What is agentic commerce?

The shift from search to agent is already happening. Most businesses are not ready for it.

The definition

Agentic commerce is when an AI agent completes a purchase or transaction on behalf of a person without that person navigating a website, comparing options, or clicking buy. The person states a need or a goal. The agent researches, evaluates, selects, and, increasingly, completes the transaction. The person approves the result, but they do not do the work.

That is different from search. A search engine returns options and lets you pick, while an agent returns a decision and executes it. There is no scrolling through five tabs and no form to fill out. You ask, and the agent finishes the job.

Agentic commerce is already live in industries where convenience beats brand loyalty. Ride sharing, appointment booking, restaurant reservations, home services, and commodity purchases happen this way now. The agent asks, compares, and books for you, and in these categories that is already normal.

How an AI agent shops differently than a person

A person browses and compares websites side by side. An agent queries, reading structured facts from every source it trusts and weighting them by recency, consistency, and authority. A person clicks through checkout flows because they have a mouse and they understand captchas. An agent needs an interface built for machines, or it cannot finish the purchase.

A person will forgive an inconsistency, but an agent will not. If your website says you are open until 8 pm and your Google listing says 7 pm, the agent does not call to ask. It moves to the competitor whose facts agree. If your pricing is not published anywhere a machine can read it, the agent does not request a quote; it buys from the business that listed the number.

A person will put in some effort to understand what you sell. An agent has twenty other options that all load in under two seconds, so it will not bother. Only the businesses a machine can read get evaluated, and everyone else is left out of the conversation. For a closer look at how shopping agents pick products today, see agentic shopping.

This is the efficiency people wanted when they hired an assistant. The assistant does the tedious work, and the person approves the result. What most companies have not realized yet is that the assistant cannot buy from you if your business is only built for humans.

The three tests every AI agent runs

Every agentic commerce transaction, in every industry, runs through the same three gates. I call them Found, Chosen, Booked. Fail one and you are out; you need all three to get the sale.

Found

Can the agent locate your business and read accurate facts about what you offer? Ranking matters less here than whether you are in the dataset the agent queries when it runs the search at all. If ChatGPT, Gemini, Perplexity, and the next ten assistants to launch cannot see you, you are not competing for that customer.

Being found means your name, services, locations, hours, and contact information exist in structured formats machines can parse, and that those facts match everywhere they appear. The agent should never have to interpret your marketing copy to understand what you do. You tell it plainly, in the same language every other business in your category uses, and you tell it the same way in every place a machine looks.

Chosen

Once the agent has found you, it has to decide whether you are the right choice, and that comes down to trust. Can it confirm that you deliver what you claim, that people who hired you were satisfied, that your facts are current, and that sources it already trusts have verified you?

The agent weighs reviews, citations, structured data, recency, and consistency. Marketing language barely registers with it. It looks for the same signals a careful person would check if they had unlimited time and perfect recall. What gets a business through this test is an actual record that supports the claim, whatever the copy says.

Booked

This is the gate where most companies lose. The agent has found you, it trusts you, and now it needs to complete the transaction. Can it? If the only way to buy is a form designed for a person with a mouse, or a phone number that rings six times and goes to voicemail, or a chatbot that asks for information the agent already provided, the answer is no, and the agent moves on to the competitor who gave it a way to finish.

Agentic commerce requires a door the agent can walk through. That door is either a machine interface, live inventory and pricing an agent can query and transact against directly, or a voice channel, a line the agent can call or hand its user to when the conversation is what closes the sale. Most businesses need both.

What agentic commerce looks like in practice

A parent asks their assistant to book a pediatric dentist who takes their insurance, is within ten miles, and has an opening this week after school. The assistant queries every dentist in range, checks insurance networks, reads availability, and returns two names with confirmed appointments. The parent picks one. They never saw the websites. They do not know the names of the practices that were in the search but failed one of the three tests.

A founder asks their assistant to order standing desks for the office, same day delivery, under a budget. The assistant searches commercial furniture suppliers, checks inventory, compares delivery windows, and completes the purchase with a vendor that published live stock levels and accepted payment through an API. The founder approved the line item without ever visiting the site. How an agent pays without anyone typing in a card number is explained in agentic payments.

A traveler asks their assistant to hold a table for four at a restaurant near their hotel, top rated, takes reservations. The assistant checks availability through the reservation system, confirms the booking, and sends the confirmation. The traveler never called or compared menus. The agent handled it because the restaurant had built for machines as well as people.

In each case the winning business passed all three tests. The ones that lost were missing one piece, and they never knew the customer existed. For launches that are already live, with dates and sources, see these agentic commerce examples.

The payment rails being built for agents

An agent that buys for you needs a way to pay, and the merchant needs a way to trust it. The card networks and the largest AI companies spent 2025 building both. Mastercard launched Agent Pay on April 29, 2025, with tokens issued to AI agents that have been registered and verified. Visa announced Intelligent Commerce on April 30, 2025, which lets consumers set spending limits for the agents acting for them.

