What Is Agentic Commerce? A Practical Guide to AI Commerce and AI Shopping Agents

Updated 22 October 2026. This guide was first published in February. Since then Amazon retired Rufus, OpenAI pulled checkout back out of ChatGPT, Google turned its Universal Commerce Protocol into a live cart, Meta launched Muse, and Walmart opened its store to outside agents. The definitions below still hold; the examples are now current.

What is agentic commerce?

Agentic commerce is shopping in which an AI agent acts for the shopper. The shopper states a goal in plain language ("a carry-on suitcase under $200 that fits budget airline limits"), and the agent searches, filters products against the constraints, explains a shortlist and, where the store and the shopper allow it, places the order. The shopper sets the goal and approves; the agent does the legwork.

AI commerce is the wider term. It covers every part of selling that AI now shapes: AI answers in search, recommendation engines, assistants inside marketplaces, and the agents that act. Agentic shopping is the same thing seen from the shopper's side. All three describe one shift: the first selection of products is increasingly made by software, not by a person scrolling a category page.

A chatbot answers questions. An agent takes steps. That difference is why the word "agentic" matters to a seller: an agent does not just mention your product, it can decide to buy it or skip it.

This is no longer a forecast. As of October 2026, Amazon's shopping assistant can buy an item automatically when it hits a target price, Google checks out from eligible merchants inside AI Mode and the Gemini app, and Walmart has published rules for which outside agents may shop on its site. You can follow those moves week by week on our AI commerce news page.

How agentic shopping works, step by step

Every AI shopping agent, whatever the brand, runs roughly the same loop.

1. Request. The shopper describes a need, often with constraints: budget, size, compatibility, delivery date, values.
2. Interpretation. The agent turns the request into checkable conditions: price under $200, dimensions within a limit, in stock, ships by Friday.
3. Retrieval. The agent pulls candidates from the product data it can reach. That data is large and structured: Google says its Shopping Graph holds over 60 billion listings (Google, October 2026), and Shopify's Catalog covers over 1 billion products (Shopify, October 2026).
4. Filtering. Each condition removes products. A product with no stated dimensions cannot pass a dimensions check, so it drops out even if it would have fit.
5. Shortlist and explanation. The agent shows a few options and says why each one fits.
6. Approval and payment. The shopper confirms, or a standing permission covers the purchase, for example a spending limit the shopper set in advance.
7. After the order. Tracking, returns and reorders increasingly run through the same agent.

Step 4 is where most products are lost, and it is the step a seller controls most. If an attribute is missing, the agent rarely guesses; it moves on. Our guide to product attributes and structured data covers which fields carry the most weight.

Agentic commerce vs traditional e-commerce

The two models differ less in technology than in who does the comparing.

The last row is the one to remember. On a category page, a listing with a weak title still shows up and might still sell. In an agent's shortlist there is no page two.

Who runs AI shopping agents (as of October 2026)

The landscape changed more in 2026 than in the three years before. Here is where each major platform stands. For a deeper side-by-side, see AI shopping assistants compared.

Amazon: Alexa for Shopping replaced Rufus

On 13 May 2026 Amazon retired the Rufus brand and folded its shopping chatbot into Alexa for Shopping, which works in the Amazon app, on Amazon.com and on Echo Show devices in the US (CNBC, May 2026; Amazon's own Rufus page now notes the rename). Ahead of Prime Big Deal Days in October, Amazon added deal alerts, target-price alerts and Auto-Buy, which completes a purchase with the shopper's default payment method once the price is reached (Amazon, October 2026). That is agentic commerce inside the largest marketplace: the agent buys without the shopper returning to the page.

Amazon is also closing its store to outside agents. It blocked Meta's Muse from Amazon.com on 20 September 2026, saying the agent did not identify itself (GeekWire, September 2026).

OpenAI: ChatGPT for discovery, checkout back on merchant sites

ChatGPT launched Instant Checkout in September 2025 on the Agentic Commerce Protocol it built with Stripe (Stripe), and added shopping research, a guided buyer's guide, in November 2025 (OpenAI). In March 2026 OpenAI scaled Instant Checkout back: merchants can now use their own checkout, and ChatGPT concentrates on product discovery, with retailers such as Target, Sephora, Best Buy and Wayfair connected for discovery (Digital Commerce 360, March 2026). We cover the reasons in what happened to ChatGPT Instant Checkout.

Google: AI Mode, Gemini and Universal Cart

Google published the Universal Commerce Protocol (UCP) on 11 January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart, to power checkout on eligible product listings in AI Mode and the Gemini app; the retailer stays the seller of record (Google, January 2026). On 15 October 2026 Google launched Universal Cart in the US across Search and Gemini: one cart for items from many stores, with price tracking and checkout powered by UCP, either in the cart with Google Pay or handed over to the merchant's own cart (Google, October 2026). Our UCP explainer for merchants walks through eligibility.

Meta: Muse

Meta launched Muse, a personal agent that browses, fills in forms and completes tasks, in the US on 8 September 2026. It checks out with Link by Stripe and asks the person before a purchase (Meta, September 2026). Muse matters to sellers because it works in a browser on ordinary store websites, so the product page itself, not only a feed, is part of what the agent reads.

Perplexity, Microsoft Copilot and the rest

Perplexity and PayPal launched Instant Buy for US users in November 2025 (PayPal, November 2025). In August 2026 a US appeals court lifted the injunction Amazon had won against Perplexity's Comet shopping agent; the underlying case continues (Engadget, August 2026). Microsoft launched Copilot Checkout in January 2026 with PayPal, Shopify, Stripe and Etsy, with the retailer remaining merchant of record (Retail TouchPoints, January 2026).

