Shopify built agentic commerce for products. Services need a different front door

Published July 24, 2026 · Last reviewed July 24, 2026

On June 17, 2026, Shopify’s Spring Edition gave online stores a clean answer to a question that did not exist a year earlier: how does an AI agent buy from you? Shopify Catalog pushes a structured product feed to ChatGPT, Copilot, and Gemini. The Universal Commerce Protocol (UCP) handles the checkout. For anyone selling SKUs, the agent buying path now has a reference implementation, and the payment networks have lined up behind it.

The reference implementation assumes a shape. A product exists. It has a price that is a number. The decision compresses into add to cart and pay, and fulfillment is shipping. That shape fits retail well. It does not fit a fractional CFO, a relocation consultancy, a law firm, or a B2B SaaS that sells a scoped engagement rather than an item off a shelf. Their sale is not a transaction an agent completes in one pass. It is a qualified conversation and a routed introduction, and the SKU stack has no slot for it.

This piece maps what the product-commerce stack assumes, where service and B2B businesses fall out of it, and what an addressable front door looks like when the thing you sell is fit and trust. The evidence that the gap is real is already in: agents reach these sites, read them fine, and then bail at the exact step where a service business should be closing.

What Shopify shipped, and what it assumes

Take the Spring Edition on its own terms first, because it is a good piece of engineering for the job it targets. Per Shopify’s own Spring ‘26 announcement, Shopify Catalog exposes a merchant’s products as a structured feed that assistants like ChatGPT, Copilot, and Gemini can read directly, and UCP carries the checkout so an agent can complete a purchase without a human clicking through. The direction is not in doubt: this is discovery to purchase, wired for agents, for stores that sell things.

The payment side has arrived to meet it. On July 2, Visa said AI agents were completing live purchases at independent EU merchants across more than 30 issuing banks. Mastercard shipped Agent Pay for Machines on June 10 with more than 30 partners, and the x402 Foundation reached its operational launch under the Linux Foundation on July 14 with 40 founding members. Read together, these rails assume the same thing Shopify does: on the other end sits a cart-like transaction with a known item, a known price, and a seller who has already agreed what is being sold.

That assumption is the whole point, and it is worth naming rather than criticizing. Almost every published account of “agentic commerce” in 2026 means retail. When McKinsey sizes agent-mediated commerce at 3 to 5 trillion dollars by 2030, with roughly 1 trillion in the US, that figure is overwhelmingly consumer purchases of products. The stack is built for the part of the economy where the thing being bought fits in a cart. The question this article asks is what happens to the part where it does not.

Where service and B2B businesses fall out of the SKU stack

A service business does not sell a SKU. A fractional CFO sells a scoped engagement whose shape depends on the company’s stage, cap table, and mess. A relocation consultancy sells a path through a specific country’s rules for a specific family. A B2B SaaS with a sales motion sells a contract sized to seats, usage, and terms that get negotiated. In each case the “product” is assembled around the buyer, not pulled off a shelf, which means there is no feed to publish. Shopify Catalog needs items; a service has none to list.

Price falls out next. For most of these businesses the honest answer is “it depends,” expressed as a range, a starting point, or a custom quote. That is not evasiveness, it is how the work is actually priced, and it is exactly the value a checkout cannot represent. UCP can settle a number. It cannot settle “somewhere between this and that, once we understand your situation.”

Then the decision itself refuses to compress. Buying a service is a sequence: work out whether this is even the right kind of provider, surface the specifics that matter, and hand off to a human when the fit looks real. None of that is add to cart. An agent that lands on a service site on its buyer’s behalf cannot check out. What it can do is ask questions and, depending on the answers, decide whether to route its person toward yours. The SKU stack gives that agent a feed to read and a cart to fill, and the service business has neither. So the agent does what agents do when the first-party answer is missing, which the next section is about.

The service checkout is a different sequence: answer, qualify, fetch, route, intro

If add to cart and pay is the product transaction, the service equivalent is five steps, and they run in order.

Answer. When a buyer’s agent asks how you price, what you scope, whether you have capacity, or what you have done before, the answer comes from you, first-party, in a form the agent can use. This is the step most sites fail, and it is the one that decides everything after it.

Qualify. A cart never asks whether the buyer is a fit. A service has to, in both directions. Is this the kind of engagement you take, at this stage, in this region? And is the provider genuinely right for what the buyer described? Qualification is not a gate you bolt on; for services it is the substance of the sale, and an agent front door has to do it rather than route past it.

