What Is Agentic Commerce? A Plain-English Guide (2026)

Most explanations of this term are written by companies selling the software, which is why they read like a brochure. Here is the version we give clients: agentic commerce moves the browsing, the comparing and often the buying from a human on your website to a piece of software acting for that human. Everything downstream of that one change is what makes it interesting, and occasionally uncomfortable, for anyone running a store.

This guide stays on commerce. If you want the underlying concept first, read what agentic AI is and, if you keep tripping over the vocabulary, agentic AI versus AI agents.

What agentic commerce actually is, in plain terms

Break the phrase in half. Commerce is buying and selling. Agentic means software that can plan and take action toward a goal rather than just answer a question. Put them together and you get software that can be told "find me a waterproof jacket that packs down into a carry-on, in navy, that ships before Friday" and then go and do it, end to end.

Three things separate it from what came before. First, autonomy: the agent decides which options are worth showing you rather than dumping a results page in front of you. Second, action: it can actually transact, not just recommend. Third, delegated authority: you gave it permission to spend on your behalf, within limits, in advance.

A few things it is not. It is not a chatbot on your storefront answering size questions, that is conversational commerce. It is not a recommendation engine ranking products for a human to click. And it is not a shopping assistant that finds a product and then hands you a link, although that is the transitional stage most assistants are in right now. The moment the software completes the purchase is the moment it becomes agentic.

How an agentic purchase actually happens

There are four stages, and it is worth understanding each because your store touches all of them differently.

1. Intent.The shopper describes a need in ordinary language, usually with constraints attached: a delivery deadline, a size, a compatibility requirement. Under Google's AP2 protocol this can be captured formally as an Intent Mandate, a cryptographically signed record of exactly what the shopper authorised.

2. Discovery and comparison. The agent gathers candidates. This is where your store either exists or does not. The agent is reading structured product data, feeds, specifications, reviews and whatever trusted sources mention you. It is not reading your homepage animation or your lifestyle photography.

3. Checkout.The agent requests a checkout from the merchant. Stripe and OpenAI's Agentic Commerce Protocol defines this handshake: the agent passes a secure token rather than raw card details, and the merchant confirms the order. Stripe states the principle bluntly, that customers should be able to securely buy where they discover while businesses sell through new channels without giving up trust, brand or control.

4. Fulfilment and aftercare.The merchant fulfils, exactly as it would for any other order. If something goes wrong, the agent may raise the return on the shopper's behalf, quoting your published policy back at you. Which is a good reason for that policy to exist as clear text rather than a scanned PDF.

The important detail, and the one that calms most merchants down: the business remains the merchant of record. You keep control over which products can be sold, how they are presented, how transactions are processed and how orders are fulfilled. The agent introduces the buyer. It does not take over your business.

Traditional ecommerce vs agentic commerce

The transaction looks identical from your order dashboard. Almost everything upstream of it is different.

That last row is the one to sit with. In traditional ecommerce a weak product page still gets seen and can still convert someone who arrived motivated. In agentic commerce an incomplete product record does not lose the sale, it removes you from the comparison entirely, silently, with no bounce rate to warn you.

What is an example of agentic commerce?

The cleanest live example is Instant Checkout in ChatGPT. Stripe and OpenAI announced it on 29 September 2025. A shopper in the US asks ChatGPT for something, a supported product appears in the conversation with a Buy button, and the purchase completes inside the chat. It launched with US Etsy sellers and extended to Shopify merchants including Glossier, Vuori, Spanx and SKIMS. Payment runs through a Shared Payment Token so the assistant never touches the underlying card.

A second example, less glamorous and probably more valuable: reordering. A facilities manager tells an assistant to restock the same filters as last quarter, matching the part number, from whichever supplier can deliver by the 15th. That is agentic commerce doing something a human genuinely does not want to do, and it is the shape most B2B adoption will take.

