What Goldman Sachs’ Agentic Commerce Report Means for Order Operations

Goldman Sachs Research published a long report on agentic commerce on September 24, covering everything from advertising and payments to fraud, identity and ticketing. Most of it is written for investors trying to work out which platforms, networks and retailers win as AI agents start shopping on consumers’ behalf. Read from the seat of a brand that has to ship the orders, one thread runs through the whole thing: as agents take over more of the shopping journey, the advantage shifts toward the companies that execute best after the click.

Goldman frames the central question as a change in what matters. For two decades, the fight in ecommerce was over who owns the storefront. In Goldman’s view, it is becoming a fight over who owns discovery, who owns the shopper’s intent, and who owns the infrastructure that turns that intent into a trusted, completed transaction. Pipe17 works in that third layer, so this post walks through the report’s main arguments and what each one means for order operations.

A slow shift that starts with the easiest purchases

Goldman expects agentic commerce to arrive gradually. The report compares it to the history of ecommerce itself, which took decades to win share from physical retail as trust, payments and delivery caught up. It expects agentic adoption to follow a similar long arc, moving category by category.

The first categories to move, in Goldman’s framework, are the ones where a mistake is cheap and preferences are simple: household supplies, packaged goods, groceries and similar replenishment purchases. Subjective, high-consequence categories such as apparel, furniture and luxury goods come last, because a wrong choice there is costly and personal. Goldman sizes the nearer-term opportunity in the highest-likelihood US categories at about $2.6 trillion of spending, and estimates that even low single-digit agentic penetration of those categories could add around a point to overall ecommerce growth.

For operations teams, the order in which categories adopt matters as much as the size. The first wave of agent-placed orders will look a lot like repeat and replenishment orders: frequent, predictable, and unforgiving of stockouts or late deliveries. Brands that already run subscriptions or auto-replenishment programs have a head start, because those programs depend on the same things agents will: accurate availability, reliable promise dates, and substitution rules when a SKU runs out.

Agents pick the merchants they can trust to deliver

The part of the report most relevant to brands is its view of how agents choose where to buy. Goldman notes that many agents are built to optimize for price, delivery speed, availability and assortment. It expects agents to gravitate toward merchants that offer the highest likelihood of a successful transaction: reliable fulfillment, deep inventory and consistent pricing.

Goldman draws a parallel with the shift from desktop to mobile. On mobile, the prize became the app install, because an installed app turned occasional visitors into repeat customers. In agentic commerce, Goldman suggests, the equivalent prize is becoming the default merchant an agent chooses again and again on a shopper’s behalf. Companies that earn that position could see higher retention and purchase frequency at lower acquisition cost.

The report’s conclusion on this point is the line that stood out most to us: over time, agentic commerce may move competitive advantage away from customer acquisition and toward operational excellence. For most brands, marketing spend has been the main lever for growth. If Goldman is right, the lever that decides whether an agent recommends you twice is whether your first order arrived complete, on time, and as described.

That puts familiar operational metrics in a new light. Inventory accuracy by location decides whether an agent sees stock it can promise. Delivery promise accuracy decides whether the agent’s quote to the shopper holds up. Order status updates decide whether the agent can answer “where is my order” without the shopper opening a ticket. Each of these was always a customer experience issue. In an agent-mediated channel, each one also becomes a ranking signal.

More channels, smaller baskets, more orders

Goldman expects agentic commerce to fragment ecommerce further. Agents that search across many merchants could surface smaller brands that never had the ad budget to compete for search traffic, which Goldman sees as a continuation of the direct-to-consumer trend. The same behavior could split baskets across merchants: an agent might buy the jacket from one brand and the matching shoes from another because each is the best option for that item. The report expects that to push average order values down.

The report also describes a split between open and closed ecosystems. Some players are opening their catalogs and checkout to third-party agents, while others restrict access to protect their customer relationships and retail media economics. Goldman points to Amazon’s recent decision to block Meta’s Muse agent as an example of the closed approach and to Shopify’s integration with AI shopping surfaces as an example of the open one.

Put those together and the operational picture is clear. Brands will likely see more orders per dollar of revenue, arriving from more surfaces: their own storefront, marketplaces, retail partners, and a growing list of agent platforms, each with its own rules. Every one of those surfaces is another sales channel that needs inventory, pricing, order intake, routing and status flowing correctly. A brand that treats each agent platform as a custom integration project will fall behind a brand that can add an agent surface the way it adds any other channel.

The commerce stack beneath the agent still does the work

One of the report’s more interesting arguments concerns Shopify. Goldman argues that investors who see Shopify mainly as a website builder underestimate it, because its value increasingly sits in the commerce infrastructure underneath: catalog management, inventory synchronization, checkout, payments, fulfillment, order management and post-purchase workflows. In Goldman’s reading, agents change the discovery layer while the operating system beneath it becomes more important as commerce fragments across AI interfaces.

We agree, and the same logic extends past any single platform. Most growing brands sell through more than one system. They run a Shopify storefront alongside Amazon and other marketplaces, wholesale and EDI accounts, retail stores, and one or more 3PLs or warehouses, with an ERP keeping the books. Agents add another layer of channels on top of that. The infrastructure that turns an agent’s intent into a delivered order has to reach across all of those systems, including every one the order touches after it leaves the storefront.

Goldman also maps the protocols forming around this layer. It describes the Model Context Protocol as the way merchants expose catalogs and inventory to agents, Google’s Universal Commerce Protocol as spanning discovery, checkout and post-purchase support, and payment protocols from the card networks as handling agent verification and purchase intent. Pipe17 has invested on the standards side through onX, the open order network standard it co-founded, and through the Pipe17 MCP Server, so that the order operations behind a brand can speak to agents directly.

The open questions all land after checkout

For all its optimism about the long-term opportunity, the report is candid about what is unresolved. Goldman lists the questions every party will need answered before autonomous purchasing scales: who owns fraud losses, who owns chargebacks, who owns returns, and who is liable when an agent buys the wrong thing. It notes that today’s ecommerce liability sits mostly with merchants, and that elevated return rates in agentic channels could erode the profitability of the channel for them.

Goldman also sees consumer intent verification as a major unsolved issue. The emerging frameworks from the card networks and their partners are designed to create an audit trail of what the shopper authorized. That record covers the front of the transaction. The back of it, what shipped, when it arrived, whether it came back and when the refund went out, lives in the order record. Brands that can connect an agent’s intent record to the full fulfillment and returns history of the order will be in a far stronger position when those liability questions get settled.

What this means for brands, and where Pipe17 fits

Goldman’s report is written for investors, but its implications for operators are practical. If advantage moves toward operational excellence, the priorities are the ones order operations teams already know, with less margin for error: accurate, per-location inventory that every channel reads from; delivery promises that hold; order status that flows back to wherever the order originated within minutes; returns tied to the original order and tagged by channel; and the ability to add a new sales channel, including an agent surface, without rebuilding integrations.

That is the job Pipe17 was built for. Pipe17 is the Order Operations Platform: it connects sales channels, fulfillment and back-office systems, and orchestrates every order from checkout to delivery, built to be run by the business team. For brands preparing for agent-placed orders, that means an agent platform becomes another channel in the same flow as the storefront, the marketplaces and wholesale, with the same inventory, routing, status and returns logic behind it.

Goldman’s own timeline suggests there is time to prepare, since adoption will build over years and start in the simplest categories. The brands that use that time to tighten their order operations will be the ones agents keep coming back to when the volume arrives.

Frequently Asked Questions