Picture a company spending real money to build a store in the middle of the desert.
That is roughly the bet a lot of companies in agentic commerce are making right now. Engineering and capital are pouring into brand new destinations, new places to shop, on the assumption that the volume will show up once the store exists. The shopping agent. The chat where you ask for a fitness watch and one arrives at your door. ChatGPT, Phia, Daydream, a dozen more behind them. Beautiful demos, real funding, enormous reach. What none of them has shown yet is people actually buying at scale, because a brand new destination has to do the single hardest thing in commerce before anything else: convince a person to show up and start shopping somewhere they have never shopped before.
That is a store in the desert. Gorgeous building. Nobody walking past.
The last place you go looking for food and water is the desert. You go to the supermarket in the city. The traffic is already there. The intent is already there. People walked in knowing they came to buy something.
I came into agentic commerce last May, from the research side. At MyCustomAI, the early work was on agentic commerce optimization, and on a simple question... could a checkout agent actually complete a purchase on a real merchant site? From there I moved to the infrastructure side at Rye, helping build the rails that let an agent check out anywhere, on any catalog, without a merchant lifting a finger. Two sides of the same thing, the agent behavior and the plumbing.
So when the conversation turns to volume in agentic commerce, the question I keep landing on is simple. Why keep building stores out in the desert, when we could make the supermarket people already walk into ten times better?
Where the volume actually is
Think about the places you already spend money without anyone having to talk you into showing up.
You open your credit card app to check a balance, and a rewards portal is sitting right there, offering points if you buy through it. You open a cashback app out of habit before a purchase you were going to make anyway. You open your airline's app and find a shopping mall bolted onto it that pays you miles. You open a retailer's loyalty app because you have points to spend before they expire.
None of these had to convince you to shop. You arrived with the intent already loaded. And each one already has the four things a brand new destination has to build from scratch: a logged in user, a stored payment method, a daily reason to open the app, and your trust.
That is the supermarket. The traffic and the intent are already inside.
So the move is straightforward. Let the agent complete the purchase right there, inside the cashback or rewards or loyalty experience the person already opened, across any catalog, with no new integration work for the merchant. Native checkout becomes a feature on top of intent that is already there.
When you put checkout where the intent already lives, the volume question mostly answers itself.
Embedded finance already ran this play
Embedded finance learned this a decade ago. Affirm and Klarna put financing right on the buy button people were already clicking. Stripe and Marqeta let any app issue cards and move money inside the product it already had. Banking showed up inside Shopify, inside ride-share apps, inside the tools people opened every day. The financial product met the intent where it already lived, and the volume followed.
Agentic commerce runs on the same law. The purchase gets completed where the buyer already is. A lot of today's funding is going the other way, into destinations that have to manufacture the visit before they can do anything else.
Everyone was waiting on OpenAI
For a while, the whole space was watching OpenAI to lead the agentic commerce revolution. They shipped Instant Checkout, the ability to buy from a retailer right inside ChatGPT. Within about six months they pulled most of it back. Only around a dozen of Shopify's millions of merchants ever went live, and OpenAI shifted toward shopping inside connected apps a user has to intentionally open, plus a bigger push into product discovery instead of one-tap checkout. (CNBC)
The reaction was instant: if OpenAI can't make shopping in chat work, nobody can.
People were asking ChatGPT plenty of product questions. They just were not finishing the purchase there. Lots of research, very little checkout.
That does not mean chat commerce is dead. The shopping and product queries running through these tools are real and growing, on the order of tens of millions a day, by industry estimates. People are clearly using them to figure out what to buy. Transacting there will come. It just is not where the volume lives today, and building your company as if it already is means standing in the desert waiting for the crowd to arrive.
Intent without a checkout
The obvious objection is reach. ChatGPT has hundreds of millions of users. Why call that a desert?
Reach is not the same as a finished purchase. Hundreds of millions of people open ChatGPT, and a real and growing share are asking what to buy. The shopping intent is there. What is not there, at least not yet, is a native way to complete the purchase. OpenAI stepped back from that and now routes buying into merchant-built apps you have to open on purpose, like the one Sephora built. So we don't actually know whether chat would convert at scale with native checkout. That version hasn't been built yet.
A cashback or rewards app does not carry that question. The intent is there and the checkout already works. The supermarket beats the empty store in the desert, every time.
What the rails are really for
The importance here is the infrastructure, more than the discovery. Discovery is great, and LLM-driven discovery is a genuinely new way to come across products, which is exactly why getting your brand surfaced in those answers, GEO, is becoming its own discipline. But the bigger shift is that agents are going to be making purchases of every kind, everywhere, not just in the places we think of as shopping.
This is also what the infrastructure was for, though it took me a while to see it from the inside.
The headline feature was always simple: give the system a product URL, get back a completed order, with no merchant integration. The deeper value showed up in where it let commerce happen. When checkout takes zero integration work from the merchant, any surface can host it. A cashback app, a rewards portal, a loyalty program, a chat inside an app people already trust. A surface that already owns the customer can turn itself into a storefront in an afternoon, across millions of products, without asking a single merchant to do anything.
And to be clear, this is not about getting merchants to adopt a new protocol or standard. There is a lot of noise right now about competing agentic-commerce protocols. The rails I am describing work on the checkout a merchant already has. The surface that owns the customer plugs in, and the merchant does nothing.
The rails turn the supermarket into a place that can actually sell you the whole catalog.
What I would do about it
Find the surfaces that already have a logged in, paying, daily-active audience with money on their mind, and put agentic checkout inside them. Cashback, rewards, loyalty, banking apps, anywhere the buying mindset is already switched on. Meet the intent that already exists.
If you are a brand, you have two jobs. Be findable when an agent does the discovery, which is where GEO is becoming essential and a wave of companies is now building. And be reachable where the buying already happens, so that when an agent completes a purchase on someone's behalf, your product is one it can actually buy. Being legible to a destination agent is table stakes. Being available where the buying already happens is where the volume is.
The next wave of commerce shows up quietly, inside the apps you already open, at the moment you were already going to spend. The companies that win it are the ones stocking the shelves where the people already are, instead of building another store in the desert.
Frequently Asked Questions
Where is the volume in agentic commerce today?
From what I saw on the infrastructure side, the volume that converts today shows up in the surfaces where people already shop with intent: cashback, rewards, loyalty, gifting platforms, really many different existing commerce experiences. They already have the logged in user, a stored payment method, a daily reason to open, and trust, so adding agentic checkout converts. Brand new shopping destinations have to manufacture that demand first.
Did OpenAI's Instant Checkout fail?
OpenAI pulled most of Instant Checkout back within about six months, with only around a dozen Shopify merchants live, and shifted toward merchant-built ChatGPT apps and product discovery. Users asked many product questions but rarely completed purchases in chat.
Does embedding agentic commerce require new protocols?
No. The infrastructure works on a merchant's existing checkout with no integration or protocol adoption. The surface that already owns the customer plugs in, and the merchant does nothing.