A third: constrained gift buying. "Something for a six year old who likes dinosaurs, arriving before Saturday, avoid anything with small parts." Multiple hard constraints, a deadline and a safety filter. That is a miserable job for a person with fifteen browser tabs and a natural job for an agent.

What companies offer agentic commerce?

It helps to sort them by which part of the chain they own. Some provide the surface where shoppers buy, some provide the payment rails, and some provide the merchant side.

1. OpenAI. Instant Checkout in ChatGPT is currently the most developed consumer buying surface. It is where the phrase stopped being theoretical.
2. Stripe. Co-author of the Agentic Commerce Protocol with OpenAI, published under the Apache 2.0 licence, plus the Shared Payment Token that carries authority to charge without exposing card details. Merchants already on Stripe can enable agentic payments with a very small code change.
3. Google. The Agent Payments Protocol, AP2, announced 17 September 2025 with more than 60 partner organisations. It solves a different problem to ACP: proving that a shopper genuinely authorised this agent to make this purchase, using signed Intent and Cart Mandates.
4. Mastercard. Agent Pay, its agentic payments programme, introduces Agentic Tokens that bind a tokenized card credential to a specific agent and a specific consent policy.
5. Visa. Intelligent Commerce takes a comparable route, issuing credentials an agent can use inside limits the cardholder sets.
6. Shopify. The largest single pool of agent-reachable merchants, which is why it was an early route into Instant Checkout at scale.
7. Salesforce. Agentforce Commerce brings agent capability into enterprise commerce and service workflows, and Salesforce has publicly backed the Agentic Commerce Protocol.
8. Amazon. Running its own agent-assisted shopping inside its own storefront, which is a deliberately different strategy: keep the agent, and the customer, inside the walls.
9. Microsoft. Copilot shopping experiences and merchant programmes, positioned around agents that discover and act on behalf of both consumers and businesses.
10. Perplexity. Shopping answers with buy actions attached, which fits its answer-engine model more naturally than it fits a traditional search results page.

Notice how few of these are ecommerce platforms and how many are payment companies. That tells you where the hard problem was: not showing products to an agent, but letting an agent pay for them without anyone getting defrauded.

What are the 4 types of ecommerce, and where does agentic fit?

People ask this alongside agentic commerce, so it is worth settling. The four standard categories are B2C (business to consumer), B2B (business to business), C2C (consumer to consumer, usually via a marketplace) and C2B (an individual selling to a business, such as a freelancer or a stock photographer).

Agentic commerce is not a fifth category. It is a transaction method that can sit on top of any of the four. A B2C shopper can use an agent. A B2B buyer can. A marketplace seller can be discovered by one. Anyone presenting it as a new type of ecommerce is confusing the channel with the counterparty.

What are the risks of agentic commerce?

Worth being straight about these, because the vendor material tends not to be.

• Wrong purchases.An agent misreads a constraint and buys the wrong size, the wrong variant or at the wrong moment. Returns volume is the merchant's problem regardless of whose mistake it was.
• Accountability gaps. If a fraudulent or incorrect transaction happens, who carries it: the shopper, the agent provider, the payment network or the merchant? Google names this openly as one of the three problems AP2 exists to address, alongside authorisation and authenticity. It is being worked on, not solved.
• Losing the customer relationship. If the agent owns the conversation, you may never get the email address, the browsing signal or the chance to upsell. You get an order and a shipping label.
• Commoditisation. Agents compare on facts. If your differentiation is entirely emotional and none of it shows up in specifications, reviews or delivery performance, you will lose comparisons you used to win on presentation.
• Data quality turning into real losses. A stale feed that shows stock you do not have creates orders you must cancel. In traditional ecommerce that is an annoyance. In agentic commerce it damages the reliability signal agents use to rank you next time.
• Impersonation and fraud. An agent claiming an authority it does not have is a new attack surface. Tokenized credentials and signed mandates are the industry answer, which is exactly why they were built before mass adoption rather than after.
• Concentration risk. If a small number of assistants become the main discovery surface, being delisted or deprioritised by one of them is a serious commercial event with no appeals process you control.

