Commerce Is Moving to Agents. But Who Controls the Rails?
Shopify just drew a line. Agentic commerce is coming. But only on its terms.
AI agents that search, decide, and buy are no longer a far-off concept. With tools like ChatGPT, Perplexity, and Gemini AI Overviews already part of daily workflows, users would soon start delegating their shopping decisions to AI. And the first battleground is Shopify, as it powers around 30% of e-commerce stores in the US.1
Last week, Shopify quietly updated the robots.txt file across all storefronts.2 For context, this file tells web crawlers (like Googlebot, Bingbot, or AI agents) which parts of a site they are allowed or disallowed from accessing and indexing. The new directives include limitations on automated scraping and specifically discourage buy-for-me agents from executing full checkouts autonomously, without human involvement. At first glance, this may appear to be a rejection of autonomous agents. But banning AI outright is neither enforceable nor future-proof, especially as agentic workflows already grow rapidly across sectors.
Our take is that Shopify doesn’t want agents scraping storefronts or mimicking human clicks. It likely wants them to use officially sanctioned tools. That’s where the Checkout Kit comes in: a set of APIs and SDKs that give structured, controlled access to search, cart, and checkout flows.3
A case in point is Shopify’s recent partnerships with OpenAI and Perplexity for enabling seamless commerce experiences in chat native interfaces. 4 5
The message to developers is simple: stop crawling, start integrating. Soon, agents won’t need to scrape product pages. They’ll tap into official endpoints: search, filter, vet, and checkout, all within Shopify’s ecosystem. And with the right guardrails, even end-to-end AI-led transactions may be just around the corner.
The familiar platform playbook
This move fits into a broader pattern. Platforms have long opened up early, captured developer energy, and then pulled down the curtains once they gained momentum. For example, Facebook gave developers viral channels and then throttled reach. Similarly, Google fueled the SEO ecosystem and then buried the organic visibility under ads.
Now Shopify is doing the same, encouraging structured integrations via Checkout Kit, while quietly closing off unregulated flows. Today, it’s a recommendation. Tomorrow, it could be monetized once the infrastructure has scaled beyond the inflection point.
And that matters. Because agentic commerce today looks a lot like e-commerce in the dotcom era - underestimated and experimental. For instance, Skeptics once said fashion wouldn’t work online. Too tactile. Too many returns. Then came free shipping, return policies, and AR try-ons. Further, they said electronics wouldn’t scale online. High consideration and complex purchases that involved a lot of research. Then came verified reviews and comparison layers.
AI agents will evolve the same way. Developers will solve trust. They’ll build explainability, human-in-the-loop flows, and post-purchase support. The tooling will catch up. What matters is who controls access when that shift happens.
Sellers are stuck, and they know it
Meanwhile, merchants are trapped between two expensive platforms:
Amazon, which charges 15% to 40% in listing and transaction fees
Google and Meta, where ad costs keep rising and RoAS keeps falling
Running your own storefront sounds like freedom, until you realize most of your revenue still flows through one of these toll booths. Either you give away margin to Amazon, or you burn cash to win traffic on paid ads.
This is the real bottleneck.
And it’s why agentic distribution, if done right, can create a third novel channel. One where merchants aren’t forced to buy attention or pay rent to sell.
CommerceMesh: open rails for agentic commerce
The CommerceMesh Protocol solves the structural problem. It gives sellers an open, programmable interface to expose product catalogs, inventory, pricing, trust layers, and checkout endpoints, in a format that AI agents can understand and act on.
It isn't a marketplace. There’s no central listing or gatekeeper. It’s just infrastructure like SMTP is for email, or HTTP is for websites. It allows any AI agent, across any platform, to query what’s in stock, what it costs, and how to buy it, without scraping or guessing.
Overway is building on top of CommerceMesh to make this future real and fast. Its Discovery Node lets sellers convert their existing catalog into a structured JSON endpoint that agents can access. From there, AI agents can:
Discover products across sellers
Compare real-time availability
Evaluate trust markers
Trigger native checkout flows
Sellers don’t need to change their storefront. Overway handles the conversion, exposure, and indexing, making any catalog discoverable by the agent economy.
This changes the equation. Sellers would no longer have to pick between Amazon’s high fees and Meta’s expensive clicks. They can now participate in a growing surface of agent-led demand, without giving up control or margin.
Additionally, this avoids expensive scraping both for the developers and the servers (Shopify in this case), as with CommerceMesh, we give them a structured product catalog that AI agents can query.
No crawling, just querying.6
Sources
Demand Sage. “Shopify Statistics 2025 – Users & Revenue.” Apr, 2025. Link
X.com. “Some of you might have noticed…” Tweet by Ilya Grigorik. Jul, 2025. Link
Shopify. “The World’s Best Checkout. Anywhere.” Accessed Jul 15, 2025. Link
Search Engine Journal. “OpenAI Quietly Adds Shopify As A Shopping Search Partner.” Jul, 2025. Link
OpenTools AI. “Shopify Boosts AI Shopping Experience with Perplexity Partnership” May, 2025. Link
Web Automation. “How much does web scraping cost - the ultimate guide.” Aug, 2022. Link
1 Original Post
Originally posted at https://www.overway.ai/post/commerce-is-moving-to-agents-but-who-controls-the-rails