230: A Shoe Company And A Cookie Company Now Say The Exact Same Thing To AI Agents

On August 5 Shopify switched on a second way into every store on its platform, built for AI agents instead of people. Nothing to install, no merchant asked. I opened three unrelated stores, Allbirds, Brooklinen and Partake Foods, and asked each what it could do for a machine. All three answered with the same roughly 800 words, identical character for character, and no merchant wrote a syllable of it. What those words say, and what it costs a brand when the visitor has no eyes.

Chapters

• 00:40 Thirty years of every shop trying to sound different
• 02:13 What Shopify switched on, overnight, on every store
• 03:19 Three stores, one identical answer
• 05:06 The tools are stage direction for a machine
• 08:13 Nobody is using any of it yet
• 13:45 Your voice does not transmit, your data does

Key Numbers

• Three unrelated stores returned the same ~800 words of tool text, identical character for character
• The adapter file is version 0.1.0
• Etsy put AI agent platform traffic under 1% of its total in Q2 earnings
• Shopify reports AI-referred orders tripled, which is referred traffic, not agents buying
• Shopify reports over a million merchants on the platform

Three Takeaways

1. Shopify wrote what your store says to machines, and every other store says it too. These are instructions, not descriptions. The checkout tool tells the agent to "follow it." Another tells it not to ask the shopper about missing options when it decides they only want to look. Someone chose when the customer gets consulted, and it was not the customer or the merchant.
2. This is the right way to build it, which is separate from whether anyone noticed. The tools read the same database as the storefront people see, so the two cannot drift apart. One good default beats a million bad ones. It is still worth knowing a default was set for you.
3. If the machine is the visitor, your voice does not transmit and your data is what is left. The photography, the badges, the reviews, the copy someone agonised over: none of it survives the call. Back comes a title, a price, a size run, availability. What is left to compete on is whether your prices are right and your stock is accurate. That work pays off whether or not the agents arrive.

Mentioned

• My WebMCP reference guide: https://nohacks.co/blog/what-is-webmcp
• What I found on day one: https://nohacks.co/blog/shopify-gave-every-store-an-agent-api
• Shopify's docs for building the buyers: https://shopify.dev/docs/agents
• Talia Wolf and the Emotional Targeting Framework
• Episode 229, llms.txt against 137,000 domains

Newsletter, one a week: https://nohacks.co/subscribe
Say hello: hi@nohacks.co

No Hacks runs no sponsorships and is funded by advisory and audit work. 

If your website needs to work for machines as well as people, start with a fixed-scope Machine-First Architecture audit: https://nohacks.co/audit

Transcript

[00:00:00] Hey, before I start, I wanna say something. I'm thankful for all of you reaching out after my latest episode, and yeah, it feels good to get feedback. Recording this it's lonely. I'm sitting in a corner of my living room at ten fifty-seven PM on Tuesday recording this, so I can release the episode on Wednesday.

[00:00:20] Every time you reach out , and tell me that it resonated, every time, just send me a message at hi@nohacks.co. It means the world. So thank you for that. Now, this week's episode, it's interesting. Something interesting happened last week. Shopify did something super interesting, and I'll tell you all about it.

[00:00:40] It has to do with WebMCP. I like WebMCP as a concept. I think it's something that has potential, has a future, unlike LLMs.txt and markdown files the reason I think what Shopify did last week, and I'll tell you all about it, but the reason I think it's important is because for the last, how long have we had e-commerce for?

[00:01:01] Thirty years? For the last 30 years, every shop online has been trying to sound different from every other shop online. They all had brand guidelines, copywriters that they hired or paid, agencies charging a fortune to make a betting company sound different than a shoe company, .

[00:01:21] So last week, Shopify announced something, and I tested it this week. I wanted to see what these Shopify stores can do for a machine. So th-there's a very simple way to do that. I'll explain wh-how it goes. But the funny thing is that all three shops that I asked, and I got the list of shops from Claude.

[00:01:40] I asked it for three random, Shopify stores. All three of them responded with the same answer, the same roughly 800 words. Not similar or anything like that. No completely identical. And none of those merchants wrote anything about that answer. It just came straight from Shopify. It's fully identical.

[00:02:02] So let's look at that. Let's look at why that matters and then what the future of this thing is. This is NoHacks with me, Slobodan Manic, where we talk about the agentic web with no hype.

[00:02:13] Here's what I've been up to and what I was poking at. On August 5th, Shopify switched on a second way into every store on the platform that's using their liquid templates, whatever they call it. So it's not a redesign that we see. It's a completely separate backdoor, side door, whatever you wanna call it, built for AI agents instead of people.

[00:02:36] There's nothing to install It's live on every page. Nobody's asked about it. The page now hands your browser a list of things an agent can do with the store. It's a list of real functions with names like Search the Catalog or Go to Checkout, and then an assistant sitting in that browser, and that could be Gemini in Chrome, that could be Perplexity Comet, that could be OpenAI Atlas.