Is agentic commerce the future?

Partly, and the honest position is somewhere between the hype and the shrug. The evidence for taking it seriously is that the infrastructure got built unusually fast and unusually cooperatively: two competing open protocols in the same month of 2025, the card networks shipping agent credentials, and the largest merchant platforms plugging in. Infrastructure that expensive does not get built for a fad.

The evidence for calm is that shopper behaviour changes slowly. Most people still open a browser. Categories where touch, fit, taste or impulse dominate will resist delegation for a long time. And agent-driven buying currently represents a small share of total ecommerce.

The useful framing: agentic commerce is an additional channel that rewards preparation and punishes neglect quietly. You do not need to bet the business on it. You do need your catalogue to be legible, because the work that makes you agent-ready is the same work that improves marketplace listings, shopping feeds and classic search. There is no wasted effort here.

How to make your store agent-ready

In rough order of return. None of it is exciting, all of it compounds.

1. Complete your product identity fields. GTIN or MPN, brand, exact model, variant. Without these an agent cannot confidently match your product to the thing the shopper asked for.
2. Fill in the attributes a comparison needs. Dimensions, weight, materials, compatibility, capacity, care, warranty length. If a human would need to ask, an agent will simply skip you.
3. Publish Product and Offer structured data. Machine-readable markup on every product page, matching what your feed says. Contradictions between the two are worse than a gap.
4. Keep the feed genuinely live. Availability and delivery timeframes need to be current, not nightly-ish. Accuracy is a ranking signal now, not just a customer service nicety.
5. Write shipping and returns policies as plain text. Not an image, not a PDF, not hidden behind a tab that loads by script. The agent will quote it, so make sure it says what you mean.
6. Restructure product copy facts-first. Specifics at the top, persuasion underneath. Keep the brand voice, just stop making a machine hunt for the material composition.
7. Make reviews and returns performance visible. These are heavily weighted signals, and they are among the few places emotional brand equity still reaches an agent.
8. Enable agentic payments with your provider. If you are on Stripe this is a small change. If you are elsewhere, ask your provider directly what their ACP or AP2 path looks like and when.
9. Instrument the measurement. Tag assistant referrals, watch branded search, and ask your payment provider to flag orders arriving through agentic surfaces. Imperfect data beats none.
10. Fix your top sellers first. Do not attempt the whole catalogue. Take the products that generate most of your revenue, get them genuinely complete, then work down.

If your platform makes any of this painful, that is a signal about the platform rather than about agentic commerce. Our ecommerce development work increasingly starts with a data audit for exactly this reason, and Commerceflo is the AI operator we built to keep catalogue, feed and content accuracy from drifting once it is fixed. UK merchants get the same audit and build process through our UK ecommerce development service.

The one-paragraph version

Agentic commerce is shopping where the software buys. The rails exist, the protocols are open, the card networks are on board, and the first real transactions have already happened. For a store owner it is not a strategy question yet, it is a hygiene question: can a machine read your catalogue accurately enough to recommend you. Most stores currently cannot. Fixing that is worth doing whether or not agents ever become your biggest channel.

Sources

• Stripe, "Stripe powers Instant Checkout in ChatGPT and releases Agentic Commerce Protocol codeveloped with OpenAI" (29 September 2025): Instant Checkout launch, Etsy and Shopify merchants, Shared Payment Token.
• Stripe, "Developing an open standard for agentic commerce" (29 September 2025): Apache 2.0 licence, delegated payments, and the merchant remaining the merchant of record.
• Google Cloud, "Announcing the Agent Payments Protocol (AP2)" (17 September 2025): authorisation, authenticity and accountability; Intent and Cart Mandates; 60+ partner organisations.
• Agentic Commerce Protocol specification site: the open standard itself, its scope and how merchants implement it.

Bhavesh Barot is the founder of FactoryJet. He runs every discovery call himself. If you want a straight answer on whether your catalogue is readable by AI agents, start with our agentic commerce guide or book a 30-minute call.