Agentic Commerce Is Rewriting Who Controls the

The demo is irresistible. You tell your AI agent to buy you coffee. It buys you coffee. It even asks politely before charging your card. The whole thing takes thirty seconds and feels like a magic trick.

That's exactly the problem with treating it as the story.

Stripe's recent cascade of agent commerce announcements—Link wallet for agents, shared payment tokens, a machine payments protocol, the full agentic commerce suite—generated a predictable wave of coverage about AI agents making purchases. What got significantly less attention is the structural question those products are answering: what happens to commerce when the buyer's agent arrives at a merchant's door already knowing what it wants to buy, already holding payment authority, and with zero interest in being marketed to?

That's not a checkout story. That's an architecture story.

The Funnel Was Never Really About Selling

To understand why this matters, it helps to remember what the traditional sales funnel actually was. Not the diagram—the institution.

Every website, pricing page, demo form, and abandoned cart email in the history of internet commerce served a single underlying function: making human intent observable inside a seller-controlled environment. The buyer searched, clicked, hesitated, returned, compared, and eventually converted—and at every step, the seller was watching, measuring, and optimizing. More than 8,000 companies in the martech space during the 2010s existed almost entirely to exploit that visibility.

The funnel wasn't a marketing tool. It was a surveillance apparatus with a checkout at the end. We just called it UX.

Nate B. Jones, who covers AI strategy and broke down Stripe's announcements in a detailed recent video, frames it this way: "The old internet asked, 'How do we get the customer into our store?' The next internet asks, 'How do we become usable by the customer's agent when the customer never comes to the store at all?'"

That shift—from attracting to being usable—sounds subtle until you sit with what it demolishes. Performance marketing, conversion rate optimization, landing page A/B tests, retargeting, lifecycle email sequences: all of it was built on the assumption that a human being would enter a seller's environment in a state of incomplete decision-making and could be nudged toward purchase. An AI agent doesn't browse. It doesn't get nudged. It arrives with a brief.

The "Authentic Coffee" Problem

The best illustration of how differently agents process commercial intent comes from an example Jones walks through: asking an agent to buy "authentic coffee."

Type that into a search engine and you get a keyword problem—whatever content has ranked for adjacently related terms. Type it into Amazon and you get a ranking problem—whatever products paid or earned their way to the top. The phrase is essentially useless to seller-controlled infrastructure because seller-controlled infrastructure was built to surface what sellers want you to see, not to interpret what you actually mean.

But to a well-built agent that knows you, "authentic coffee" becomes something else entirely. Jones describes it as "a purchasing brief"—the agent can translate that vague phrase into origin country, roast level, processing method, flavor profile, freshness window, roaster reputation, brew compatibility, price range, and prior purchase history. The buyer's language stays fuzzy. The commercial intent becomes precise. And critically, the seller never got to shape any of it.

This is why the common framing of "agentic visibility" as "SEO for agents" misses the point so badly. Search engine optimization was about winning a human's attention in a seller-friendly environment. Agent-readiness is about being legible to software that is explicitly not trying to be persuaded—it's trying to match. As Jones puts it: "A person will tolerate ambiguity. A person will infer from aesthetics. A person will click around... All an agent needs is for the business to be legible enough to operate against."

That's a meaningfully higher bar. It means product catalogs, pricing, return policies, fulfillment constraints, inventory levels, and service tiers all need to become machine-readable not as a nice-to-have but as a baseline condition of being found.

Payment Authority Goes Nomadic

The second structural shift is where Stripe's product announcements get genuinely interesting from an infrastructure perspective.

In the old model, payment authority was extracted inside the seller's environment. The buyer arrived, shopped, and handed over payment details at checkout. The seller's funnel was literally the place where the buyer's wallet opened.

Link wallet for agents inverts this. A user grants a buyer's agent programmatic access to their payment credentials. The agent generates a spend request, the user approves it, and Link returns either a one-time card (for operating against the web as it currently exists—Jones calls this "an adapter for the existing commercial internet") or a shared payment token that points toward a more natively machine-to-machine transaction model. The agent never touches raw credentials. The payment authority travels with the task.

What that means in practice: a merchant may increasingly find that when a transaction arrives, the commercial decision already happened somewhere else. The buyer never browsed. The seller never had an opportunity to present alternatives or upsell. The transaction shows up pre-authorized.

Jones is direct about the significance: "The seller may not be receiving a browsing customer. The seller may be receiving an authorized purchasing attempt by a bot."

That's not science fiction. That's what the product is designed to enable.

The trust architecture this requires is genuinely complicated—the buyer trusts the agent, the seller trusts the transaction, the agent platform trusts the seller, the wallet protects the credentials, the payment network enforces the controls, and the user needs enough context to approve actions without becoming a bottleneck. Each of those trust relationships has to work. The failure modes are non-obvious and not yet well-understood at scale.

Not Just Stripe

What makes this more than a single company's product launch is the convergence happening across the entire commercial infrastructure stack. Microsoft is pushing shopping into Copilot. Meta is collapsing the distance between ads and checkout. Visa and Mastercard are building their own agent payment and tokenization systems. PayPal is building commerce services around wallet trust and merchant protection. Google's Universal Commerce Protocol is extending Merchant Center with product attributes—compatible accessories, substitutes, answers to common product questions—that look less like search keywords and more like structured data for agent reasoning.

They're all building toward the same thing: commerce that begins inside the buyer's interface rather than the seller's storefront.

The Walmart-ChatGPT instant checkout experiment is instructive here, even in its failure. According to reporting cited by Jones, it converted three times worse than products where ChatGPT sent the shopper back to Walmart's own website. Walmart's product and design lead called the experience "unsatisfying." OpenAI's own post-mortem acknowledged the initial version lacked flexibility, and it shifted the product toward discovery rather than direct purchase.

That result doesn't invalidate agentic commerce—it clarifies what the actual problem is. People aren't waiting to buy things inside a chat window. They're waiting for an agent that understands their preferences well enough to handle the commercial legwork before they have to think about it. Instant checkout was a feature in search of a context. The larger shift Jones is describing is a context in search of features—and those features are now being built simultaneously by every major player in payments and commerce.

The persuasion surface—the carefully designed journey from "I'm vaguely interested" to "I purchased"—is compressing. In some cases, it may disappear entirely, replaced by a buyer's agent that already has a theory of what you want and the payment authority to act on it.

For merchants, that means the question is no longer "how do I convert this visitor?" It's "am I even in the room when the decision gets made?"

By Marcus Chen-Ramirez, Senior Technology Correspondent