Agentic Commerce Pushback

If it isn’t E2E & autonomous, it’s not Agentic Commerce

This is another mid-week post that, in the future, will only be available to paid subscribers.

Introduction

I published a post on Agentic Commerce on October 12 that downplayed its likely impact. A few of my friends pushed back with an argument that can be summarized like this:

“I get that it might be hard to go all the way to the payment, but Agentic Commerce will greatly improve discovery”

I decided to use this new channel to engage with that line of thinking. My view is that the assumptions behind this argument are questionable:

• It won’t be E2E. The very definition of Agentic Commerce includes the payment. If the payment is not included, it isn’t Agentic Commerce. It still could be Agentic Discovery or Agentic Search but we should call it what it is

• It won’t be autonomous. Agentic methods may not improve discovery without engaging the consumer in an iterative prompt refinement process. One source described this as “conversational commerce”

• It won’t get the best deals. Most large retailers are blocking AI crawlers, so price discovery may be challenging. On-site, proprietary agents at major marketplaces and retailers don’t have this problem, but only work within a walled garden

At some point, only paid subscribers will get the remainder of this content

E2E: Without payment it’s not Agentic Commerce

In my prior post, I shared Stripe’s definition of Agentic Commerce:

“AI agents are autonomous or semiautonomous software programs that use artificial intelligence to act on a user’s behalf. What they do goes far beyond surfacing search results—they decide what to buy, when to buy it, and where. They can follow instructions, weigh trade-offs, and transact. In other words, they don’t assist with the purchase—they actually are the purchaser.”

Bolding is mine. You may disagree with this definition, but if so, take it up with Stripe! It is pretty much consensus that AC is E2E. Visa & MasterCard think so, which is why they are focusing so much on setting standards for card use in Agentic transactions.

Stripe also points out that AC goes “far beyond surfacing search results”, so just helping with discovery may be valuable, but it isn’t AC.

I am not just splitting hairs here. Without the payment, many of the implementation challenges of AC go away as do some of the proposed benefits. In fact, when Shopify imposed its rule that AC transactions must bring the consumer back to hit the final purchase button, it technically ends AC entirely within its platform.

Autonomous: Agentic search needs consumer input

Think about initiating an Agentic shopping search. First, you need to write a prompt. I used an example of winter boots in my post. First you describe the basic data on size, price range, availability and maybe color. Those are easy.

I needed to go further: I wanted pull-on boots, not boots with laces; I wanted warm boots but not for arctic conditions (it only gets so cold in the NYC suburbs); They had to be waterproof; I didn’t know I needed a “gaiter” to keep rain or slush from dripping down the tops until I looked at a bunch of boots. I wanted mid-calf when some boots are much lower or much higher. I didn’t want them to look too “clod-hoppery”, which many did. These were all real considerations. I also wanted them highly rated, from authoritative sources, because I knew nothing about winter boots.

Most of these criteria I could now specify in a clear prompt for an Agent, but only because I did all the research before. Had I given the agent an early prompt I would never have learned about the gaiter issue and I wouldn’t know to specify temperature range. I might not have specified the height without seeing the wide variety that fit my other criteria. I don’t know how an Agent would interpret “not too clod-hoppery” – maybe it would understand, maybe not.

The Agent could have run a dialogue with me, showing results for an initial prompt and then narrowing it down as I rated my options. The Walmart/ChatGPT alliance seems to work that way. That is pretty much how I arrived at an answer boots search. So perhaps it would have gotten me there a faster, but it could not be autonomous because even I didn’t really know what I wanted until seeing what I could get.

Best deals: Agents may not have the data access they need

Many marketplaces already have “filters” to help narrow down choices. I used Wayfair’s extensively when I bought my home office desk. Those filters helped narrow choices from hundreds to a few dozen – based mostly on dimensions and price. After that it came down to taste. But how would an agent know my taste in desks? I didn’t even know it – and there is no precedent to guess from. I only bought one other desk in my prior life. It would have to show me a bunch and have me swipe left or right.

It isn’t clear that third-party agents would even have access to the data they need to neutrally advise users. Most big platforms I checked block AI scraping. Further, these platforms, like Wayfair or Zappos, will deploy their own agents to help customers narrow down choices before buying. These platforms have all the right data and every incentive to reduce returns. But their agents will be proprietary – reinforcing incumbent advantages. That is what Walmart has done.

After I wrote my post, both Etsy & Shopify cut deals with OpenAI and Stripe to allow agents on site. For Etsy this makes sense as many of its items are one-of-a-kind artisanal goods – competition for these is never exact. For Shopify, small merchants have trouble being discovered and may welcome Agent search, but the rules seem to limit what AIs can take and use for training.

But why would Amazon or Wayfair allow this? Instead, they might deploy on-site AIs that use their proprietary data to drive advice – and block that data from use by third-party agents. That is essentially what Walmart just did with Chat GPT. Open AI can access Walmart data for a specific transaction, but that data cannot be used for general training or for any non-Walmart commerce.

Without big merchant data it would be hard for 3^rd-party agents to deliver the best deals. Today, most major merchants block AI scraping for these reasons.

So even in Discovery, agents are helpful, but more of a “sustaining” innovation than a “disruptive” innovation. They reinforce incumbent retailer advantages rather than leveling the playing field.

Conclusion

Having seen “Contextual Commerce” and “IOT Commerce” go through a hype cycle without leaving a trace, I may be cynical. In one of my research searches, I saw the term “Conversational Commerce” to describe the Walmart/Chat GPT implementation. That term is more accurate than Agentic Commerce for what is really going on.

I may be entirely wrong on this. Maybe the big retailers open up their data stores. Maybe Agents figure out a way to measure taste. Maybe “conversational” prompts solve the specificity problem without imposing too much friction. Maybe merchants, like Shopify, eventually allow agents to make autonomous purchases. Lot’s of maybes.

Going out on a limb, I suspect most of us will not be using an E2E, non-proprietary, autonomous agents to buy stuff for the 2026 holiday season. You will be able to say “told you so” in January 2027 if my instincts are wrong. I would be happy to publish opposing views now if you want to submit them. But, no hype please, just reasoned arguments.