Agentic Commerce (Protocol): feeds, checkout and the new sales channel

Karri: Hello and welcome to ReCommerce podcast, the podcast where we talk all things ReCommerce and circular businesses. Today we are focusing on agentic commerce, something that has been in the news lately as OpenAI and Stripe released an agentic commerce protocol where the idea is to bring commerce into the world of AI and LLMs. Today I'm joined by Tuomo Laine, the CEO and co-founder of TWICE Commerce. Welcome.‍

Tuomo: Thank you, Karri. I'm happy to be the recurring guest here.‍

Karri: Nice to have you here. Let’s start from the basics. What is agentic commerce?‍

Tuomo: I’d simplify it as a way of allowing your end customers to buy from you via applications like ChatGPT or other agentic AI platforms. From the end user’s perspective, they can more easily compare items and listings. They go further in the sales process inside the AI chat. When they reach a purchase decision, agentic commerce enables them to purchase inside that chat rather than being directed to your website. It’s a way to let customers buy more easily inside an AI platform like ChatGPT.‍

Karri: So it brings the logic you already use in ChatGPT—you don’t need exact filters. It’s conversational and hopefully understands what you want even if you don’t know the exact terms.‍

Tuomo: Exactly, it’s more immersive. Versus classic googling and bouncing across sites, you can even do checkout inside the ChatGPT session.‍

Karri: Is the idea that you do the actual purchase inside there, or does it replace more touchpoints like returns and order management? Or is it just that one interaction to purchase and then it moves into the traditional online store or email flow?‍

Tuomo: It differs by AI platform, but some key players right now are OpenAI, Stripe, and I’d add Shopify. OpenAI is the leading AI platform—ChatGPT here—Stripe provides payment technology (cards, Apple Pay, shipping details), and Shopify brings a large merchant base and product feeds to be showcased in the ChatGPT session. The agentic commerce protocol helps with discoverability—if you’ve shared your catalog, your products can be discovered and shown in chat. With Stripe or another PSP that supports agentic commerce (Stripe is the forerunner), checkout can happen inside the AI platform. Even if a customer hasn’t attached payment to ChatGPT, a hosted checkout can run in ChatGPT, including shipping. Shopify’s role is making it easy for Shopify merchants to share products for discovery. Post-checkout, platforms like Shopify or TWICE do the heavy lifting: catching the checkout, creating the order in the commerce OS, sending confirmations, and handling returns or planned returns (for rentals). I haven’t seen agentic handling long-term order lifecycle like reverse logistics; current focus is the initial purchase, not the follow-up stages.‍

Karri: So the initial phase is discoverability, comparison, and possibly checkout inside ChatGPT, but then the workflow looks the same for users and merchants as a regular online purchase.‍

Tuomo: Exactly. From the merchant’s perspective, think of agentic commerce simply as a new sales channel with interesting characteristics. One is higher conversion due to the nature of AI platforms: customers do deeper due diligence in a conversation with the agent. If you’re discoverable and recommended, conversion is likely.‍

Karri: We see something similar on our site: sign-ups from ChatGPT traffic convert far better than from Google, though the volume is lower because more of the early journey happens in ChatGPT.‍

Tuomo: They’re high-intent buyers. If they see your product, intent is high. Letting them act on that inside ChatGPT is likely good for business. But as with any channel, there are benefits and costs. You’ll want to manage listings, pricing, and availability per channel because some channels may charge fees. Treat it like other channels: you want to be visible in Google product cards and also in ChatGPT, but you want control over how you’re visible there.‍

Karri: From a merchant’s perspective, how transparent is this channel? Google gives you ranking data you can optimize. Will agentic commerce provide feedback like impressions or engagement? And what’s the incentive for the agentic provider—why show your product? If two stores carry the same item, do they show cheapest price for users or the best commission for them?‍

Tuomo: Thinking of it as its own channel is right. If it’s your only channel, that’s risky, because it may be more of a black box than Google. Providers themselves may not be able to deterministically explain why the model chose Product A over B. You can still test queries and build your own analytics. Platforms may eventually productize visibility metrics, but ranking is probabilistic. They can add moderation or policy layers and, yes, “pay to win” patterns could emerge. Safe merchant bets look similar to other channels: provide a high-quality product feed. The quality and freshness of your feed correlate with visibility. Relevance depends on well-documented attributes, frequently updated data, and truthful stock/availability. Platforms want to protect their users with good experiences, so they rank recent, structured feeds higher. That’s where Shopify and TWICE help—maintaining structured commerce data with sound taxonomy. Even if it’s early and somewhat black-box, the work merchants can do now is exactly what benefits them across channels.‍

Karri: Is it already a good idea to be there? It’s early, U.S.-first; there’s not much public data yet.‍

