What Is Agentic Commerce? A Store Owner's Definition
Five of the pages ranking for what is agentic commerce come from IBM, Mastercard, Stripe, Salesforce and McKinsey. Every one of them defines the term properly. None of them ends with something you can do on Tuesday.
At rank 8 sits a Reddit thread titled "Has anyone had success with agentic commerce yet?" That thread is the honest state of this market, and it is more useful than most of the explainers above it.
So here is the version for someone who runs a store. Your checkout, your product feed and the markup around them already answer this question, whether or not you have looked.
What is agentic commerce, in one sentence
Agentic commerce is buying where an AI agent does the finding, the comparing and sometimes the paying, acting on an instruction a person gave it earlier.
That is the agentic commerce definition every vendor page lands on, in slightly different words. The meaning worth arguing about is narrower: an agent is a customer that cannot be persuaded. It reads what your site exposes in machine-readable form and acts on that. Your photography, your copy and your trust badges are not inputs to its decision.
Agentic e-commerce is the same idea with the word retail bolted on. The distinction does not matter. The mechanism does.
“I truly believe agentic AI is going to be one of the biggest transformations since the beginning of the cloud.”
Agentic AI vs agentic commerce: where the line sits
These two get used as if they were the same thing, and they are not. Agentic AI is the capability. Agentic commerce is the transaction that capability completes.
"Agentic AI in enterprise planning is moving organizations beyond today’s reactive copilots and toward goal‑driven AI agents that can sense, decide, and act—within enterprise guardrails—to drive business outcomes across finance, supply chain, and operations."
Agentic AI describes software that plans and acts toward a goal on its own, in any domain at all. Booking a flight, filing a ticket, refactoring code. Agentic commerce is the narrow case where the goal is buying something and the agent runs to autonomous completion, ending in a payment against your store with no person approving it.
The difference is not academic, it changes what breaks. An agentic AI system reading your site is a legibility problem: your content has to be parseable. Agentic commerce is an operations problem: your price, your stock count, your shipping rule and your returns policy have to be correct at the second of purchase, with nobody in the loop to catch it.
Most stores are already exposed to the first one. The second arrives with money attached.
How agentic commerce works, step by step
Strip away the vendor decks and the loop is short. Four of the five steps happen somewhere you do not control.
| Step | What happens | What the agent reads | Where stores drop out |
|---|---|---|---|
| 1. Instruction | Someone asks in plain language: "running shoes under $120, arriving by Friday" | Nothing of yours yet | Nowhere |
| 2. Retrieval | The agent pulls candidates from feeds, merchant APIs and page markup | Product feed, JSON-LD, structured attributes | Thin or stale feed data. You are never considered rather than rejected |
| 3. Ranking | Candidates scored against the stated constraints | Price, delivery window, return terms, availability | Attributes that exist but are not legible, or disagree with the page |
| 4. Transaction | Some flows pay with a stored credential, others hand back to your checkout | Commerce endpoint or your checkout flow | Forced accounts, CAPTCHA, steps that need a human hand |
| 5. Confirmation | Order lands, outcome feeds the next recommendation | Your confirmation data | Silent failures nobody reconciles |
Step 2 is the one merchants control and mostly ignore. It is also the cheapest to fix. Everything after it is somebody else's platform decision.
Agentic commerce vs traditional online shopping
The clearest way to see what changes is a straight comparison of the two checkouts. Same store, same catalogue, different buyer.
| Step | A person shopping | An agent shopping |
|---|---|---|
| Finds you | Search, ad, social, direct | A feed, an API, or a model that already read your pages |
| Reads the page | Images, reviews, layout, trust signals | Structured data and text, no design at all |
| Compares | Two or three tabs, then a gut call | Every option it can reach, on stated attributes |
| Price it acts on | What the page shows now | What your feed said, which may be hours old |
| On a mismatch | Notices, adjusts, sometimes still buys | Buys the wrong thing, or drops you silently |
| You recover by | Abandoned cart email, retargeting | Nothing. There is no cart and no address |
Read down the right column and you can see where the work is. Every row that goes wrong goes wrong faster than a human buyer would let it, and none of your usual recovery tools reach it.
Agentic commerce examples, and who is actually building it
The named activity sits with payment networks and platform vendors. Mastercard announced Agent Pay in April 2025, added further agentic-commerce tooling that September, and in June 2026 launched Agent Pay for Machines. That is the honest answer to what is agentic commerce Mastercard: a payments programme with a public release history you can read.
IBM, Stripe and Salesforce each run their own definition page. McKinsey published a piece on the agentic commerce opportunity. People search for "Agentic Commerce Protocol" often enough that it shows up in related queries, which tells you the standards conversation has started.
Notice what those agentic commerce examples have in common. Every one of these companies sells infrastructure. None of them is a mid-market store reporting revenue from agent traffic, which is exactly what that Reddit thread is asking for and not finding.
How big is agentic commerce, really
Nobody measures agent-completed transactions. Everything published as a market size is a forecast, and the ones worth reading say so.
| What it is | What it actually covers | Can you act on it |
|---|---|---|
| Measured AI-referred traffic | Visits and conversion from generative-AI sources, reported quarterly by Adobe Digital Insights | Yes. It is a measurement of real sessions |
| Measured growth rates | Adobe recorded a 1,200% jump in generative-AI referral traffic to US retail in March 2025 | Yes, as direction. The base was tiny |
| Market-size figures | Analyst projections for 2030 from several firms | No. A forecast is not a number about your store |
Traffic referred by an AI assistant and a purchase completed by an agent are different events. Most of the confident percentages circulating measure the first and get described as the second. If you repeat one, say which it is.
