still matters — agents will factor in user preferences and brand affinity — but the hard sell, the urgency tactics, the dark patterns in checkout flows? Those don’t work on an algorithm.
Some businesses will hate this. Others will thrive because their products genuinely are the best match for customer needs, and an unbiased AI agent will surface them more consistently than the current ad-driven discovery model ever could.
The protocol war will sort itself out. The payment infrastructure will mature. The privacy frameworks will (eventually) catch up. What won’t change is the direction: commerce is moving from “browse, compare, buy” to “describe what you need and approve the result.”
Whether that future arrives in 18 months or three years, the smartest thing you can do right now is make sure your products are agent-discoverable. Update your structured data. Get on platforms that support ACP or UCP. And start thinking about what your value proposition looks like when the customer never visits your website.
The checkout button had a good run.