What Is Agentic Commerce? A Retailer's Guide to AI Shopping Agents

Agentic commerce is shopping where an AI agent, not the shopper, does the searching, comparing and buying. In 2026 it stopped being a demo and became infrastructure: Google, OpenAI, Amazon, Meta, Shopify, Stripe, Visa and Mastercard all shipped agent-ready shopping or payment rails within twelve months. McKinsey expects AI agents to orchestrate up to $1 trillion of US retail revenue by 2030.

For retailers, the open question is no longer whether shoppers will use AI shopping assistants. It is who owns the customer relationship once an agent sits between your store and the buyer. This guide explains what agentic commerce is, how AI shopping agents work, which agentic commerce examples matter in 2026, and what retail teams should do about it now.

In short:

• Agentic commerce means AI agents act for the shopper: they find, compare and, within set limits, buy.
• The rails are live: Google's Universal Commerce Protocol (UCP), OpenAI and Stripe's Agentic Commerce Protocol (ACP), and agent payment programs from Visa and Mastercard.
• Trust is the bottleneck. Most "agentic" purchases today still end with a human approving checkout.
• The retailers who win will run their own agents: ones that work for the retailer, know each customer, and act before a third-party agent goes shopping elsewhere.

What Is Agentic Commerce?

Agentic commerce is a model of online shopping in which AI agents act on a shopper's behalf. Instead of returning a list of links or product suggestions, an agent understands the shopper's goal, searches across retailers, compares options, and completes checkout within limits the shopper sets, such as budget, brand, size or delivery date.

"Agentic" means the software has agency: it takes actions, not just answers. A chatbot tells you which moisturizer suits dry skin. An agent remembers you are running low, checks three retailers for your usual product, applies your loyalty points and places the order once you approve, or without asking if you have told it to.

You will also see the model called agentic shopping, zero-click commerce or AI-orchestrated commerce. They describe the same shift: the shopping journey moves from pages a human browses to tasks an agent completes.

Agentic commerce has two sides, and retailers need to understand both:

• Buyer-side agents work for the shopper. ChatGPT, Google Gemini and AI Mode, Amazon's Alexa and Buy for Me, Perplexity and Meta's Muse all fall here.
• Seller-side agents work for the retailer. They run on the retailer's own data and decide what each customer should be offered, when, and through which channel. This is where agentic CRM comes in.

Agentic Commerce vs. Ecommerce vs. AI Shopping Assistants

The difference is who does the work. Classic ecommerce leaves every step to the shopper, an AI shopping assistant advises, and an agent executes.

Most AI shopping today sits in the middle column. The move to the right column is what the industry calls agentic commerce, and it changes where the sale happens: often off your site entirely.

How AI Shopping Agents Work

An AI shopping agent turns a goal into a completed order in five steps. Each one is a place where a retailer can win or lose the sale.

1. Intent and mandate. The shopper states a goal ("a fragrance-free SPF 50 under $40, delivered by Friday") and the rules the agent must respect: budget, preferred brands, payment method, whether it may buy without asking.
2. Discovery. The agent reads product data, not product pages. It pulls structured catalog feeds, attributes, stock and price from retailers that expose them through protocols such as UCP or ACP, or through merchant feeds.
3. Evaluation. It compares options on fit, reviews, price, delivery, return terms and loyalty value. Products with thin or missing attributes drop out here, because the agent cannot prove they match the request.
4. Checkout. The agent completes the purchase through the retailer's checkout, using a tokenized payment credential issued for agents. The shopper approves, or the mandate allows it to proceed alone.
5. After the purchase. Tracking, returns and, increasingly, reorders. This is where a one-off order becomes a routine, and where the question of who "owns" the customer gets decided.

The Protocols Behind Agentic Commerce

Agents and retailers need a shared language for catalogs, carts and payments. Four standards and two card-network programs carry most of the traffic in late 2026.

You do not need to implement all of them. Most retailers reach agents through their commerce platform (Shopify, commercetools, Salesforce Commerce Cloud) and payment provider, so the real work is data quality, access policy and what happens after the first order.

Agentic Commerce Examples in 2026

The clearest agentic commerce examples of the past year come from platforms racing to own the checkout, and from retailers deciding which agents they let in. Newest first:

Three patterns stand out. Platforms want to own the cart. Large retailers are pushing back on agents they do not control. And the most practical wins so far are retailer-run agents, such as Business Agent and ASA on AWS, that keep the conversation on the retailer's terms.

Agentic Commerce by the Numbers

Shoppers already use AI shopping tools at scale, and AI-referred visitors buy more. What lags is trust in letting an agent pay.

