Agentic Commerce: what happens when AI starts shopping for us
By 2030, between 10% and 20% of online purchases in the United States could be completed by AI agents, without a human ever touching the cart. These are Morgan Stanley estimates, and the value at stake ranges between 190 and 385 billion dollars. McKinsey raises the bar even further: for the US B2C market, the figure is around 1 trillion, with a global impact between 3 and 5 trillion. This is not science fiction. It is the present already taking shape.
It is called Agentic Commerce, and it is the deepest transformation e-commerce has faced since mobile. In this article, we explain what it is, why it is happening now, what it concretely means for those who manage an online shop and, above all, how to start preparing.
From browsing to buying: what Agentic Commerce really means
In traditional e-commerce, the journey is familiar: the user searches, clicks, compares, adds to cart, pays. Every step requires a deliberate human action.
In Agentic Commerce, this pattern breaks. The user provides instructions on what they want, with what budget, by when, and with which preferences, and an AI agent carries out the entire process on their behalf, searching for options across multiple sites, making a decision and then completing checkout. All this without the user returning to the product page.
In practice, you are no longer the one buying. A software entity buys for you.
This is not a future feature of some niche app. It is the direction in which the largest investments in the global tech sector are converging.
The investments driving the rise of Agentic Commerce
Amazon, Microsoft and Google are investing a combined total of over 650 billion dollars in AI infrastructure by 2026. Amazon alone has allocated around 200 billion to AWS data centers. Apple is moving in parallel: the evolution of Siri, strategic acquisitions, new agreements to improve the interaction between people and devices.
The reason is structural: AI requires enormous computing power, and those who build the infrastructure today will have a competitive advantage that will be difficult to close tomorrow. Agentic Commerce is not an isolated application; it is one of the natural consequences of this massive investment in computational capacity and advanced language models.
The real challenges for online merchants
For those who sell online, this scenario is not just an opportunity. It is also a series of new problems to solve, and some of them call established logics into question.
Traditional tracking loses effectiveness. If the purchase takes place inside an AI conversation, without cookies, without browsing, without traceable clicks, many of today’s analytics and attribution tools become blind. Who brought that sale? Which campaign contributed? The answers become much more complicated.
The discovery phase gets shorter. Today, a user may visit 5 websites before buying. An AI agent can do it in seconds, without ever generating a recognizable session. The merchant risks losing that window of direct contact with the potential customer that today feeds remarketing, nurturing and loyalty strategies.
AI behaviors do not follow human patterns. Fraud prevention systems are calibrated to recognize anomalous behaviors compared to a human norm. An AI agent that completes 50 purchases in one hour on behalf of 50 real users could be blocked or classified as suspicious. New parameters are needed.
Legal responsibilities are still unresolved. Who is responsible if an AI agent buys the wrong product? How is consent defined in an automated purchase? How does the payments industry handle a transaction without direct user action? Today, the answers do not exist, and this is a real risk for merchants.
Zero-party data becomes strategic. Without cookies and without tracked sessions, the only reliable information will be the one provided directly by the customer: preferences, history, priorities. Those who have built a zero-party data asset will have an advantage. Those who have not will start from scratch.
How e-commerce platforms are responding
The main platforms have already started to adapt. It is worth observing their direction, because it defines the technical perimeter within which merchants will have to operate.
Adobe Commerce is working on open protocols to make catalogs and inventories easily readable by AI: data structure, API exposure, compatibility with external agents
Shopware is developing solutions to translate agent requests into concrete actions, with direct connections to ERP and PIM systems while maintaining operational control over the fulfillment process.
Shopify, together with Google, is testing shopping experiences that can be completed directly inside AI chats, without the user ever having to leave the conversation to complete the order.
The common trajectory is clear: e-commerce must become readable and actionable by autonomous AI systems, not only by human beings.
What merchants can start doing today
There is no need to wait for Agentic Commerce to become mainstream before starting to move. Some actions have immediate value, improve the current user experience and, at the same time, build the foundation to remain competitive in this new scenario.
Take care of product data quality. An AI agent works from structured data: precise titles, complete descriptions, consistent attributes, well-mapped variants. Poor or inconsistent product pages make your catalog invisible or misinterpreted. This already applies today to SEO and advertising feeds, and it applies even more when the one reading your catalog is an AI that has to make decisions.
Make updated data available in real time. Stock availability, delivery times, prices: an AI agent needs accurate information at the moment it operates. Outdated or unsynchronized data leads to wrong decisions, and a purchase completed on a promise you cannot keep is both an operational and reputational problem.
Simplify the checkout process. Less friction, fewer steps, fewer technical barriers. A complex checkout is already a problem for human users: for an AI agent that has to complete the purchase programmatically, it can become an insurmountable block.
Start building your zero-party data asset. Preferences, purchase history, direct communications with customers: these data points will be the compass when traditional tracking is no longer enough.
Test integrations with AI agents. Early experiments bring asymmetric advantages: you learn earlier, make mistakes earlier, correct earlier. Waiting for the market to consolidate means giving ground to those who moved in advance.
Start by making your e-commerce AI-friendly
The operational summary is that the goal remains to build an optimal experience for the human user, but with an increasingly strong focus on making it understandable, navigable and actionable also for autonomous AI systems.
“AI-friendly” is not a marketing label: it is a technical and strategic choice that concerns data structure, exposed APIs, catalog quality and the simplicity of the purchasing process.
Those who start working on it now will have a concrete advantage when Agentic Commerce moves from emerging phenomenon to market standard. And judging by the numbers, that transition is closer than it seems.
