Agentic Commerce is Here

This story was originally published on LinkedIn.

AI is beginning to change the way we shop online.

This week, OpenAI announced “ChatGPT Shopping,” a new feature that lets users browse and buy products directly within ChatGPT. It’s the company’s first big move into e-commerce, and it follows Microsoft’s launch of the CoPilot Merchant Portal last week, designed to help retailers sell through AI-powered tools. Perplexity got a jump start on Microsoft and ChatGPT when it launched a full shopping feature, “Buy with Perplexity,” allowing users to purchase items directly from chat results in November of last year.

Together, these moves signal the start of a major shift: shopping won’t always happen on traditional websites or apps anymore. Instead, transactions will increasingly happen inside the conversations we’re already having with AI tools.

In some ways, this idea isn’t entirely new. Platforms like TikTok have shown how product purchasing can happen natively in a discovery experience, as we wrote about in our post on Distributed Commerce a year and a half ago. Users see something they like and buy it without traversing to the merchant’s site.

But what’s emerging now is even bigger: AI is enabling intent-based shopping, where you tell the AI what you want — like you would type into Google — and it can instantly find and sell it to you, all in one seamless flow.

This isn’t just a set of press releases — we’ve talked to multiple top 30 retailers/brands that are in active discussions with ChatGPT or Perplexity to feed their catalogs and to discuss how transactions could flow. Commerce giants like Shopify are moving very quickly as well, and have served as the underpinning for some of the early transactional capabilities that are in the market today.

While the momentum is clearly there, we are still quite early, and the current implementations are clunky at best. The Perplexity offering seems to be the furthest along, but purchasing only works on a select set of merchants, resulting in a somewhat disjointed purchasing experience. For instance, one of my recent purchases of athletic shorts resulted in two ShopPay authentication code text messages, despite the promise of a zero-touch checkout process.

We’ve outlined our thoughts on three potential models that merchants and the engines could pursue. Affiliate, Facilitated, and Marketplace. Each has their own pros and cons, with unique opportunities for enabling technology/startups.

Affiliate Model

This is the simplest approach — where users use AI search engines to find the right product and the consumer is then sent to the merchant website to make a purchase. This is the process initially taken by ChatGPT.

Pros:

• Simple: This is consistent with historical approaches to product discovery on the web. Consumers will understand and be comfortable with this flow

Cons:

• Not Revolutionary: This approach doesn’t take advantage of the power of these AI platforms or enable the integration of search and purchase into one motion

Capabilities Required:

• Feed Management: AI search platforms will need/want real-time feeds with product information, inventory availability, and pricing to deliver great results.
• Visibility Optimization Tools: Merchants will need tools to help get their products into the search results of these AI engines.
• Startup Examples: Profound, xFunnel, Evertune, Bluefish, and eComtent.

Facilitation Model

This is the most complex model, but perhaps also the most interesting. In this model, the consumer never leaves the AI search platform to make the purchase, but the underlying retailer/brand remains the merchant of record. Perplexity is doing this now for a small percentage of products using its “Buy With Perplexity” feature.

Pros:

• Seamless Experience: Once a consumer finds the product or service that they want, they simply click “buy” rather than being redirected into an external shopping experience
• Merchants Retain Ownership of the Customer: Merchants (likely) do not give up (full) ownership of the customer.

Cons:

• Complex Checkout Process: It will not be simple to process the checkout. Merchants have a wide variety of site architectures, cart functionalities, and checkout flows. The AI search engines will either have to connect programmatically to the checkout system for each merchant (hard to get full coverage) or figure out how to navigate the visual/user facing version of checkout using technology like Operator (hard to drive 98%+ accuracy or deliver a quick consumer experience) to create a cart and checkout.
• Complexity Around User Accounts: If a consumer wants to make a purchase using their account on a merchant site, there will be further complexity.
• Consumer Messaging Required: Consumers will need to be trained that their purchase came from the underlying merchant, and not the AI Search engine.
• Potentially Disjointed Experience: Unless the AI search engine can make the checkout experience consistent across merchants (regardless of the merchant’s e-commerce engine), the experience could be disjointed (as it is today).

Capabilities Required:

• Same Capabilities from the Affiliate Model: Feed Management and Visibility Optimization
• Cart Creation Tools: The AI search platforms will need to be able to create a merchant cart on behalf that is populated by SKUs identified by the consumer in a chat session. This will either happen utilizing APIs, or with an AI Agent traversing the visual shopping experience of a site.
• Cart Checkout Tools: These tools will also need to be able to “check out” — essentially connecting a payment to a cart. While much of the payment processing infrastructure will likely remain unchanged, functionality will need to be built to “ check out” without going through the merchant’s typical visual checkout process.
• Startup Examples: Firmly, Violet.io

Marketplace Model

In this model, the AI search platforms would operate as a 3P marketplace where the search platform would be the system of record and the fulfillment is handled by the brand (or retailer) that has inventory of the product. This would be the most hands-on approach, and would require the AI search engines to take on customer service, payments, chargebacks, returns, etc.

Pros:

• Familiar Model: Consumers understand marketplaces.They understand that (for the most part), the marketplace will stand by the transaction if something goes wrong, so they just need to trust the marketplace
• Less Technically Complex than Facilitation Model: While there is significant work in setting up a marketplace, it is a known/solved problem that wouldn’t be an issue for any of the major players.

Cons

• AI Search Platforms Become Retailers: It is not clear that these players would relish becoming retailers.
• May Leave Out Many Merchants: Some merchants will be glad to participate in a marketplace, but some will only want to sell direct. As a result, not all products would be available, and the experience would be disjointed.

Capabilities Required

• Same Capabilities from the Affiliate Model: Feed Management and Visibility Optimization
• Marketplace Enablement Tech: AI search platforms will require e-commerce marketplace capabilities — minus the visual UX. This could be a stripped down version of an existing marketplace tech platform, or something built from scratch
• Startup Examples: Mirakl and Marketplacer (legacy players). If AI platforms choose this path, new players will likely emerge to power UX-free marketplaces.

Our Takeways

What’s clear is that AI-powered shopping is no longer a theoretical concept — it’s unfolding in real time. While the current experience is far from perfect, the pace of innovation and the level of interest from major retailers and platforms suggest we’re at the beginning of something big.

Whether it’s through affiliate links, seamless in-chat transactions, or full-blown marketplaces, AI search engines are poised to become powerful new channels for commerce. For merchants, this moment is both a challenge and an opportunity: figuring out how to show up in these new environments, how to maintain control of the customer relationship, and how to enable transactions in a way that feels native to AI-driven interfaces.

We’re still early, but the building blocks are being laid. We are currently looking for any exciting startups that will help move this forward. In particular, companies that are helping merchants expose their product catalogs to the AI engines and startups that are building an API-based checkout layer that would enable more seamless transactions generated from AI searches.