Text Team
Best content from Text
See what's new at Text. We share updates on how we help businesses turn customer conversations into revenue, qualified leads, and measurable growth.
Agentic commerce is what happens when AI agents take part in shopping and buying. A customer can ask an assistant to find a product, compare options, ask a store questions, and help complete the purchase. The merchant still has to make the business understandable: product data, policies, payment, delivery, chat, and service answers all become part of the buying system.
That is why agentic commerce is bigger than ecommerce with a chatbot attached. It changes who reads your store, where purchase decisions happen, and what counts as useful sales and support content.
Quick answer: agentic commerce in one sentence
Agentic commerce is commerce where an AI agent can act for a buyer or help a business transact with that buyer, from discovery and comparison to payment, fulfillment, and follow-up.
Stripe describes agentic commerce as a way to “build agents that pay” and to “sell products, services, and APIs to agents” in its agentic commerce documentation. The Agentic Commerce Protocol repository frames ACP as an open standard for connecting buyers, their AI agents, and businesses to complete purchases.
Those definitions point to the same shift: the customer may still be human, but the first reader, comparer, negotiator, or checkout assistant may be software.
The three layers of agentic commerce
It helps to separate agentic commerce into three layers.
1. The buyer-side agent. This is the assistant working for the shopper. It may search, compare products, ask questions, summarize tradeoffs, monitor price, or prepare a purchase decision. The buyer may approve the final transaction, or the agent may act within permissions the buyer has already set.
2. Merchant readiness. This is everything the business makes available for the agent to read and use: product attributes, prices, inventory, policy pages, delivery promises, reviews, support content, chat answers, and escalation paths. If these sources are vague or inconsistent, the agent has weak input.
3. Transaction infrastructure. This is the layer that lets the purchase move from recommendation to action: catalog feeds, checkout APIs, delegated payment credentials, order status, authentication, and protocols such as ACP or UCP.
A lot of confusion comes from talking about one layer as if it were the whole system. A shopping assistant alone is not enough. A perfect product feed is not enough. A payment protocol is not enough. Agentic commerce needs the buyer’s intent, the merchant’s reliable data, and a safe way to complete the transaction.
What changes for merchants
For merchants, the practical change is that product and service information has to work for another reader.
A human shopper can tolerate a fuzzy product page. They can infer what “works with most systems” might mean, click a return-policy link, open a chat, and decide whether they trust the answer. An agent has less judgment by default. It retrieves facts, compares fields, and works from the sources it can reach.
That is why agentic commerce turns old ecommerce housekeeping into a revenue problem. Product specs, compatibility, variants, exclusions, shipping conditions, warranty rules, and return windows have to be explicit. Discounts need approved logic. Chat answers need to come from current policies rather than improvisation.
For the product-selection side, see what makes an AI agent pick your product. The short version: agents are not perfectly rational buyers, and they do not ignore every old persuasion cue. They work with whatever signals the system gives them. That makes clean, verifiable product data more important, not less.
Where chat fits
Agentic commerce is often discussed as a checkout or product-feed problem. It is also a conversation problem.
A buyer’s agent may ask the same questions a person would ask before buying. Does this fit the item I already own? Can it arrive by Friday? Does the discount apply to this bundle? What happens if the size is wrong? If the store-side AI can answer from approved sources, the conversation can reduce uncertainty. If it cannot, the agent may leave before a person ever sees the page.
The same pattern shows up in AI shoppers in ecommerce chat: product pages, help articles, policies, and order systems become raw material for the conversation. The chat layer does not replace the store. It exposes whether the store is understandable enough to act on.
Where protocols and payments fit
Protocols matter because the open web was not designed around agents completing commercial tasks. Pages can describe a product, but an agent also needs structured information, permissions, identity, payment, order confirmation, and a way to handle exceptions.
OpenAI’s Agentic Commerce Protocol documentation describes ACP as a connective layer between merchants and ChatGPT users, using structured catalog data and inventory to surface relevant products in context. Stripe’s documentation also points to catalog feeds, shared payment tokens, and the Universal Commerce Protocol as building blocks.
Business teams do not need to become protocol experts to start preparing. They do need to understand what those protocols reward: structured catalogs, accurate availability, clear policies, secure payment permissions, and systems that can answer the questions a buyer’s agent will ask.
How agentic commerce differs from ecommerce automation
Traditional ecommerce automation usually starts inside the business. A merchant automates emails, recommendations, support routing, discount rules, or abandoned-cart flows. The business decides the path and the customer moves through it.
Agentic commerce adds a buyer-side actor with its own task. The agent may skip the path you designed. It may compare your product against competitors, ask one precise question, request proof, or bring its own user context into the decision.
That does not make your site less important. It changes the standard. The store has to be readable by people and usable by agents. The better your underlying data and policies are, the less the agent has to guess.
How to get ready without rebuilding your store
Start with the parts any buyer needs to trust.
Make product attributes specific. List dimensions, compatibility, materials, variants, inventory, delivery conditions, and exclusions in places your systems can retrieve.
Turn policies into answerable rules. A clear return window with exceptions beats a friendly paragraph that says “standard rules apply.”
Keep offer logic approved. If an AI assistant can negotiate, it needs boundaries for eligibility, margin, bundles, and escalation.
Treat chat as an agent-facing contact channel. A buying agent may use your chat widget to confirm product fit, delivery, policy, discount, or handoff questions before the shopper opens the product page. Make sure the answers come from current sources and that the conversation can escalate with context.
Test the conversation. Ask an agent to compare, question, and prepare a purchase. Then inspect which sources it used and where the answer broke.
Treat transcripts as demand research. The questions agents ask will show you the product, policy, and content gaps that humans were working around.
Agentic commerce is still early, and the vocabulary will keep shifting. The useful move now is not to chase every new assistant. It is to make your business legible to software acting for a real customer.