[00:03:01] Oops, they killed it, so it cannot be OpenAI Atlas. I'm sorry. Sam Altman has a bad habit of killing products that generate a lot of hype when they're launched. Different story for a different podcast. Listen to Better Off if you wanna more-- learn more about that. So now an assistant sitting in that browser can call these functions directly.

[00:03:19] Not, not Atlas, but all those other browsers can call those functions directly instead of just going through the mess that HTML is on most websites. So I went and I looked at the three websites, and the websites are Allbirds, the eco wool shoes company that somehow became an AI data center company or wanted to become...

[00:03:41] it's a weird story, and I recommend you Google it if you haven't read about it. Brooklinen, and they sell bedsheets and partake foods. It's a small company. They make, uh, allergy-friendly cookies, something like that. So a giant shop, a, a mid-sized shop, and a small shop with nothing connecting them except the fact that they all run on Shopify.

[00:04:03] I opened each one of these in a Chrome tab. There's nothing special switched on. I just asked the page what it had registered on it, every time, the same answer. I put the three lists side by side. The descriptions of the tools are identical. Not similar completely identical.

[00:04:20] And, uh, yeah, that's what a shoe company, a bedding company, and a cookie company have in common. They, they just give the same interface to every user that looks for that kind of interface. And look, the problem here is that a company like Allbirds, they spent years learning how to sound like Allbirds. They spent years figuring out what their messaging is, what their emotional targeting is and who their customer base is. But on this side of the website, the webMCP side, the one for AI agent, they-- agents, they sound exactly like a company that sells cookies And identical is strange But, when you look at those tools and what they say, it gets even stranger.

[00:05:06] So this is checkout tool, word for word, the description for the checkout tool. It says, "Verifies the cart is not empty and moves the browser to the checkout page. The tool result then confirms the destination and instructs you not to navigate again. Follow it." This is description of how something will happen to an agent.

[00:05:25] This is direction basically written for a machine in, in, in capital letters in it and all of it. So it just literally says, "Follow it." And then there's another one. There's a tool that shows a product in a particular size or color. It says, "Do not ask the user for missing options when their intent is to view."

[00:05:44] That's the description of this tool. So somebody had decided when the shopper gets asked a question, this was not the shopper, this was not the merchant, this was, uh, whatever it was. So who is this script for? What is this about here? There's a whole section on Shopify's documentation at shopify.dev/docs/agents.

[00:06:07] It's not addressing the people building stores. It's addressing the people building the buyers. There's a whole section of their documentation that's really not... It's weird when you think about it. It's not teaching the people with Shopify stores how to run a store and how to do whatever. It's not teaching Shopify customers anything about Shopify.

[00:06:34] It's teaching, it's teaching the buyers how to shop. And this is something we haven't, like-- Imagine if this was ten years ago, and if Shopify had a series of tutorials for people shopping online, "This is how you click a checkout button. This is how you do this. This is how you do..." It would be ridiculous. Shopify is essentially trying to - control both sides of the marketplace. They're helping you sell and they're helping other people buy. And I'm, I'm not painting them as a villain. This is not the story that should be used to paint Shopify as a villain. If you want that, just go and read about what their CEO said about voting rights and wealthy people.

[00:07:20] Just if you need a Shopify villain story. This is a sensible thing for Shopify to do because if they teach... This is what they have to do, I think. If they teach people how to build these agents that's the only way people are going to properly switch to agentic commerce and then all that. So these tools, they, they read the same database as the storefront that we can see, that you can see.

[00:07:46] So it's a good thing that Shopify is the one that is controlling the WebMCP tools. I, I'm not-- Yes, of course. The fact that Allbirds sounds the same way as Partake Foods. The fact that they sound the same is-- sounds weird when you say it like that, but the fact that it's Shopify doing this and, and controlling the output of this means that it's not going to get too messed up, especially in the early days.

[00:08:13] I, I think this is really important. It's directionally correct. It's a good thing that this is happening. But also the question is what exactly is happening? And that's the second side of this story. ' Cause all this, Shopify building the agentic infrastructure for agents to shop and also helping people build the agents that will shop, None of this counts until someone or something starts using it. You heard me talk about llms.txt last week. It doesn't matter. It doesn't matter if it sounds like a good idea, if it sounds directionally correct. This is my favorite phrase when I talk about AI, by the way, because AI will always sound directionally correct.

[00:09:02] It will always sound like it makes sense if you don't wanna scratch the surface. Now, this sounds directionally correct more than llms.txt. I think WebMCP, and I said this already in this episode, I think it has potential. But who's calling these tools?

[00:09:18] All the platforms that reported their second quarter earnings, they all said that agent volume is low. Etsy said it was under one percent. Shopify says its AI-referred orders tripled. That's their number from their company, and they have a lot to gain for it. But this is AI referred. This is not agentic.

[00:09:37] This is not the same thing. This is people arriving from an AI answer, then they're buying with their own hands. Not the same. Yes, also influenced by AI, that visitor, but it's not AI buying and, and, and a lot of people are confusing those. AI referred is not the same as agentic and AI agent doing something on the website.