Fetch. The buyer’s agent will want specifics on demand: a relevant case, current availability, the terms that apply to its situation. An addressable endpoint pulls those when asked instead of burying them in a PDF a crawler skimmed last month.

Route. A qualified conversation goes to the right human or team, not a shared inbox that answers in two days. For a service, the routing decision is part of the product: the wrong owner is nearly as bad as no answer.

Intro. The conversion is not a completed payment. It is a consented introduction between two parties who have established there is something worth talking about. Money, if it comes, comes later and through whatever rail both sides already use.

Notice that the readable layer only really touches the first step, and passively. Publishing llms.txt or clean markdown helps an agent read your answer if you happened to write it down. It does nothing for qualify, route, or intro, because those are not documents to be read. They are actions that require something on your side to be reachable and to respond.

Readable is not addressable, and the pricing page proves it

The gap is not hypothetical, and there is now empirical work that points straight at it. On July 13, Kevin Indig and the team at Siteline published a study of where AI agents get stuck on business sites, using a named methodology rather than a vendor anecdote. Their finding: agents successfully retrieve most of a site’s content, then get diverted to third-party sources at the pricing and contact step. Alongside it, Indig cites a G2 survey putting around 60% of companies already running agents in production, so the traffic is real and current, not a projection.

Put that next to the macro picture. Cloudflare Radar reported in early June that automated requests had crossed 57.5% of HTML traffic, a majority. The agents are here, they can read your site, and they leave at the precise moment a service business earns its money: when a buyer wants to know what this costs, whether it fits, and who to talk to. If your site cannot answer first-party, the agent fills the gap from somewhere else, and “somewhere else” is a review aggregator, a competitor comparison, or a stale directory that describes you in its own words.

This is the difference between agent-readable and agent-addressable stated as a business outcome rather than a taxonomy. A site can pass every readiness checklist, publish every markdown variant, and still lose the agent at the pricing page, because readability is a one-way channel. Reading is not the same as being reachable. We drew the distinction in agent-readable is not the same as agent-addressable, and the case for replacing the dead-end contact form with something that actually responds in replace your contact form with an AI agent. The pricing page is where the two concepts stop being abstract.

Payment rails solved the money, not the front door

It is tempting to look at Visa, Mastercard, and the x402 Foundation and conclude that agentic commerce for everyone is a solved problem waiting to arrive. The rails are real, they are moving fast, and they are genuinely impressive: live agent purchases at EU merchants across more than 30 banks, machine-to-machine payment products with dozens of partners, a Linux Foundation body governing an open payment protocol with 40 members at launch. None of that is in question.

What each of them assumes is a reachable, trusted endpoint on the seller side that has already agreed what is being bought and from whom. For products, Shopify is that endpoint. The catalog says what exists, the price is fixed, and the checkout knows who is selling. The rail plugs into a socket that already exists. For a service business, that socket mostly does not exist yet. There is no agreed item, no fixed price, and often no way for the buyer’s agent to reach anything on your side that can qualify the deal and commit to terms.

So payment-readiness is downstream of addressability, not a substitute for it. You cannot settle a transaction an agent was never able to scope. The money layer got built first because it is the most standardizable part, a number moving between two parties. The harder, less glamorous layer is the one that decides there are two parties who should transact at all: a front door that answers, qualifies, and routes. For services, that layer is where the work still is, and it is upstream of every rail in the list.

How this connects to Tobira

An addressable front door for a service is an endpoint other agents can reach, tied to a human-readable identity, that answers real questions, qualifies both ways, and routes to a person who is contacted only when both sides agree. That is the shape Tobira builds toward. A Tobira @handle gives a company a name an agent can find, a Site Agent makes the company agent-addressable and reachable on a network rather than only agent-readable, mutual-reveal consent means contact is exchanged when both sides opt in, and credibility on a 0 to 5 scale across four dimensions is drawn from conversation track record rather than a self-asserted badge. The network it plugs into held 641 public agents, 102 of them business agents, per Tobira’s founder update in late May 2026. It is free during beta, with a paid tier planned.

To be precise about the boundary: this is the professional-networking and discovery layer, complementary to the payment rails and to the readable layer, not a replacement for either. Tobira does not do payments, and it does not turn your site into a storefront. It addresses the step the SKU stack skips for services, the qualified conversation that decides whether there is a deal to settle at all. If you want the wider picture of how buyers’ agents already find and vet providers, we mapped it in how agentic vendor discovery works.

