Agentic commerce explained: what happens when AI starts shopping for the customer?

Online shopping used to assume a human would do the browsing. We searched, opened tabs, compared reviews, checked stock, calculated shipping and finally chose. Agentic commerce changes that sequence by allowing an AI system to take over some of those steps and return with a shorter path to a decision.

That does not mean every customer is about to hand over a credit card and say “buy whatever you think is best.” The more useful way to understand the shift is as a curve: AI can research a purchase, narrow options, assemble a basket, ask for approval, complete a transaction where supported, and eventually handle recurring or low-risk decisions with less human input.

What is agentic commerce?

Agentic commerce is shopping in which an AI agent can perform a sequence of commercial tasks for a user rather than only answer a question. It may interpret the user’s intent, compare products, apply constraints such as budget or delivery time, retrieve current product data, build a cart and interact with commerce systems to move the purchase forward.

McKinsey describes this as a shift from fragmented shopping toward an intent-driven flow. Its 2026 work argues that the level of automation will vary by category and estimates that, under moderate scenarios, AI agents could mediate $3 trillion to $5 trillion of global consumer commerce by 2030. That is a forecast, not a guarantee, but the infrastructure needed for the shift is already being built.

How is this different from a chatbot recommending products?

The important shift is not the chat window. It is the connection between conversation and systems. Google’s Universal Commerce Protocol (UCP), for example, is designed as a common language between consumer surfaces, agents, businesses and payment providers across discovery, checkout and order management. Stripe is building agentic commerce infrastructure around product publishing, checkout and agent payments. Shopify is syndicating merchant catalogs into AI channels. OpenAI supports merchant product feeds and shopping discovery in ChatGPT.

What does the automation curve look like?

1. Research assistant. The agent explains the category, reads specifications, compares reviews and helps the shopper understand trade-offs.
2. Shortlist builder. The agent narrows the market to a few products that fit the user’s constraints and explains why.
3. Basket builder. The agent combines multiple items, checks compatibility, quantity, availability and delivery timing.
4. Approval-based buyer. The agent prepares the transaction and asks the user to approve a purchase or a final set of choices.
5. Delegated buyer. For low-risk or recurring purchases, the user may give the agent pre-agreed rules around price, brands, timing or substitutions and allow it to act with less intervention.

The last stage will not suit every purchase. A customer may delegate printer ink and groceries long before they delegate a wedding ring, a medical device or a major luxury purchase. The level of human involvement will depend on risk, emotion, complexity, reversibility and trust.

Check out what our CEO is writing about: AI changed marketing, but not in the way people think

What changes for brands when the agent becomes part of the customer journey?

The biggest change is that some of the work traditionally done by a shopper will be done by software. That makes the information behind the experience as important as the visible page. If an agent has to compare five products against six constraints, it needs facts it can reliably retrieve and interpret.

1. Product data becomes part of the brand experience

Titles, variants, identifiers, dimensions, ingredients, compatibility, price, stock, delivery windows and return rules stop being back-office fields. They become inputs into recommendation and purchase decisions. If the data is incomplete or inconsistent, the agent may skip the product or present it badly.

2. Differentiation has to survive compression

A shopper may no longer read a full product page before seeing the recommendation. An agent could reduce dozens of pages to a few sentences. Brands therefore need differentiators that can be expressed as clear facts and evidence, not only visual mood or vague adjectives.

3. Availability and fulfillment become marketing signals

If two products are similar but one can arrive tomorrow, has simpler returns and is actually in stock, those operational details can influence the recommendation. The separation between marketing, merchandising, ecommerce and operations becomes thinner.

4. The website is no longer the only storefront

Google is enabling UCP-powered shopping across AI Mode and Gemini, Shopify is building Agentic Storefronts across AI channels, and ChatGPT can surface merchant products from structured catalog data. The merchant site still matters, but discovery may begin somewhere else and the transaction path may be partly orchestrated by an agent.

What is actually live in 2026?