Google announced its Agent Payments Protocol, AP2, on September 16, 2025, and said it was working with more than 60 organizations on it, including American Express, Mastercard, PayPal, Adyen and Coinbase. Google framed the open problems as authorization, authenticity and accountability, the last one meaning who answers when a purchase made by an agent turns out to be fraudulent or wrong. On September 29, 2025, OpenAI and Stripe released the Agentic Commerce Protocol along with Instant Checkout in ChatGPT, and Stripe issues a payment token that only works for one merchant and one cart total.

For a merchant, these systems mostly solve the Booked test, the moment the agent pays. They do not decide which stores make the short list an agent builds before it buys. How the tokens and signed mandates work is covered in agentic payments.

I measured that short list for one category. In a study I ran on November 6, 2026, ChatGPT named none of 50 randomly chosen independent US coffee roasters when asked five common questions about buying coffee online, and Perplexity named one. Most of those stores were readable by a machine, with 36 of the 50 publishing price and availability in structured data, and that alone did not get them recommended.

Which industries are already affected

Any industry where the buying decision can be reduced to criteria is already in play. That includes appointment based businesses, home services, hospitality and travel, healthcare providers, local retail, and professional services where the deliverable is standard enough that an agent can evaluate fit.

If your customers currently find you by searching, reading reviews, comparing a few options, and then booking or buying, an agent can do that entire sequence faster and without showing them your site. If your close rate depends on a person seeing your brand, your design, or your differentiation before they commit, you have more time, but less than you think.

The agents are learning how to evaluate subjective criteria. They ask clarifying questions, they infer preferences from past behavior, and they weight factors a person would care about but might not articulate. As the models improve, the range of purchases they can complete without human input expands. If you wait until your industry is clearly affected, you will be catching up to competitors who built for agents while the behavior was still forming.

What it means for your business right now

If you sell anything a person currently finds by searching, you are competing in an agentic commerce market whether you have prepared for it or not. Agents already shop your category. What you need to know is whether you pass the three tests, and whether you would even notice losing to a competitor who did.

Start with the facts. Every platform a machine reads, your name, your offerings, your location, your hours, and your contact information say the same thing. One inconsistency breaks trust, and the agent moves on.

Answer the real questions. Find the fifteen to twenty questions a buyer asks an assistant before they choose a company like yours. Make sure your site answers each one in plain language a machine can quote. Write them as answers, and leave the marketing language out.

Build a door. Can an agent complete the transaction without a human on your side intervening? If the answer is no, you are only in the market people shop when they have time to call, and the agent is shopping somewhere else.

Businesses that pass all three tests will become the default answer, because agents remember who passed and they go back. That position is still open in most categories today.

Being early matters less than being ready, because your customers already ask agents what they used to ask search engines. So can the agent buy from you today? Finding that out is the first thing an agentic commerce advisor does.

FAQ

What is agentic commerce?

Agentic commerce is when an AI agent completes a purchase or books a service on behalf of a person. The person states a need, the agent researches, evaluates, and transacts, and the person approves the result. The person does not browse websites, compare options, or fill out forms. The agent does the work.

How is agentic commerce different from ecommerce?

Ecommerce is built for people. A person browses a website, adds items to a cart, and clicks through a checkout flow designed for eyes and a mouse. Agentic commerce is built for machines. An agent queries structured data, evaluates options by weighted criteria, and completes the transaction through an API or a voice channel. The person never sees the site.

What are examples of agentic commerce?

Booking appointments through an assistant, ordering products by asking an agent to find the best match for specific criteria, reserving tables at restaurants via voice or chat interfaces, and purchasing services where the agent researches providers, checks availability, and confirms the transaction without the person navigating a website.

How do AI agents decide which company to buy from?

Agents run three tests. Found: can the agent locate the business and read accurate facts. Chosen: can it verify trust signals like reviews, citations, and consistency. Booked: can it complete the transaction without human intervention. The businesses that pass all three get the sale. The rest are not in the conversation.

What is Found. Chosen. Booked.?

Found. Chosen. Booked. is the three gate framework every agentic commerce transaction runs through. Found means the agent can locate your business and parse your facts. Chosen means it trusts you based on signals it can verify. Booked means it can complete the purchase or reservation. Fail any gate and the agent moves to a competitor.

How do I prepare my business for agentic commerce?

Fix the facts first. Make sure your name, services, hours, and contact information are identical everywhere a machine can read them. Answer the real questions buyers ask before they choose a company like yours, in plain text an agent can quote. Build a door the agent can complete a transaction through, either a machine interface or a voice channel that answers. Then measure whether agents recommend you when someone asks for your category.