Last week’s issue argued that AI commerce is being constrained by accountability, not discovery. That remains true. But this week showed something just as important: the market is not waiting for a perfect accountability layer before it rebuilds the merchant side of commerce for machines. The real movement is happening one layer lower. Merchants, platforms, marketplaces, and payment networks are starting to reformat catalogs, checkout rails, and operating systems so agents can actually use them.
That is a material narrative shift from the previous issue. The question is no longer only whether machine-driven commerce can be trusted. The question is whether merchants can make themselves legible, callable, and executable inside machine-led environments before those environments become the default buying layer. This week’s signals point to the same conclusion from different parts of the stack: discovery is being reformatted as structured data, checkout is being reformatted as protocol compatibility, and store operations are being reformatted as agent-executable actions.
The importance of that shift is easy to underestimate because much of it looks like tooling. It is not just tooling. It is the early construction of a machine-facing commerce layer. When that layer matures, the merchant that still thinks in pages, campaigns, and manual dashboards will not merely look outdated. It will become harder for systems to discover, compare, transact with, and optimize on that merchant’s behalf.
Visa is positioning itself as the protocol router for agentic checkout
On 8 April, Visa launched Intelligent Commerce Connect, a new layer inside the Visa Acceptance Platform that gives businesses a single integration for agent-initiated payments. The important detail is not just that Visa wants to support AI-driven shopping. It is that the product is explicitly network-, protocol-, and token-vault-agnostic. Visa says it can support major agent protocols including Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol, and Universal Commerce Protocol, while also helping make merchant catalogs discoverable inside AI platforms and handling orchestration and PCI compliance for enablers processing transactions on merchants’ behalf.
Structurally, this means the payment layer is moving up the stack. Visa is not waiting to be the final settlement rail after someone else defines agentic commerce. It is trying to become the compatibility layer that reduces protocol fragmentation for merchants and agent builders. That matters because the hardest part of early machine commerce is not only trust. It is integration chaos. If a major payment network can abstract that chaos behind one merchant connection, the balance of power shifts away from whichever protocol gets the most attention and toward whichever infrastructure provider can make fragmented protocols operational at scale.
What this changes is the merchant decision model. The old question was, “Which AI shopping platform should we support?” The new question is, “Which infrastructure partner can make us compatible across many agent environments without forcing repeated rebuilds?” The second-order implication is even bigger: payment incumbents can become control points for discoverability, authentication, spend controls, and checkout execution in agent-led commerce, even when they do not own the consumer interface.
Break: Direct integrations stop being a strategic moat when the payment layer turns protocol fragmentation into a service.
Source: Visa announcement
Shopify is turning the merchant backend into an execution surface for AI
On 9 April, Shopify released its AI Toolkit, which connects AI tools to Shopify documentation, API schemas, code validation, and store management through CLI store-execute capabilities. Shopify also released a plugin that auto-updates so the connected agent stays current as new capabilities ship, and it listed support for tools such as Claude Code, Cursor, Gemini CLI, VS Code, and Codex CLI.
This is more important than another developer convenience release. Shopify is moving from “AI can advise the merchant” to “AI can operate the merchant stack.” That is a structural shift. A merchant backend used to be a human interface where a person navigated menus, edited products, and manually triggered changes. Shopify is now formalizing the store as something agents can understand and act on through machine-readable schemas, validated code paths, and execution permissions. In plain language, the store is becoming callable.
What this changes is how commerce operations scale. If AI can directly execute store tasks across product data, merchandising, code, and store management, the unit of work is no longer “a team member updating the store.” It becomes “an agent acting against structured store capabilities.” The second-order implication is that platform advantage increasingly comes from how safely and cleanly a commerce system exposes actions to machines, not just from how easy the dashboard feels to humans. That raises the bar for every commerce platform that still treats AI as a sidebar assistant instead of an operational interface.
Break: The admin dashboard stops being the primary operating surface when the store itself becomes executable by agents.
Source: Shopify changelog
Amazon is converting its marketplace into an intake layer for off-Amazon inventory
Amazon announced that merchants can now connect to its AI-powered Shop Direct experience through third-party product feeds from Feedonomics, Salsify, and CEDCommerce. Shop Direct already includes over 100 million products from more than 400,000 merchants, and Amazon says tens of millions of products are available through the Buy for Me feature, where Amazon can complete the purchase on the customer’s behalf using encrypted payment and shipping information. The same synced feed can surface products in traditional search results and in Rufus.
The structural importance here is easy to miss if this is read as an acquisition tactic. It is bigger than that. Amazon is lowering the cost of becoming ingestible by its AI shopping layer without requiring merchants to rebuild a native Amazon retail presence for every product. That means Amazon is widening its role from marketplace owner to discovery and transaction broker for inventory that lives elsewhere. Once that model scales, the boundary between “inside Amazon” and “outside Amazon” matters less than whether Amazon can index, route, and intermediate demand.
What this changes is the merchant’s relationship to marketplaces. The old model assumed platform expansion required listing natively, managing the marketplace stack, and often sacrificing direct control. This new model says a feed, real-time inventory sync, and agent-compatible purchase path may be enough to participate. The second-order implication is that distribution will increasingly favor infrastructures that can ingest merchant data quickly and turn it into machine-usable selection across both search and assistant surfaces. The winning marketplace may not be the one that owns the most inventory. It may be the one that can broker the most machine-usable inventory.