Retailers opening a front door for agents

On 21 October 2026 Walmart announced an Agent Gateway for approved agents, built on MCP and starting as a beta, plus Agent Profiles that let customers approve every agent purchase or set a spending limit such as $50 or $100. Agents may not see card details and are blocked if they break Walmart's terms (Walmart, October 2026).

The protocols behind it: ACP, UCP and AP2

Agents need a shared way to read a catalog, build a cart and pay. Three open protocols compete and overlap:

• ACP, the Agentic Commerce Protocol, co-developed by OpenAI and Stripe (September 2025). It defines how an agent passes an order to a merchant, with a payment token scoped to one merchant and one basket total (Stripe). Since March 2026 OpenAI uses it mainly for product discovery.
• UCP, the Universal Commerce Protocol, from Google (January 2026). It covers discovery, purchase and post-purchase support, and is built to work alongside A2A, AP2 and MCP (Google).
• AP2, the Agent Payments Protocol, from Google (September 2025). It uses signed "mandates" that prove what the shopper authorized and limit what the agent may spend (Google Cloud).

Most merchants will not implement any of these directly. Platforms do it for them: Shopify, payment providers such as Stripe and Adyen, and feed tools translate one catalog into each protocol. Our ACP vs UCP vs AP2 comparison explains which one reaches which agent.

Agentic commerce benefits for businesses

The upside for a merchant is real, but it is narrower than the hype suggests.

• Shoppers arrive decided. An agent that recommends your product has already matched it to a stated need, price and constraint.
• New reach without new storefronts. One clean catalog, syndicated through a platform, can appear in Google AI Mode, ChatGPT, Copilot and Muse.
• You stay the seller. In Google's UCP checkout, Copilot Checkout and Walmart's gateway, the retailer remains seller or merchant of record and handles fulfillment and returns.
• Less friction for repeat buys. Price-triggered and reorder purchases (Alexa's Auto-Buy, Walmart's "usual items") favor products with stable data and reliable stock.

The costs are just as concrete. The agent holds the conversation, so you see less of the customer's journey. Products with gaps in their data are silently excluded. And each platform sets its own rules: Amazon blocks outside agents, Walmart admits approved ones, and Google limits checkout to eligible merchants.

In agentic commerce there is no page two. A product either passes every check the agent runs, or it is not in the answer.

Kris Krokos

What to do now: a seller's checklist

You do not need to build an agent. You need a catalog an agent can read and trust.

• Fill every attribute that a shopper's constraint could test: size, dimensions, material, compatibility, certifications, what is in the box.
• Keep price and stock in sync across your site, feeds and marketplaces. An agent that finds a mismatch drops the product or the store.
• Make the product page readable without a feed. Muse and other browser agents read the page itself, so specifications belong in text and structured markup, not only in images.
• Check your platform's agent settings. Shopify, Google Merchant Center and payment providers now carry options for agent channels and direct checkout; know which are switched on for your store.
• Write descriptions that answer real buying questions ("does it fit", "does it work with", "how long does it last") in plain words.
• Keep return and warranty policies explicit. Agents explain why they picked a product, and clear policies are part of that reason.
• Watch who is allowed in. Decide whether you want outside agents buying from your store and under what limits, as Walmart did.

For a full walk-through, see the AI commerce topic page.

The future of agentic commerce

Three things look settled as of October 2026, and one is still open.

Settled: permission becomes the gate. Walmart's gateway, Amazon's blocks and the draft Personal Agent Protocol published by Sierra with Meta in October (Sierra, October 2026) all point the same way: agents will have to identify themselves and act within limits the shopper and the store set.

Settled: checkout moves to whoever can carry the risk. OpenAI stepped back from running checkout; Google, Walmart and the payment networks are stepping in with mandates, spending limits and purchase protection. Expect checkout inside agents to grow where a large platform or payment network stands behind it.

Settled: product data decides visibility. Every agent above filters on structured facts. That will not change with the next model release.

Open: who owns the customer. Platforms say the merchant stays seller of record, but the agent holds the request, the comparison and often the reorder. How much of that relationship merchants keep is the question to watch through the 2026 holiday season.

FAQ

What is agentic commerce in simple terms?

It is shopping where an AI agent does the work for you: it searches, compares products against your needs and, if you allow it, places the order. You set the goal and approve the purchase.

What is the difference between agentic commerce and e-commerce?

In traditional e-commerce the shopper searches, filters and compares on a store's site. In agentic commerce an AI agent does that comparison and returns a short list, so a product's visibility depends on whether its data passes the agent's checks.

Is Amazon Rufus still available?

No. Amazon retired the Rufus name on 13 May 2026 and replaced it with Alexa for Shopping, which works in the Amazon app, on Amazon.com and on Echo Show devices in the US.

Can ChatGPT buy products for you?

Mostly not inside ChatGPT itself, as of October 2026. OpenAI scaled back Instant Checkout in March 2026; ChatGPT now helps shoppers find and compare products, and most purchases finish on the merchant's site or in a retailer's own ChatGPT app.

Do merchants need to adopt ACP or UCP?

Usually not directly. Most merchants reach these protocols through their commerce platform, payment provider or feed tool. What every merchant does need is complete, accurate product data.