josh whiteDo We Even Need Shopify Anymore? Selling to AI Agents
On the night of October 9 I posted a little exercise on X. Make a list of all your software subscriptions, Shopify apps and platforms, write down which features you actually use in each one, then send it to Grok Bot and ask if it can build a spec that covers the same features. Right after that I asked "Do we even need Shopify anymore?"
Someone replied that you could even replace Shopify itself, and I listed out what Shopify really is (access to apps, secure handling of private customer data, theme editing, payment processing) and called it toast. I was being a little provocative.
Four days later, on October 13, Shopify shipped something that made me think about it a lot more carefully.
what actually launched
Shopify called it Shop MCP. It lets an agent or app search Shopify Catalog, build carts and complete Shop Pay checkout through a single connection, and Shopify says the catalog has over 1 billion products in it. Search doesn't even need an API key. When the shopper is signed in with Shop, the agent can use their saved addresses and Shop Pay, and it completes the purchase after the shopper confirms the total and the payment method. Those orders land in the merchant's Shopify admin like any other order.
A week before that, Sierra announced the Personal Agent Protocol, an open standard it says it's developing with Meta and industry partners including Shopify, Stripe and Walmart for how personal agents deal with businesses. The idea is that the consumer decides what access their agent gets, and the company sets the limits on what the agent can do through its website, its APIs or its own customer service agent. It's built on OAuth. Sierra published a v0.1 draft of the spec on October 9 and says it can still change in ways that aren't backward compatible before 1.0, so this one is early.
so do we need it?
For a $1-10M wellness brand right now, my honest answer is yes, and maybe more than before. If agents are going to shop on behalf of your customer, Shop MCP makes the stores in Shopify Catalog with Shop Pay behind them some of the easiest for an agent to check out from. Moving off Shopify to save on fees right as that's happening seems like a bad trade to me.
Where I think my post still holds up is the app stack sitting around Shopify. A lot of brands are paying for a long list of apps and tools and using a small slice of each one. That's what I'd run through the spec exercise first, because some of those features are things an agent can now handle for you, or build for you, for a lot less money.
write for a reader that can't see your hero video
An agent shopping for your customer reads product data and policies. It isn't going to watch your hero video or scroll your founder story, so the boring stuff starts to matter a lot more.
Product titles and descriptions should say plainly what the product is, what's in it, the serving size, how many servings and the price per serving. If a human needs three clicks and a zoom on the supplement facts panel to figure that out, I'd assume an agent is going to struggle with it too.
Inventory has to be right. An agent building a cart for somebody isn't going to wait around on a backorder, it'll move on to the next store that has it.
Returns, subscriptions and shipping policies need to be written clearly enough that a machine can answer a question about them. Sierra literally uses checking a returns policy as an example of what an agent does before the shopper even signs in.
Then there's claims, which is the one I'd worry about most in supplements. You already live under structure/function rules, and agents are going to summarize whatever is on your page. If your language is loose, the summary probably will be too, and it's your brand that takes the hit when it is.
attribution is about to get weird
One thing nobody seems to have figured out yet is attribution. Shop MCP orders show up in admin like any other order, which is great for ops and not so great for knowing where that customer came from. There are also no webhooks at launch, so agents have to keep checking on checkout status themselves.
If agent-driven orders start showing up as new customers with no ad touch behind them, your blended nCAC (new-customer marketing spend divided by new customers) is going to look better than paid really is. You could end up adding spend because the blended number looks great, or cutting a channel that was doing more than you thought. I haven't seen yet how these orders get labeled in admin, so I'd check your sales channel and order source reports now while the volume is tiny, and get a baseline before it matters.
Same goes for payback, which is nCAC divided by monthly contribution margin per customer. Somebody whose agent reorders for them every month could be your best cohort or your worst one, and you won't know unless you can actually separate them out.
what I'd do this quarter
I'd run the spec exercise on the app stack and leave the platform alone for now.
Have whoever is technical on your team install Shopify's `shop` CLI and search for your own category. Shopify says each result shows the store, the price and whether it's in stock, so you'll see roughly what an agent sees when it goes shopping for a magnesium or a greens powder, and whether you show up.
Rewrite your top five PDPs as if the reader is a very literal machine that can't see images, and get the claims reviewed while you're in there.
And keep an eye on the Personal Agent Protocol as the draft spec moves toward 1.0. If Shopify, Stripe, Meta and Walmart are all at the table, it's probably going to end up mattering for a small brand whether you pay attention to it or not.
If you want help getting agents set up for your own brand, whether that's running the spec exercise on your stack or connecting Grok Bot to your Shopify data, book a call at joshwhite.co/book-a-call and we'll walk through what makes sense for you.
ai setup sessions
Want help setting up your first agent?
I run paid AI setup sessions for brands at this stage. We pick the first agent, connect it to your store and books data, and figure out who on the team owns it. A free call is the easiest way to see where it makes sense to start.
Not ready for a call? The free tools are a good place to start.