Tuomo: For an average merchant, say ~$400k in revenue, you probably have bigger levers today. Don’t spend three days a week on this. Instead, future-proof your operation for when agentic becomes common and more platforms adopt the protocol. Concretely: make sure your catalog, pricing, and availability live on top of a solution that accounts for agentic. Shopify is leading here; at TWICE we’ve designed taxonomy, individualized inventory, and cross-channel listings for this future. Even if you don’t use a commerce OS, ensure you have a structured, central source of truth so you can generate a product feed when ready. Invest in data structure and quality—platforms like Shopify and TWICE guide you.‍

Karri: So don’t go all-in yet, but make sure your infrastructure is ready so you won’t need to switch platforms later just to participate.‍

Tuomo: Exactly. And it never hurts to test the experience as a user—run demos, imagine how your products would appear, and consider tweaks that make you more discoverable.‍

Karri: In TWICE, what concrete steps can merchants take to be ready—taxonomy, for example?‍

Tuomo: In our new platform, every stock item can be categorized via taxonomy. Based on that, we recommend attributes to fill (brand, color, pattern, etc.). These are used by other platforms to talk to end customers and categorize products correctly when filters are applied. Listings need similar care—clear titles, descriptions, and commercial text that external platforms can analyze. Many platforms, including Shopify and TWICE, offer AI assistance to suggest attributes and copy so you can review instead of writing from scratch. Relevance also depends on real-time availability—critical for booking businesses. Your catalog should be connected to assets so you always know what’s available when, and at what price. That’s where TWICE helps; Shopify doesn’t do temporal availability natively. With that foundation, it’s easier to publish listings to multiple channels while keeping availability accurate.‍

Karri: For availability, you probably don’t list every timeslot—you expose an endpoint the AI can query.‍

Tuomo: Right. Platforms like ours provide APIs for querying availability and rendering offering + availability. We also package availability into popular product feeds, like Google products or emerging agentic feeds, even if those weren’t designed for bookings. We translate bookable offerings into purchasable feed structures—that bridge work is where TWICE adds value. Google has booking feeds (initially for hospitality), but they can be repurposed for rentals. TWICE effectively translates temporal catalogs into more linear feed formats to keep user experiences smooth.‍

Karri: How do agentic commerce and recommerce fit together? Standardized products are easy, but in resale every item is unique; in rentals, availability is time-bound. Many AI demos focus on travel/booking.‍

Tuomo: They fit great. Travel is a demo focus because it’s research-heavy and complex—exactly the kind of complexity agents help with. Recommerce has similar complexity: condition variance, location, shipping cost, and trade-offs across listings. An AI agent can hold a natural conversation about preferences instead of forcing you across six marketplaces with inconsistent filters. For this to work, platforms like TWICE must expose product feeds compatible with agentic flows. That’s on us—and we’re excited to do that work. I see agentic becoming a major channel for resale and recommerce.‍

Karri: It would also normalize filters and attributes across sites—the AI can unify different condition scales and fields so buyers can actually compare.‍

Tuomo: Exactly. The market is getting better tools: full operating systems like TWICE and specialized services. These are the enablers for growth and a better experience for everyone. Merchants should pick the right platform partners, and technology providers like TWICE will do much of the heavy lifting. We’re implementing fundamental data-quality work now so agents can use that data to deliver experiences small companies can’t replicate alone.‍

Karri: From my experience building AI agents, it all comes down to how the source data is structured—where the source of truth lives, how rich it is, and how reusable it is across agents.‍

Tuomo: 100%. We’re moving from value in specialized UIs to value in structured product and inventory databases with flexibility on who maintains which attributes (e.g., third-party appraisals). The priority is protecting that structured, shareable data so both general-purpose and specialized AI platforms can use it. Yes, we talk about APIs, but increasingly we wrap them with protocols (like MCP) so models can access data in a controlled, understandable way—with good docs. This is a more mature phase for the market. Data modeling is where you create durable value. Giants like OpenAI will run general platforms; smaller specialized AI platforms will appear for specific jobs.‍

Karri: So the best thing a merchant can do now is get data in order and maybe expand attributes—turn “catalog quality” into a competitive advantage.‍

Tuomo: Yes—with one caveat. Simplify where possible. Recommerce offerings can get complex (e.g., granular pricing rules). For agentic channels, simpler pricing usually improves outcomes because agents and protocols are often optimized for linear commerce. I’d encourage merchants to define the simplest pricing that’s still true to the business for these channels. Simple tends to win—and likely improves your chances of being surfaced by platforms focused on fast, easy user experiences.‍

Karri: Great tip. Hopefully we’ll hear from merchants about whether more data or simpler data performs better for agents. Thanks again for the insights—this is a space we’ll revisit as it evolves.‍

Tuomo: Definitely. Thanks for having me.‍

Karri: Thanks for listening to the ReCommerce podcast. This podcast is produced by me, Karri Hyakkönen. You can find us on Apple, Spotify, YouTube, or wherever you like to listen. If you liked it, please share and leave a review on Apple Podcasts, or like and subscribe on YouTube. See you next time.