The ecosystem is real, but uneven. Different platforms support different levels of discovery, checkout and geographic coverage, and the architecture is still changing.

• Google. UCP was introduced in 2026 as an open standard for agentic commerce. Google says select merchants can enable UCP-powered checkout on eligible product listings in AI Mode and Gemini in supported markets, while remaining the seller of record.
• Shopify. Shopify says merchants can syndicate product information into connected AI channels through Shopify Catalog and manage agentic storefronts. It also offers an Agentic plan for merchants that do not want to migrate their full ecommerce stack.
• OpenAI. ChatGPT shopping uses merchant and product metadata from third parties or directly from merchants, and OpenAI supports direct product feeds for fresher catalog information. Checkout options and partner availability continue to evolve, so merchants should verify the current implementation before building around one flow.
• Stripe. Stripe’s Agentic Commerce Suite is designed to let businesses publish products to agents, support checkout and accept agent-initiated payments while the business remains the merchant of record.

The practical conclusion is simple: prepare the data and systems first. Do not redesign the entire commerce stack around one protocol or one platform before the use case, geography and eligibility are clear.

How should a brand prepare for agentic commerce?

1. Clean the product catalog. Make titles, descriptions, identifiers, variants, dimensions, ingredients/specifications, imagery and category data complete and consistent.
2. Keep price and availability current. An agent needs fulfillment-grade truth, not a marketing page that says “available” while the commerce backend says otherwise.
3. Make policies explicit. Shipping, returns, warranty, subscriptions, delivery limits and eligibility rules should be easy to retrieve and understand.
4. Strengthen product proof. Use accurate reviews, testing, comparisons, certifications, expert guidance and original content that explains why the product is right for a specific use case.
5. Use structured commerce channels where relevant. Merchant feeds, shopping catalogs and protocols can provide fresher information than waiting for a crawler to rediscover every change.
6. Protect the merchant relationship. Understand who owns checkout, customer data, consent, returns, service and post-purchase communication in every agentic channel you join.
7. Test the full journey. Ask an AI agent to find the product, compare it, explain it, locate the right variant and reach checkout. The broken step is often not the content page; it is inventory, variant mapping, delivery or a policy the system cannot interpret.
8. Measure the business result. Track qualified AI referral traffic, product discovery, conversion, order quality, margin, returns and new-customer behavior. Do not celebrate “AI visibility” if the orders are unprofitable or the experience creates support problems.

What should product pages change?

Do not turn every product page into a machine manual. Keep the human experience strong, but make the decision facts explicit enough that both a person and an agent can find them.

This is not a replacement for strong marketing strategy and customer experience. It is another reason those disciplines need to connect to ecommerce and operations rather than working as isolated teams.

What new competitive risks appear?

• Brand disintermediation. The agent may become the interface and compress the brand into a few attributes, reducing exposure to the full website experience.
• Price pressure. Agents can compare options faster and more consistently, which may make weak differentiation more visible.
• Bad data at scale. An incorrect attribute can be repeated across multiple AI channels if the same catalog source feeds them all.
• Channel dependency. Relying too heavily on one AI platform can recreate the same distribution risk brands already know from marketplaces and social platforms.
• Trust and authorization. Agents need clear user permission, transaction controls and secure payment/account flows. Brands also need clarity on fraud, disputes and responsibility when an agent acts.
• Over-automation. A technically possible automated purchase can still be a bad customer experience if the decision is emotional, high-risk or difficult to reverse.

A simple agentic commerce readiness check

What should marketers watch next?

Three things matter more than the latest demo: which categories consumers are willing to delegate, which standards become interoperable across major platforms, and whether merchants can retain enough customer relationship and economics to make the channel attractive.

The winning brand may not be the one with the most impressive AI shopping assistant. It may be the one whose product truth is easiest to verify, whose offer survives comparison, whose inventory and fulfillment data are reliable, and whose customer experience still feels good when the interface is no longer completely under the brand’s control.