Read together, the numbers say two things. Demand is real: AI shopping traffic is growing fast and converts better than the average visit. But fully autonomous buying is still early. Liability for a wrong purchase is unresolved, and US Federal Reserve Governor Christopher Waller has named trust between buyers and sellers as the main barrier to scale. For now, most agentic commerce is agent-assisted: the agent does the research, and a human taps "buy".

That gap is the retailer's window. The habits shoppers form with agents over the next two years will decide whose products, and whose customer relationships, those agents default to.

Agentic AI in Retail: Buyer-Side Agents vs. Seller-Side Agents

Every agentic purchase involves two kinds of intelligence: one working for the shopper, and one that should be working for you. Most coverage of agentic commerce focuses on the first. Retailers need to invest in the second.

[Diagram: Two agents meet at your store. Only one works for you. The buyer-side agent (Gemini, ChatGPT, Alexa) takes the shopper's task and shops your store via UCP or ACP. On the retailer side, your seller-side agent (AI CRM Manager) reads your first-party data and reaches each customer first with a reorder, a cross-sell, or nothing.]

The buyer-side agent shops your catalog on the shopper's terms. The seller-side agent reasons on your own data and reaches each customer before they hand the next order to someone else's agent.

What Buyer-Side Agents Change

A buyer-side agent optimizes for its user: the right product, the best price, the fastest delivery. That is good for shoppers and uncomfortable for retailers, for four reasons.

• The session disappears. If an agent shops through a protocol, there is no browsing session to retarget, no abandoned cart email, no onsite upsell.
• Brands flatten into attributes. An agent compares specs, price and reviews. Merchandising and storytelling count for less unless they are encoded in your data.
• Loyalty gets tested every order. An agent told to "reorder my serum" can check five retailers in seconds. Habit, the quiet engine of repeat revenue, no longer protects you by default.
• The relationship can move. Whoever holds the shopper's preferences, sizes and routines owns the next order. Today that is increasingly the agent platform.

Why Retailers Need Their Own Agents

The answer is not to block every agent, and it is not to bolt a chatbot onto your site. It is to run a seller-side agent that knows each customer better than any third-party agent can, and acts first.

Your first-party data is the advantage. You know what each customer bought, when they are likely to run out, which shades and sizes they return, and what they ignored last time. A buyer-side agent knows only what the shopper tells it. A seller-side agent that reasons on your data can reach the customer before they ask an agent to go shopping: the replenishment reminder that lands the week the product runs low, the cross-sell that completes the routine, the decision to send nothing at all.

This is the move from martech to agentic CRM. Most "AI" in the retail stack today is a task agent: it drafts, scores and suggests, then waits for a person to build the segment and push the campaign live. A worker agent owns the outcome. It decides the next best action for each customer and executes it through the marketing automation tools you already run.

The gap matters. The share of retail marketers who say they can prove ROI on AI fell from 54% to 38% in a single year, according to MarTech.org. Tools that recommend but do not act leave the bottleneck where it was: team bandwidth.

Replenishment Is the First Purchase Agents Will Automate

The easiest purchase to hand to an agent is the one you have already made. Repeat purchases of consumables, such as skincare, fragrance, supplements, pet food and household goods, are routine, low-risk and predictable. That makes them the first category where shoppers will let agents buy without asking.

For beauty, wellness and other consumables retailers, this is the main risk and the main opportunity of agentic commerce. If you know the routine and act on it, the reorder stays with you. If you wait for a calendar campaign, the shopper's agent may reorder from someone else.

How Retailers Should Prepare for Agentic Commerce

Preparing for agentic commerce is mostly a data and decisioning job, not a new storefront. Eight steps, roughly in order:

1. Make your catalog machine-readable. Agents choose on attributes: ingredients, skin type, size, compatibility, substitutes, bundle logic. Fill the gaps in your product feed first. Our guide to how retailers win visibility in ChatGPT and AI search covers feed enrichment in detail.
2. Turn on the rails your platform already offers. Check agent checkout in your commerce platform, UCP eligibility in Google Merchant Center, and ACP support through your payment provider. For many retailers this is configuration, not development.
3. Write an agent access policy. Decide which agents may browse and buy, how they must identify themselves, and what you rate-limit. Amazon's dispute with Meta shows this decision is coming whether you plan for it or not.
4. Prepare payments and fraud for agent-initiated orders. Work with your payment provider on agent tokens, transaction scoring and who carries the loss when an agent orders the wrong item.
5. Measure AI-referred traffic as its own channel. Track sessions and orders from ChatGPT, Gemini, Perplexity and Copilot separately. Adobe's data shows they convert better than average, so they deserve their own targets.
6. Win the reorder before the agent searches. Predict when each customer will run out and reach them first with the right product, channel and timing. This is the single most defensible move in agentic commerce.
7. Move from segments to 1:1 decisions on your own channels. Calendar campaigns sent to broad segments will lose to agents that act on individual intent. Your owned channels need the same per-customer reasoning.
8. Put your team on strategy and guardrails. Let agents execute. Have people define business goals, brand rules and the skills agents use, then review outcomes.

