Last night’s announcement from OpenAI, Shopify and Stripe was designed to make jaws drop. A neat demo of someone asking ChatGPT about a product, getting a recommendation, and completing the purchase without ever leaving the conversation. It was smooth and fast, even I was impressed.

And then came the predictable LinkedIn headlines: “The death of online stores.” “The future of shopping.” “The end of checkout as we know it.” bla bla bla, endless ai-generated slop and click bait.

But as anyone who has ever worked in the detail of an ecommerce integration knows, a slick demo is one thing. Making it work in the real world is another.

The ACP documentation makes it sound so simple: provide a product feed, expose a checkout API, accept a delegated payment token, and voilà. In reality, orders don’t fall into a merchant’s systems that easily. Each merchant or platform has its own quirks, its own integration points, and often its own legacy baggage. It feels like there is a long way to go and a lot of work to do still.

Take product data. The spec assumes merchants can serve up clean, structured feeds that refresh every 15 minutes. But anyone who has worked with feeds knows the bigger challenge isn’t generating them, it’s making sure the data inside them is accurate and consistent. Prices need to match what’s in ERP. Stock levels need to be reliable across warehouses. Attributes and imagery need to line up so the product shown is the product shipped. Feeds exist everywhere, but trust in their integrity is often shaky and that’s not a problem solved by a protocol spec. It’s one thing having a feed for ads or listings, it’s different when they are required for placing orders.

Payments? The delegated token model is neat, I like it a lot, I wrote about it here, but it doesn’t solve SCA in Europe, or a merchant’s risk team, or the external fraud provider that needs to score the order before capture.

Fulfilment? The spec says merchants should push shipping updates back to ACP. That’s fine if you’ve built webhooks into your OMS. If not, you’re in for more work. Split shipments, backorders, partial refund, none of it is in the demo, all of it is in the real world.

Don’t get me wrong, we’re all-in on this future. But the current conversation oversimplifies what it takes to make it operational. The death of ecommerce platforms, as usual, has been exaggerated; they’re still where all the heavy lifting gets done.

Agentic Commerce will have a place, I don’t doubt, but I personally don’t think it will replace all the other channels. It certainly won’t be plug-and-play. It will require infrastructure, connectors, and middleware to bridge the gap between protocols and the real world. It is certainly an amazing new distribution layer. And like every other channel before it, social, search, marketplaces, it will only work if someone builds the pipes to connect the shiny surface to the messy merchant backend.

Where Shopthru Comes In

This is where we see Shopthru’s role. We live in that messy middle, the plumbing infrastructure. We build the middleware that takes protocols like ACP and makes them practical – normalising feeds, standardising order ingestion, translating OMS events back to where they need to go. The unglamorous but essential stuff. At Shopthru, we’ve been building exactly the stuff ACP glosses over:

• Middleware for order ingestion: we standardise order creation across Magento, Shopify, WooCommerce, custom stacks.
• Feed normalisation: we pull product data from messy ERPs, PIMs, ecommerce systems and turn it into clean, structured feeds.
• Attribution guardrails: we protect publishers and content creators from losing credit in AI-driven transactions.
• Transactions: we facilitate the payment through the merchant’s payment processing, preserving MoR and their fraud and security rules.
• Analytics: we give retailers visibility into how embedded/AI-driven orders perform vs. traditional funnels.