Break: Marketplace participation stops starting with listing pages when a synced feed is enough to make external inventory discoverable and purchasable.
Source: Amazon Shop Direct update
Google just made feed quality an AI visibility requirement, not a merchandising hygiene task
On 8 April, Google’s Ads & Commerce team stated plainly that AI-driven shopping experiences such as conversational shopping in AI Mode, virtual try-ons, and shoppable connected TV depend on the basic product data merchants provide to Google. Its message was direct: if a Merchant Center feed is messy or incomplete, customers will not be able to find those products.
That is not a minor operations note. It is an eligibility statement. Google is effectively telling merchants that product data quality is now part of interface access. In older commerce models, weak feeds hurt ranking, ad quality, or merchandising efficiency. In this model, weak feeds reduce a merchant’s ability to exist inside AI-native buying experiences at all. The feed is no longer a backend asset that supports commerce. It is part of the storefront that the model sees.
What this changes is the meaning of product ops. Titles, attributes, variants, availability, and pricing are no longer just channel-management details. They are machine-legibility infrastructure. The second-order implication is severe: the merchant with the best brand, site, and campaign may still lose machine-led discovery if its product data is incomplete, inconsistent, or poorly structured. In AI commerce, representation quality becomes shelf space.
Break: Product feeds stop being cleanup work when they determine whether AI systems can see you at all.
Source: Google Ads & Commerce
Commercetools is betting that protocol volatility will create a new orchestration market
On 9 April, commercetools framed AI Hub as the answer to a fast-moving, unstable agentic commerce landscape. Its argument was explicit: building separate integrations for each AI channel is slow, risky, and likely to age badly as protocols and consumer behaviors shift. AI Hub’s role is to translate products in a commercetools project into formats compatible with multiple AI channels and to let brands support a complementary stack of protocols, including ACP, MCP, and UCP, rather than betting on a single one.
This matters because it reveals where a new layer of enterprise value is forming. If the agentic ecosystem fragments into multiple protocols, assistants, and checkout paths, merchants will need an orchestration layer that keeps product data, checkout logic, and order flows consistent across all of them. That is not a temporary integration problem. It is a new systems category. Commercetools is saying, in effect, that agentic readiness will not be won by choosing the right front-end channel, but by building a translation layer between merchant logic and machine environments.
What this changes is the enterprise roadmap. Early AI commerce experiments are often treated like channel pilots. This signal says that approach will not hold. The second-order implication is that the next winners in commerce tech may be the platforms that can keep merchant logic stable while every external AI surface keeps changing. That is a different kind of defensibility than storefront design, search relevance, or loyalty tooling. It is cross-protocol control.
Break: A channel strategy stops working when every AI surface speaks a different protocol and your catalog has to travel across all of them.
Source: commercetools AI Hub and ACP guide
The System That Is Emerging
This week’s developments point to one structural truth: commerce is being reformatted so machines can read it, call it, and execute against it. That is the hidden layer beneath the announcements. Visa is abstracting protocol chaos at checkout. Shopify is exposing stores as executable systems. Amazon is widening its intake layer for external inventory through feeds. Google is making feed quality a condition for AI visibility. Commercetools is building a translation layer for protocol volatility. These are not isolated product moves. They are coordinated answers to the same systems problem.
What is shifting is the merchant representation model. For two decades, digital commerce assumed the storefront was the main interface and everything behind it was support infrastructure. That assumption is starting to break. In agent-led commerce, the structured catalog, the callable checkout, the protocol bridge, and the execution surface matter just as much as the page a human sees. Control is moving toward whichever systems can translate merchant intent into machine-usable form without losing pricing control, availability accuracy, payment security, or operational consistency.
For operators, the implications are practical:
- Product data quality is now a distribution decision, not a merchandising chore.
- Checkout compatibility is becoming a protocol and tokenization problem, not just a PSP decision.
- Commerce platforms will be judged by how safely they expose actions to agents, not only by human usability.
- Multi-channel AI participation will require orchestration layers that preserve merchant control while protocols keep changing.
- Marketplaces are learning to intermediate external inventory without fully owning it.
Core Truth: In AI commerce, the merchant that cannot be interpreted and operated by machines will lose distribution before it loses demand.**
Tool of the Week Feedonomics
|
|
This week’s issue is really about one thing: the quality and portability of merchant data. Feedonomics is relevant because it sits exactly at that layer. Its positioning is direct, “Transform wild product data into optimized, structured listings for ad channels, marketplaces, and AI surfaces,” and it frames agentic commerce as the process of turning discovery into checkout. That makes it more than a feed-management tool. It is a practical bridge between merchant catalogs and machine-facing discovery surfaces. In a market where Amazon is opening Shop Direct through feed partners and Google is making structured product data a condition for AI visibility, this layer becomes strategically important.
Source: Feedonomics
|
Trend to Watch
|
|
Watch for the separation of the merchant data layer from the merchant presentation layer. That sounds technical, but the commercial meaning is straightforward. More platforms will let merchants participate in AI-led discovery and transaction flows without requiring a full native storefront build on each destination. As that happens, structured product data, checkout endpoints, and orchestration logic will become portable assets that can travel across assistants, marketplaces, and payment environments. The website will still matter, but it will increasingly be only one expression of the commerce system, not the system itself. |