[00:09:56] AI referred is just a primed user that with a lot of knowledge about your brand or more than just using Google search or social. They're two different visitors. The one is a person, the other one is not. It's the one without a person and, and those users where there's no person, where it's just an agent completing things, the ones that should be using this WebMCP infrastructure and, and all of the shopify.dev/docs/agents tutorials, they're not there yet.

[00:10:26] It's very early. It-- This is barely even finished.

[00:10:29] This is version zero one zero. So we have a million of these storefronts and or more. How many Shopify stores are there? I haven't checked. I should have checked. They're saying the same thing to an audience that is just me testing. I can hear the crickets. There's nobody else seriously doing this. I- if I had to bet, I don't, but if I had to, I would say this will happen in the next twelve to eighteen months, if not sooner. If we are going to have AI agents browsing the web and using the web and using the websites and buying, and there's a lot of money being poured into that happening in there. Just the last week there have been so many announcements in this space. A lot of companies are trying to get us into this and trying to help us use AI agents to do things for us online and I hear it, but I love to shop online.

[00:11:24] I love to compare prices. So, so go ahead and do it. That's fine. There are things that nobody likes to do online, filling out forms, research, whatev-whatever it is. There are many things where agentic browsing and agentic, uh, shopping even will be handy. The other reason why I think, and this has that advantage over LLMs.txt, is for those visitors, for non-human visitors, AI agents trying to understand a webpage, trying to do something at a webpage, this is the cheaper way.

[00:11:58] WebMCP is a significantly cheaper way, and I wrote about this. If you, if you just Google... i'll put it in the description, but just Google No Hacks what is WebMCP. I have a very long reference guide that I just updated a few days ago. So you wanna learn about WebMCP and, and what it is, what it can do for you, how it changed over time since it was announced earlier this year, just you can find that guide and read it there.

[00:12:21] And again, WebMCP has an advantage compared to HTML Because you can still do things. It-it's kind of like an API that happens in the browser. So let's say, let's say this starts happening. Let's say this is adopted. Let's say, I don't know. Let's say in, in six months we have actual orders coming through this way.

[00:12:48] And this is not some AI agents that represent... No, this is me deciding what my next running shoes are and then just having an agent and I say, "Hey, this is where I live. This is my budget. This is my size. These are the shoes.

[00:13:08] I want them delivered in the next seven days. Just go and buy it." And maybe I can say, "Before you click that final button, come back to me and check with me." Maybe I can build that in, or maybe I don't. But that's how agent commerce will work for me. I'm not saying that's how it's going to work for everyone,.

[00:13:24] But something like this with a limited wallet I think that can work So what, what changes for all those websites selling shoes here in Portugal, what changes for them if people start buying this way? Your voice that you have on your website, your messaging, your emotional targeting, this is the second time I'm mentioning emotional targeting.

[00:13:45] Shout out to my friend Talia Wolf, who is behind Emotional Targeting Framework and, and wrote a book about it literally. So everyone's spending money on these stores. There's always, there's an eye-- there, there are eyes looking at the page.

[00:13:58] There's the photography, there's the badges, the reviews, the copy, someone spent time over, someone was fired over, whatever. If the machine is the one navigating the website, just looking for the product, it's cold-blooded. It just needs to get the information it needs. Everything else is fluff, and the information it needs is the whole product.

[00:14:19] It needs to be accurate. It needs to be clear

[00:14:22] So if, if you ask the agent for a product using WebMCP, this new door that Shopify has opened on a lot of the Shopify stores, it doesn't need personality it needs accurate product information. It needs clear product information, and, and that's about it. When I asked Allbirds about one of their products using one of these functions that WebMCP has, that's what I got. Just got a, a structured JSON file, essentially.

[00:14:54] And that changes what you as a brand trying to sell on, on- online, exist online, can compete on. Because if your voice does not transmit through this, if it's lost, what is left is whether your prices are right, whether your stock is accurate, your catalog complete, whether your policies are parsable. You know, all the boring, the unglamorous That's the entire game In this reality.

[00:15:24] There's another reality which is still browsers and visuals and humans of course, but there's-- now we are getting this second one.

[00:15:31] yeah, This is boring work. I think that's the beauty of it, the fact that it's boring you just have to look at this, look at your server logs, and then just ignore the keynotes, announcements, all the stuff. Just focus on the very, very boring work. So in closing, look More than a million storefronts learned the same commands

[00:15:54] and none of them have chosen what they're going to say. The other side of this is that the room is still absolutely empty. Nobody is using these to shop. This is version one zero one. But when that room starts filling up,

[00:16:09] It's good to be ready and to have accurate information to serve these non-human users ' Cause this is a new reality. This is a new door for your website where your voice, your tone, all the stuff you invested in trying to convince humans to buy from you, it's gone. And the data and that structured product feed, that's all you have left and all that's left from you

[00:16:31] For 30 years, websites were something people looked at There were people who couldn't, and, well, we mostly didn't care about them, and we didn't build the websites that way. That era, the visual era , is ending, so it's good to start preparing

[00:16:53] If you made it this far, this is episode two hundred and thirty, and if you wanna subscribe to the newsletter, I send one every Wednesday, no hacks dot co slash subscribe.

[00:17:08] Thank you very much. I'll talk to you next week