Here is why the number is genuinely hard. An agent-assisted purchase usually arrives in your analytics as direct traffic, or as a referral from an assistant that strips its referrer. You cannot separate it from the rest without instrumenting for it on purpose.
So most stores have no idea what their agent traffic is, including the ones quoting a percentage. What you can measure is whether you are reachable, and that is available today.
What agentic commerce costs to get ready for
The spend lands in three places, and the order matters more than the total.
- Feed and structured data first - complete, current attributes and page markup that agrees with them. Mostly correcting what exists rather than building anything
- Checkout second - whether your flow survives being driven by something that is not a person with a mouse
- Platform integration last, and worth waiting on. The standards are moving, and buying into one early is how you migrate twice
The cheap work is the work that matters, and it is the work nobody is doing.
Agentic commerce risks and challenges for retailers
The explainers stop at the opportunity. From what I have seen, the parts that bite are all operational, and they show up before the volume does.
- Disputes get harder to argue. A chargeback where nobody claims to have clicked buy is a different conversation with your processor than a normal one. Ask yours today what evidence they expect on an agent-initiated order, and get the answer in writing.
- Returns arrive in volume, not in ones. An agent that misreads one attribute on a product does not misread it once. It repeats the same order until something stops it.
- A stale feed becomes a pricing error. Your site and your feed disagreeing by a few hours is survivable with human buyers who see the live page. An agent acts on the feed.
- One bad field scales. A wrong size chart or a missing shipping restriction on a single SKU is a slow trickle of complaints today. Read by machines at machine pace, it is a queue.
- You lose the recovery path. No cart, no email, no retargeting pixel. An agent that rejects your store on a data problem leaves no trace that it was ever there, and you will not know why revenue moved.
None of this argues against getting ready for agentic commerce. It argues for fixing the data first and treating the sales channel as second.
How to audit your store for agentic commerce this week
None of this needs a budget or a vendor call. Your feed clean? Your checkout drivable without hands? Go and find out.
- Open your product feed and read twenty rows. Not the dashboard summary, the actual file. Price, stock, shipping estimate and return window populated and current on every row.
- Run three product URLs through Google Rich Results Test (https://search.google.com/test/rich-results). Missing Product schema, or a price in the markup that disagrees with the page, gives an agent two answers and it trusts neither.
- Read your robots.txt for what you are blocking. Plenty of stores blocked AI crawlers in 2024 and never revisited it. Blocking the thing that would have recommended you is a decision worth making on purpose.
- Buy something from your own store in a private window, on a phone, without logging in. Count the steps that need a human hand. Each one is a place an agent stops.
- Ask your platform what it exposes to agents today. The answer tells you whether the remaining work is yours or theirs.
That is a couple of hours. Do it before you read the next forecast.
Where this leaves a store owner
Is agentic commerce the future? Nobody honestly knows, and the confident answers in circulation are forecasts being quoted as measurements. The rails exist and the purchase counts do not, so treat it as a channel worth being ready for rather than one worth reorganising around.
People typing what is agentic commerce into a search box are usually deciding whether to care yet. The caring part is already behind you and the doing part is small.
The stores that get picked by agents will not be the ones that adopted the earliest protocol. They will be the ones whose data was clean when someone's assistant went looking. Go and read your feed.
FAQ
What is agentic commerce in simple terms?
Buying where an AI agent does the finding, comparing and sometimes the paying, acting on an instruction a person gave it earlier. The practical difference for a store is that the customer reads machine-readable data rather than your copy and photography.
What is agentic commerce Mastercard?
Mastercard announced its Agent Pay programme in April 2025, added further agentic-commerce tooling in September 2025, and launched Agent Pay for Machines in June 2026. It is a payments programme with a public release history, not a single product launch.
What are some examples of agentic commerce companies?
The named activity sits with payment networks and platform vendors - Mastercard, Stripe, IBM and Salesforce all publish on it, and McKinsey has written about the opportunity. What is missing from the list is mid-market stores reporting revenue from agent traffic.
How big is agentic commerce?
Nobody measures agent-completed transactions. Adobe Digital Insights measures AI-referred traffic and conversion quarterly, which is real. Every market-size figure in circulation is a 2030 forecast, and traffic referred by an assistant is a different event from a purchase completed by an agent.
What does agentic commerce cost to prepare for?
Feed and structured data work first, which is mostly correcting what already exists. Checkout second, testing whether your flow survives being driven without a human hand. Platform integration last, because the standards are still moving.
Is agentic e-commerce different from agentic commerce?
No. Agentic e-commerce is the same idea with the word retail attached. The mechanism is identical: an agent retrieves candidates from feeds and markup, ranks them against stated constraints, and either pays or hands back to your checkout.
Is agentic commerce the future?
Nobody honestly knows, and the confident answers you read are forecasts being quoted as measurements. What is checkable today: the payment rails exist, several large companies have shipped agent-facing checkout, and no one publishes a count of purchases an agent actually completed. So treat it as a channel worth being ready for, not one worth reorganising around. The work that prepares you for it - clean feeds, correct stock, machine-readable policies - pays for itself whether or not the agents arrive in volume.