Agentic Commerce: What Retailers and Manufacturers Should Prepare Now

Agentic commerce describes commerce in which AI agents take over part of buying or selling. An agent is more than a chatbot. It pursues a goal, plans the steps to get there and uses tools such as product search, the shopping cart or the order system.

In practice, the term covers three different things.

1. Agents on the customer side. An AI assistant such as ChatGPT, Gemini or Perplexity advises the customer, compares offers and ideally fills the cart or completes the purchase itself.
2. Agents in the retailer's own sales experience. An agent in the shop or customer portal advises, answers questions about availability and delivery and, in B2B, reorders regularly needed items.
3. Agents in the back office. Agents steer promotions, merchandising, search and loyalty across systems, capture orders from emails and PDFs or prepare quotes.

The public debate revolves almost entirely around the first form. Economically, the second and third matter more today. Why terms like agentic and autonomous commerce often get mixed up is the topic of the episode Agentic vs. Autonomous Commerce (in German).

At the end of September 2025, OpenAI introduced the Agentic Commerce Protocol and Instant Checkout. The announcement promised to connect more than one million Shopify merchants, and only a fraction went live. In March 2026, OpenAI scaled back buying directly in the chat and has since moved it into merchants' ChatGPT apps.

Google took a different route. The Universal Commerce Protocol in January 2026 was followed in October 2026 by Universal Cart in Google Search and the Gemini app in the US. Customers collect products from different retailers there, pay with Google Pay or hand the cart over to the retailer's shop. Either way, the retailer remains the seller and handles confirmation and service. Since September 2026, Shopify has let browser agents check out at participating merchants, and at the end of October it built agentic payments with Visa, Mastercard and American Express into Shop Pay, without the agent ever seeing the card details. Walmart opens its assortment to approved agents through an Agent Gateway, with its own rules and spending limits set by the customer. Amazon, by contrast, blocks outside agents.

Demand remains modest for now. Agentic traffic accounted for less than one percent of retail visits in the second quarter of 2026. In a survey by the German e-commerce association bevh with 40,000 participants, 92 percent rejected fully automated purchasing. Universal Cart is so far only available in the US. For retailers and manufacturers in the DACH region, that is a head start they can use.

We review the year in detail in the episode One Year of Agentic Commerce (in German).

The protocol is rarely the problem. Whether an agent can buy reliably depends on the craft of retail. "150 in stock" in the merchandise management system rarely means 150 units are on the shelf. Maybe it is 120, twenty are in transit and ten are still in a container. On top of that come price commitments, discount codes, delivery methods, take-back of old devices and packaging regulations. An external agent has to understand and follow every one of these rules, otherwise wrong purchases occur that the retailer ends up absorbing.

Much of what drives revenue and margin is also at stake for the retailer, including upselling and cross-selling, its own advice and the relationship with the customer after the purchase. That is exactly why models are gaining ground in which the retailer remains the seller, controls checkout and decides which agents may do what. More on this in the episode Why the AI checkout fails on the craft of retail (in German).

Product discovery through AI assistants is growing faster than buying through agents. At US retailers, traffic from AI sources rose by 393 percent year on year in the first quarter of 2026, and the conversion rate of these visitors was 54 percent higher than that of other visitors. In Germany, 30 percent already use AI chatbots for product research.

For retailers and manufacturers, AI assistants are becoming a new entry point into the customer journey. Brands that appear there with complete product data, prices and availability win visitors with high purchase intent. Brands that are missing do not get recommended at all. How to make your assortment findable for people and their AI assistants alike is described under Agents for your customers.

Today, agents mainly reduce costs, especially where a lot of manual work goes into recurring processes. In B2B, agents capture orders from emails and PDFs, prepare quotes and suggest reorders before the customer asks. In retail, they bring sell-through, prices and stock into one view and steer promotions, search and merchandising across systems that run side by side today.

Responsibility stays with your team. The agent prepares, your people approve. Success is measured against a value lever such as cost per transaction, revenue per customer or capital tied up in inventory. More under Agents for your processes.

Gartner now also evaluates commerce platforms on whether they support such agents and connections through open standards like MCP. Read our analysis in the article Gartner Magic Quadrant 2026: What It Means for Your Platform Decision, or listen to the episode Gartner Magic Quadrant 2026 (in German).

Nobody can reliably say today how quickly your customers will shop through assistants. The following steps pay off in every scenario.

1. Make product data machine-readable. Complete attributes, clear variants and understandable descriptions are the basis for an assistant to find your assortment and recommend it correctly.
2. Offer availability, prices and delivery as a reliable interface. What your shop shows correctly, an agent must be able to query correctly, for example through an MCP server. Decide which agents may do what. Walmart's Agent Gateway shows what this can look like.
3. Keep your architecture independent of any single protocol. ACP, UCP and the next standards will come and go. A composable architecture connects new channels without rebuilding the core.
4. Measure AI traffic as a channel of its own. Separate visits from ChatGPT, Gemini, Perplexity and others in your web analytics and track how often AI assistants mention your brand.
5. Start with an internal agent tied to a measurable value lever. An agent for order capture or customer service shows within a few months what agents can do in your company and lays the groundwork for everything that follows.

Companies that provide product data, payment and delivery methods in a clean, machine-readable way can test a new channel within weeks without turning it into a major project. That capability is the real advantage, regardless of which protocol prevails.

A little over a year in, agentic commerce is smaller than the first announcements promised, and more concrete at the same time. Fully automated buying remains the exception for now. But Google, Shopify and Walmart are building the paths for it, in a way that keeps the retailer as the seller. Product discovery through AI assistants is growing and driving revenue, and agents inside the company are already reducing costs today.

For retailers and manufacturers, it therefore pays to lay the foundations now and gain experience with a first agent. With the Impact Check, we show you within two weeks where your company stands and what your first agent looks like.

Sources: OpenAI, Instant Checkout and Agentic Commerce Protocol (29.09.2025); Digital Commerce 360, OpenAI moves checkout into merchant apps (06.03.2026); Google, Universal Commerce Protocol (11.01.2026); Google, Universal Cart (15.10.2026); TechCrunch, Walmart Agent Gateway (21.10.2026); PYMNTS, Shopify and agentic payments (29.10.2026); Flagship Advisory Partners, Agentic Commerce Market Tracker (August 2026); Adobe Digital Insights, AI traffic Q1/2026; bevh, Interactive retail in Germany; CELUM study on AI product search (2026). The analysis reflects the assessment of foobar Agency.