The AI commerce revolution has already begun. Amazon, Shopify and other e-commerce platforms are now using AI recommendations to trigger autonomous workflows. Yet one question hasn’t been answered: who is going to control the business context, permissions and execution layer through which those AI agents operate?

I've been watching the e-commerce evolution unfold from inside the industry for more than two decades. I started during the osCommerce and Magento era, worked with the early Shopify ecosystem in 2014 and eventually sold Aheadworks in 2019. Every wave of innovation changed not only the technology merchants used, but also what they owned, controlled and delegated to software. However, one principle remained the same: the better you understand your customer, the more efficiently you can serve them, the more sales you will generate.

A century ago, the owner of the small corner shop knew everything about their customers. They knew what each person usually bought and what they were likely to need next. In modern terms, they kept customer data in their memory and recommended relevant products based on it.

In the 20th century and at the start of the 21st century, online retail mirrored the change that had happened in the physical world: small “corner shops” were replaced with large supermarkets optimized for an average buyer, yet lacking true personalization. They tried to get it back through loyalty programs and shopping recommendations, but they never really replaced that personal connection between a merchant and customer.

For example, if you purchased a tent in June, traditional marketplaces would recommend you either another tent or mosquito repellent in July, without fully understanding what products would be relevant to you at that moment and whether your camping trip, let’s say, to Finland was over or not yet.

Do Today’s E-Commerce Sellers Really Own Their Data?

While the majority would expect the answer to be yes, the reality is more complicated. Digital marketplaces like Amazon often know far more about the average buyer than brands selling their products there. Today, Amazon has over

For many merchants, Amazon has become not just another sales channel, but the foundation of their business. It gives even the smallest brands access to a global customer base, convenient logistics and an infrastructure that would be too expensive to build independently. On the other hand, it has become one of the largest repositories of commerce data third-party AI platforms may use to train their models and build the next generation of commerce.

The Data Ownership Problem

Amazon allows sellers to compete, bidding against each other for the platform’s internal traffic using Amazon Ads. However, even if a seller pays to drive a user to their own product listing, the marketplace keeps control over the customer identity, browsing behavior, purchase history and other data that would allow brands to engage with their customers in the long term and understand what they would likely need next, unlike the owner of that “small corner store.”

This creates a new opportunity for independent systems that can combine data a merchant permits them to access, catalog performance, public market insights and goals set by a seller, turning that context into controlled operations.

Three Generations of E-Commerce Control

The first generation of e-commerce was built around software owned and controlled by the seller. Online stores operated on platforms such as osCommerce and Magento, often on the merchant’s own server. Even though businesses kept control, they also had to deal with technical complexity.

The second generation moved e-commerce into cloud platforms and marketplaces. Shopify, Amazon and similar platforms made online selling faster, cheaper and more accessible, but merchants delegated their software, distribution and business context to platforms.

The third generation will be agentic. It can restore operational agency, allowing merchants to use their authorized data, set their own objectives and let AI systems act within permissions they control.

The Urgent Need For a Safe Execution Layer

In 1995, when a customer bought books on the Amazon website, Jeff Bezos hand-packed them in his garage in Bellevue, Washington, then drove to the post office and sent the parcel himself. One customer in Bulgaria, unable to pay by credit card, folded two $100 bills, tucked them into a floppy disk and mailed it to Amazon. Every step was manual, and there was an obvious need for massive automation.

Today, we see a similar automation gap when it comes to the use of AI in sellers’ daily operations. By May 2025, more than

Amazon now has around

230,000 monthly Seller Assistant users, withsellers accepting its recommendations more than 90% of the time , which is pretty powerful validation of agentic seller workflows. But the question is if 90% of sellers use the same marketplace-native AI optimization playbook, are they really gaining a competitive advantage or losing the chance to stand out? This creates the potential for independent AI commerce tools.

The Future of Agentic Commerce

For customers, sellers and large e-commerce platforms, AI agents could recreate the experience of the small corner shop. Buyer- and seller-focused agents are technically different, but they shouldn’t be considered as opposing teams. While shopper agents could help people find products they’re looking for, seller agents could make the most relevant offers within certain constraints, such as availability, margin and brand positioning.

Sustainable AI commerce starts when both the seller and the customer benefit. It creates the foundation of the next generation of personalized shopping experience, where AI is involved at every touchpoint of the customer journey, providing market insights and acting on them in real time. To make that happen, modern e-commerce needs the right execution layer, permission architecture and continuous feedback loop.

Most e-commerce products still look like SaaS tools with dashboards just because this is the interface merchants used to. I believe the more important architectural direction is agent-first, API-first and data-first. The interface may remain Gemini, Claude, ChatGPT or a custom commerce agent, but the underlying layer should provide trusted context, guardrails and execution capabilities. This is the direction we’re building toward.

The Race to Build The Next Generation of E-Commerce

History shows that during a gold rush, the biggest value appears to those providing the infrastructure that allows others to participate. In AI commerce, that infrastructure is not another chat interface, but the control layer that gives agents business context, permissions, auditability and a safe path from AI recommendation to execution.

I’ve noticed that some startups are exploring the idea of building a dynamic online storefront for each individual visitor. To optimize product placements for online retailers, they are trying to analyze the external and internal third-party data, including the information about which channel, campaign, CTA or offer brought a customer in.

In my opinion, the next generation of e-commerce won’t be about creating a storefront customized to every visitor. What if the customer’s purchase intent is so strong that the main priority is not to provide more relevant information, but to remove any possible friction before they change their mind? What if a storefront shouldn’t be the starting point at all? A customer could start their shopping journey in the AI interface, while the merchant retains the checkout, brand surface and direct customer relationship.

Another group of AI platforms is competing to help brands automate commerce operations. Amazon Seller Assistant can reason and take actions with the seller's permission. CommerceIQ deploys agents in sales, digital shelf, content and retail media. Pacvue offers execution with guardrails and audit trails. And Triple Whale’s Moby executes actions in connected commerce and marketing systems. Among customer-focused tools, Shopify Agentic Storefronts connects merchants’ catalogs to AI shopping channels.

The AI commerce race is already very active, but very fragmented. Most solutions remain tied to a specific platform, customer segment or functional workflow.

The missing piece is broad access to an independent, seller-controlled operations layer that can turn AI recommendations into real actions in a live store with certain permissions, configurable approvals, audit trails and catalog-wide optimization.

The modern seller needs an AI operations layer that works from their own authorized context, such as catalog data, listing performance, business goals, brand rules and relevant public market insights. Its purpose is not only to generate more content, but to identify operational problems, prioritize actions, prepare changes and execute them under controls defined by merchants.