In a single week, Shopify wired storefronts into ChatGPT, Santander ran live agentic payments with Visa, and Plaid connected brokerage accounts to an AI platform. Three layers of the commerce stack, all agent-ready, all at once. This is the week the infrastructure arrived.

For the past year, agentic commerce has been the industry's favourite thought experiment. Protocols were announced. Whitepapers were published. Pilots were launched in sandboxes that never touched real money.

This week was different. Three announcements, from three different companies, across three different layers of the commerce stack, all landed within days of each other. None of these companies coordinated. That is precisely the point.

The infrastructure for AI agents that discover products, initiate payments, and access financial data is assembling itself, layer by layer, without a central architect. And now, for the first time, every layer is operational.

Here is what happened, and why the convergence matters more than any single announcement.

The storefront layer. Shopify announced "agentic storefronts" that make merchants' products discoverable and purchasable directly inside ChatGPT. Meanwhile, OpenAI retreated from building its own Instant Checkout, handing the transaction to Shopify. The AI platform becomes the discovery engine. The commerce platform keeps the sale. As we explored in our analysis of the Shopify/ChatGPT deal, this sets the template for how AI platforms and commerce platforms divide the value chain.

The payment layer. Santander completed live agentic payment transactions with Visa across multiple Latin American markets. Real money, real networks, AI-initiated. Days earlier, JP Morgan Payments partnered with Mirakl on a joint agentic commerce solution. As we detailed in our coverage of the Santander/Visa milestone, this is no longer a sandbox exercise. The world's largest financial institutions are processing agent-initiated transactions.

The data layer. Plaid connected brokerage accounts to Perplexity Computer, an AI platform orchestrating 20 frontier models. On the same day, Truist Financial expanded its open banking partnership with Plaid. As we examined in our analysis of the Plaid/Perplexity integration, open banking rails are becoming the default plumbing through which AI agents access financial lives.

Discovery. Transaction. Data. All wired for autonomous AI agents, all in one week.

Any one of these announcements would be significant on its own. Together, they describe a complete commerce stack for AI agents.

Consider what an AI agent can now do, using only the infrastructure announced this week: it can discover a product inside a ChatGPT conversation (Shopify agentic storefronts), initiate a payment through a major card network (Santander/Visa), and access the consumer's financial data to inform the decision (Plaid/Perplexity).

No single company built this stack. It emerged from independent decisions by companies that each concluded the same thing: agents are going to transact, and the infrastructure needs to be ready. The fact that all three layers came online in the same week is not coordination. It is convergence.

According to Total Retail, 70 percent of retail companies are either piloting or have deployed agentic AI in some capacity. But there is a critical difference between deploying AI for internal operations and being ready for AI agents that show up as customers. This week's announcements close that gap from the infrastructure side.

The companies positioned to define agentic commerce are those that control the boundaries between layers.

Shopify controls the merchant relationship. Millions of merchants already use its infrastructure. Agentic storefronts extend that relationship into AI platforms without requiring merchants to rebuild anything. The merchants stay with Shopify. The customers just arrive differently.

Visa and Mastercard control payment authorisation. Visa is proving the operational model with live transactions. Mastercard is building the trust layer with Verifiable Intent. The network that gets both right first becomes the default rail.

Plaid controls financial data access. It connects over 12,000 financial institutions. As AI platforms proliferate, Plaid becomes not just fintech infrastructure but the data access layer for agentic finance.

Each of these companies sits at a chokepoint. Each will extract value from the agent-initiated traffic that flows through their layer. And each is building the connective tissue that makes the others' layers more valuable.

The companies most at risk are those that depended on human attention to drive commerce. Brands that built their business on search rankings, display advertising, and loyalty programmes. AI agents do not scroll. They do not see banner ads. They do not have brand preferences unless they are explicitly programmed to. They optimise for the consumer's stated intent, not the merchant's marketing budget.

As we explored in our analysis of the agent tax, the economics of selling through AI intermediaries are fundamentally different from selling through search or social. The cost layers change. The discovery mechanism changes. The customer relationship may not transfer at all.

A complete stack does not mean a mature stack. Three critical gaps separate this week's milestone from mainstream adoption.

Security. OpenAI published a paper this week on designing agents to resist prompt injection, acknowledging that the attack surface expands dramatically when agents can transact. Its acquisition of Promptfoo, an AI security platform, underscores the threat. An agent that can buy things is an agent worth hijacking. An agent that can see your brokerage account is an agent worth compromising.

Authentication. As we examined in Who Authorised the Agent?, the liability chain breaks when the entity initiating a transaction is not the entity bearing the financial consequence. Santander's live transactions make this problem immediate. When an AI agent spends your money and something goes wrong, the current dispute resolution infrastructure has no clear answer for who is responsible.

Standards. Multiple competing protocols exist for agent-to-merchant and agent-to-payment-network communication. As we covered in our analysis of the agentic commerce standards race, the companies that set the standards will set the economics. This week's live deployments will intensify that competition.

Regulation. Governments have not caught up with AI agents as economic actors. Liability frameworks, consumer protection rules, and dispute resolution mechanisms all assume a human made the purchase. That assumption is already outdated.

Before this week, sceptics could reasonably argue that agentic commerce was vapourware: impressive demos, ambitious announcements, and no real transactions.

That argument is over.

Live payments are flowing through Visa's network. Storefronts are moving inside ChatGPT. Brokerage data is accessible to AI platforms through regulated open banking rails. The infrastructure exists. The gaps are real but addressable. The direction is irreversible.

The companies that recognised this early, the layer owners, the infrastructure providers, the protocol designers, have a head start. The companies that assumed they had more time do not.

When AI agents can discover, buy, and access financial data without human intervention, who is really in control of the transaction?