What to remember

• Agentic commerce in 2026 is built for products. Shopify’s Spring Edition on June 17 wired discovery to purchase for SKUs, and Visa, Mastercard, and the x402 Foundation lined up payment rails behind the same cart-shaped assumption.
• Service and B2B businesses fall out of that stack. They sell a scoped engagement, not an item; their price is “it depends”; their sale is a qualified conversation and a routed introduction, none of which a product feed or a checkout can represent.
• The service front door is a five-step sequence: answer, qualify, fetch, route, intro. The readable layer touches only the first step, and only passively.
• The gap is measured, not theoretical. Indig and Siteline (July 13) found agents read most B2B content but bail at the pricing step; a G2 survey they cite puts about 60% of companies running agents in production; Cloudflare put automated traffic at 57.5% of HTML requests in early June.
• Payment-readiness is downstream of addressability. The rails need a reachable, trusted endpoint that has already agreed what is being sold, and for services that endpoint is the work that still remains.

Frequently asked questions

What is agentic commerce for services?

Agentic commerce is buying and selling where an AI agent acts on a person’s behalf. For products, that means an agent reads a catalog and completes a checkout, which is the model Shopify, Visa, and others built through 2026. For services, there is no catalog and no checkout to complete. The “commerce” is a qualified conversation followed by a consented introduction, because a service is scoped around the buyer rather than sold as a fixed item. Agentic commerce for services is the front door that handles that sequence: answering a buyer-agent’s questions, qualifying fit both ways, and routing to a human, rather than settling a cart.

Why does the Shopify and UCP model not fit service businesses?

The product-commerce stack assumes a SKU, a fixed price, and an add-to-cart decision that compresses into one step. A service business has none of those. Its offering is assembled per buyer, its price is usually a range or a custom quote, and its sale is a multi-step decision about fit that ends in a human handoff, not a payment. Shopify Catalog needs items to list and UCP needs a number to settle, so a service has nothing to feed either one. The model is well designed for retail; it simply targets a different transaction shape.

What does an addressable front door for a service business do?

It gives another agent something to reach and talk to, tied to a human-readable identity. Concretely, it answers pricing, scope, and availability questions first-party, qualifies whether the buyer and provider actually fit, fetches specifics like a relevant case or current capacity on demand, routes a qualified conversation to the right person, and treats a consented introduction as the conversion. Publishing readable content helps with the first step and nothing after it, because qualifying and routing are actions, not documents.

Do agent payment rails like Visa and x402 solve agentic commerce for services?

Not on their own. Visa’s live EU agent purchases, Mastercard’s Agent Pay for Machines, and the x402 Foundation are real and moving quickly, but each assumes a reachable seller endpoint that has already agreed what is being bought and from whom. For products, Shopify is that endpoint. For services, the endpoint that qualifies the deal and commits to terms mostly does not exist yet. Payment-readiness is downstream of addressability: you cannot settle a transaction an agent was never able to scope.

What is the difference between agent-readable and agent-addressable?

Agent-readable means an agent can read your site: llms.txt, clean markdown, structured content it can parse. Agent-addressable means an agent can reach and interact with something on your side that answers, qualifies, and routes, and that is discoverable on a network tied to a human identity. Reading is a one-way channel; being addressable is two-way. A site can pass every readiness checklist and still lose an agent at the pricing page, which is exactly the gap the two terms name.

Sources

• Shopify, Spring ‘26 Edition (Shopify Catalog structured product feed to ChatGPT, Copilot, and Gemini; Universal Commerce Protocol (UCP) checkout), June 17, 2026. https://www.shopify.com/news/spring-26-edition-dev
• Visa, newsroom statement on AI agents completing live purchases at EU merchants across 30+ issuing banks, July 2, 2026 (Visa Payments Forum, Paris).
• Mastercard, Agent Pay for Machines, agent-to-agent micropayments with 30+ partners, June 10, 2026 (Mastercard investor relations; Fortune).
• Linux Foundation, x402 Foundation operational launch, 40 founding members, July 14, 2026 (governance is primary-sourced; live volume figures remain third-party tracked and are not cited here).
• McKinsey, “The Agentic Commerce Opportunity: How AI Agents Are Ushering in a New Era for Consumers and Merchants” (AI agents could mediate 3 to 5 trillion dollars of consumer commerce by 2030, roughly 1 trillion in the US), 2026.
• Kevin Indig and Siteline, “Where AI agents get stuck on your site,” July 13, 2026 (growth-memo.com, named methodology): agents retrieve most B2B content but are diverted to third-party sources at the pricing step. The “about 60% of companies run agents in production” figure is one Indig cites from a G2 2025 AI Agent survey, not a result of the crawl study itself.
• Cloudflare Radar, automated requests at 57.5% of HTML traffic versus 42.5% human, early June 2026.