Where Maestro Fits: The Seller-Side Agent for Retail

Maestro is Replenit's AI CRM Manager, a seller-side agent built for retail. It reasons, remembers, decides and executes individualized replenishment, cross-sell and retention workflows for every customer, on the stack you already run.

In agentic commerce terms, Maestro is the agent that works for the retailer:

• It knows the customer. Maestro builds a living Golden Memory of each customer, product and brand from your data, and reads each purchase as a signal of intent, not a flat event.
• It decides per person, with no segment step. Using 100+ pre-built retail skills, plus skills your team builds, it chooses the next best action, the timing, the channel, or no action at all.
• It executes and explains. Every decision runs through 15+ guardrail checks, then ships through your CEP or marketing automation platform as a Golden Decision Event with traceable reasoning and a confidence score.
• It learns from outcomes. A closed feedback loop feeds every result back into the next decision, so performance compounds.
• Nothing gets replaced. Maestro sits above tools like Klaviyo, Braze and Bloomreach and executes across 120+ platforms. No migration, no rip-and-replace.

The results show up where agentic commerce bites first: the post-purchase window and the reorder. L'Occitane saw a 235% uplift in post-purchase revenue with Replenit.

Agents will shop for your customers. Make sure one is working for you. See how Maestro, the AI CRM Manager for retail, works.

Agentic Commerce FAQ

What is agentic commerce in simple terms?

Agentic commerce is when an AI agent shops for you. You describe what you want and your limits, and the agent finds the product, compares retailers and completes the purchase, either after you approve it or on its own if you allow it.

What are some examples of agentic commerce?

Google's Universal Cart and agentic checkout in AI Mode and Gemini, Amazon's Buy for Me and Alexa shopping, Meta's Muse agent, Perplexity's shopping features, and ChatGPT's product discovery. On the retailer side, Google's Business Agent and Amazon's Agentic Shopping Assistant on AWS let retailers run their own agents.

What is the difference between an AI shopping assistant and an AI shopping agent?

An AI shopping assistant answers questions and suggests products, but the shopper still decides and checks out. An AI shopping agent acts: it can compare across retailers, build the cart and place the order within rules the shopper sets.

What are AI agents for ecommerce?

AI agents for ecommerce are software agents that perform commerce tasks autonomously. Buyer-side agents shop for consumers. Seller-side agents work for retailers, for example deciding which customer should get which offer and when, and executing it through the retailer's channels.

Is agentic commerce safe?

The payment rails use tokenized credentials, spending mandates and agent identification from Visa, Mastercard, Google and Stripe. The open issues are liability for wrong purchases and agents that do not identify themselves, which is why most purchases still require a human to approve checkout.

Will AI agents replace ecommerce websites?

Not soon. OpenAI moved checkout back to merchant sites in 2026, and shoppers still visit stores to browse and discover. But a growing share of research and repeat buying will happen through agents, so retailers need to be readable by agents and active on their own channels.

How does agentic AI in retail affect CRM and retention?

It raises the bar. When an agent can reorder from any retailer in seconds, segment-based calendar campaigns lose. Retention moves to 1:1 decisions made before the customer goes looking, which is the job of an agentic CRM or AI CRM Manager.

Sources

• McKinsey forecasts up to $5 trillion in agentic commerce sales by 2030, Digital Commerce 360, October 2025
• US agentic commerce revenue forecast to reach $1 trillion by 2030, Retail Dive, May 2026
• AI referrals drive higher ecommerce traffic and conversions, MarketingTech (Adobe Analytics data), June 2026
• New tech and tools for retailers to succeed in an agentic shopping era, Google, January 2026
• Google announces new Universal Cart at I/O, Search Engine Journal, May 2026
• Google I/O 2026 unveils Universal Cart, Stellagent, May 2026
• Amazon opens Agentic Shopping Assistant to outside retailers, EMARKETER, May 2026
• Amazon and Meta exchange agentic fire, Fortune, September 2026
• Agentic commerce had its biggest launch month, FinTech Weekly, September 2026
• ChatGPT Instant Checkout: what happened to it in 2026, Hypotenuse AI, 2026
• Stripe powers Instant Checkout in ChatGPT and releases the Agentic Commerce Protocol, Stripe, September 2025
• The truth about martech in 2026, MarTech.org, March 2026