Agentic Commerce: When the Buyer Is an AI Agent

What is agentic commerce?

In agentic commerce, whoever is browsing your store may be software buying with its owner’s permission.

Agentic commerce is a purchase made by an AI agent on behalf of a person. Instead of opening ten tabs, comparing sizes and filling in the payment form, the customer tells an assistant what they need, their budget and their deadline, and the agent searches, filters and, if the customer authorizes it, pays.

What makes it possible isn’t just a more capable AI model, but a layer of standards defining how an agent reads a catalog, builds a cart and pays without exposing card details. That layer was published between September 2025 and January 2026, and for a store it means something concrete: besides persuading a person, it has to be readable by a machine.

Don’t confuse it with SEO for AI, which aims to get ChatGPT or Google to mention you in an answer. Agentic commerce goes a step further: the agent can complete the transaction with you, or at least has accurate enough data to recommend you with confidence.

What exists today: ChatGPT, Google and payment protocols?

There are two major standards, one backed by OpenAI and Stripe and the other by Google and Shopify, and both are still evolving.

Two families of agentic commerce standards exist today. The first is the Agentic Commerce Protocol (ACP), created by OpenAI and Stripe; it is still in beta, and its latest stable version, dated April 17, 2026, adds cart, feed, orders, authentication and MCP support. The second is the Universal Commerce Protocol (UCP), which Google launched on January 11, 2026 with Shopify, Etsy, Wayfair, Target and Walmart, and which builds on the AP2 payments protocol.

The landscape moved fast, and not everything that was announced stayed the same. When Shopify switched on selling to ChatGPT users for millions of merchants in March 2026, it explained that purchases complete in an in-app browser and that on desktop ChatGPT links out to the merchant’s store. In other words, for many merchants the agent recommends and hands off to the store’s own checkout. Here is the timeline as verified in official sources:

Is agentic commerce coming to Colombia and Latin America, and when?

Native agent checkout hasn’t been announced for Colombia, but visibility to agents is already in play.

As of this writing, none of the official announcements we reviewed list Colombia as a market for native checkout in ChatGPT or Google: in-agent purchasing launched in the United States. No one has published a date for Latin America, and anyone promising one is guessing.

That doesn’t mean it can wait, for three reasons. First, agents already recommend products and link to stores, and to recommend you they need clean catalog data. Second, part of the path is already global: Shopify offers its agentic plan worldwide to brands that don’t sell on Shopify, and Stripe announced its agentic suite through platforms such as WooCommerce. Third, getting a store ready for agents is almost the same as running it properly: a clean catalog, accurate prices and stock, clear policies.

What is specific to the region is payments. The protocols rely on tokenized cards and wallets such as Google Pay; PSE, Nequi and cash on delivery don’t appear in the announcements we reviewed. Until that changes, the realistic scenario in Colombia is an agent that finds and compares products and sends the customer to your checkout with Wompi, PSE or card.

What does your store need for an AI agent to understand it?

An agent doesn’t see your design: it reads your data. If the data is wrong, so is the recommendation.

An AI agent understands a store through structured, consistent data, not through its design. OpenAI’s product feed spec for merchants, for example, requires nine fields per product: ID, title, description, URL, brand, seller name, image, availability and price with currency, and recommends uploading a full catalog snapshot at least once a day. This is the readiness checklist we use:

• An exportable catalog feed: a file or API with every product and variant, with a stable ID per variant.
• Structured data on every product page: product schema with price, currency, availability, brand and GTIN where the product has one, without inventing codes.
• Price and stock synced in real time with inventory: an agent that sells something out of stock creates a complaint, not a sale.
• Readable policies: shipping, delivery times, returns and warranty in plain text on their own pages, not buried in a PDF.
• Descriptions with concrete attributes: materials, dimensions, sizes, compatibility. What a person infers from a photo, an agent needs to read.
• A payment gateway with tokenization and fraud screening, and a checkout without unnecessary steps for when an agent sends the customer over.
• Genuine reviews and verifiable contact details: trust signals for people and agents alike.

What are the risks of agentic commerce for brand and margin?

When an agent compares for the customer, your brand competes on data, and whatever you can’t prove with data counts for less.

The first risk is commoditization. If the agent compares specs, price, shipping and returns, a brand that doesn’t stand out on verifiable data ends up competing on price alone. Brand story, photography and site experience still matter, but they reach fewer customers before the decision is made.

The second is losing the direct relationship. If the purchase happens inside an agent, the merchant may see less of the customer’s browsing and get fewer cross-sell opportunities. Google stresses that retailers remain the seller of record in its checkout, and Shopify that checkout customizations carry over, but the conversation before the purchase no longer happens on your site.

The third is operational: amplified errors. A mispriced item or stale stock count is no longer seen by a hesitant person; it is acted on by an agent that trusts the data. And every new channel can bring its own costs, such as platform fees or payment processor charges, so check your margin per channel before switching each one on.

What should you do with your store in the next 90 days?

The next 90 days are for cleaning up data, not for integrating protocols that don’t yet operate in Colombia.

Over the next 90 days, the profitable move is to get your catalog ready for any agent, not to bet on one protocol. In month one, audit your product pages and structured data: prices, availability, variants and policies, and fix mismatches between what the site shows and what inventory says.

In month two, generate a complete product feed refreshed daily, and publish your shipping and returns policies on clear pages. If your store runs on WooCommerce, review the official options with your technical team: WooCommerce’s MCP integration is in developer preview, and Stripe’s agentic suite announced a rollout through the platform. If you are on Shopify, check in the admin how your products appear in the agentic channels.

In month three, measure. Check your analytics for traffic arriving from AI assistants and which pages it visits, and decide what conditions an agentic channel would have to meet before you switch it on: fees, control over customer data and compatible payment methods. And be skeptical of anyone selling “ChatGPT integration” as a boxed product: today what matters most is the quality of your data, and that doesn’t come in a plugin.