Marketing
Agentic AI in E-Commerce: When (and If) It Will Actually Arrive
Agentic AI is supposed to transform e-commerce - but are we really there yet? Despite bold announcements from the biggest players, reality looks different. We look at what's actually happening today, why AI still isn't buying on customers' behalf, and what the future of online shopping might look like.
For months now, one phrase has been circulating in e-commerce that sounds like the announcement of yet another revolution: Agentic AI.
In theory, it's a massive leap forward - artificial intelligence that doesn't just help a user find a product, but makes the decision and completes the purchase on their behalf.
It sounds like something that could completely change the way we shop online - and that's the angle we took back in July 2025, when it was still "fresh news": Agentic AI in e-commerce - a new era of autonomous selling. The problem is that, despite plenty of announcements and big statements from the largest tech players, real-world deployments are practically nowhere to be seen. Why is that? Were these just empty promises - castles in the air?
Prefer to listen? We cover this topic on our podcast, Biznes w Internecie (Business on the Internet). Below you'll find the episode dedicated to Agentic AI in e-commerce:
Does Agentic AI in e-commerce even exist?
Before we look ahead, it's worth getting our terms straight. Agentic AI is often presented in marketing language that says very little about its actual capabilities.
In operational terms, it can be defined as a system that understands the user's intent, makes a decision, carries out an action (e.g. a purchase), and takes on some form of responsibility for it.
That's an important distinction. In today's e-commerce, AI is already at work - recommending products, personalizing offers, supporting search. But it's still the user who makes the decision and completes the purchase.
Agentic AI would take on that responsibility itself. Not "suggesting," but "acting." And that's exactly the point where reality starts to diverge from expectations.
Everyone's talking about it, but... nobody's actually delivered it yet.
Recently, several major initiatives have emerged, all promising the arrival of Agentic AI in e-commerce. Shopify announced the development of AI agents within its ecosystem, Google is working on the Universal Commerce Protocol, and Mirakl is developing a project called Nexus. On top of that, there are OpenAI's experiments with in-chat shopping.
On paper, it looks like the start of a new era. In practice, though, it's hard to point to real deployments operating at any real scale, because beyond announcements of tests and waitlists to sign up for, there's no real "echo" in the media or on social media. There are no widely available case studies, no global rollouts, not even clear follow-ups after the big announcements.
This isn't a matter of lacking interest - it can instead be read as a signal that implementing Agentic AI in e-commerce is a much harder problem than initially assumed.
What happened to Instant Checkout - shopping in ChatGPT?
One of the more interesting experiments was OpenAI's approach, which tested Instant Checkout, the ability to make purchases directly inside the chat. Sellers could sign up on OpenAI's site and join the test.
This was the scenario that struck many people as the most natural, given how people default to using AI - the user chats with the AI, and the AI completes the purchase. Everything would happen inside the AI window.
After a few months of testing, though, the feature was pulled. The reason was simple: adoption was lower than expected, both among users and among sellers.
That's a very telling signal. It shows that even when the technology is available, how it's used has to match users' actual habits. And those habits don't necessarily point toward "conversation as the main shopping channel." On top of that, according to voices from the AI community, users were mainly using AI to research and compare products, but the actual purchase decision (when it happened) moved outside the AI and was still made manually by the user. That points to a lack of trust in AI at... the most critical moment. Users were happy to let AI handle actions that didn't directly "touch" them - but once it came to the point where AI would take the lead over their data and their money, users backed away.
Trust in AI for the entire purchase process remains a problem that needs a reliable, user-friendly solution. If anything, Instant Checkout may have shaped the broader Agentic AI landscape for e-commerce more than OpenAI and analysts assumed. That's borne out by the fact that the direction is now shifting toward integrating AI with existing systems and platforms.
What's actually happening right now with Agentic AI?
The most important shift isn't obvious at first glance. It's not about the user interface - it's about infrastructure.
Solutions like the Universal Commerce Protocol, being developed by Google, are trying to create a common language that lets AI systems communicate with stores. In practice, that means being able to pull information about products, availability, and prices, and to move through the purchase process.
That sounds like a foundation for Agentic AI, and it is exactly that - but the problem is that we're still at the stage of building that foundation. Deployments are limited, often experimental, and focused mainly on the US market. Globally, this ecosystem is only just starting to take shape.
At the same time, AI-driven traffic in e-commerce is growing (for example, Shopify points to an 11x year-over-year increase). More and more users are turning to AI tools to search for products and make decisions. That's still not Agentic AI, though. It's an intermediate stage - AI as an advisor, not an executor.
Why won't Agentic AI work at this stage?
At the conceptual level, everything seems simple. In practice, though, several fundamental problems come up.
The first is trust. If AI is going to buy on a user's behalf, the question of responsibility comes up. Who's responsible for a bad purchase? What happens with a return? How are disputes resolved? Without clear answers to these questions, it's hard to talk about widespread adoption.
The second problem is infrastructure. E-commerce is a complex ecosystem made up of different systems, integrations, and data sources. An AI agent needs access to up-to-date information on stock levels, prices, promotions, and logistics. Keeping that data consistent in real time is a massive challenge.
The third factor is AI decision-making itself. Models are capable of making decisions, but not always in a way that's predictable or matches what the user expects. Preferences can be unstable, and choices can be hard to justify from a business standpoint.
Finally, there's the question of user experience. Contrary to popular imagination, the market isn't heading toward a single "super-chat" that handles the entire purchase process. Instead, we're seeing entry points spread out - AI in search, AI in the store, AI in productivity tools.
What will this look like in practice?
The most realistic scenario doesn't resemble a vision of fully automated shopping.
It's more likely that Agentic AI will act as a layer on top of existing systems - e-commerce platforms, for example. It will support users in searching, comparing products, and partly in the purchase process, but won't necessarily take it over entirely.
This hypothetical Agentic AI use case can be pictured, to the disappointment of some, as a chatbot window, a store search box, or a landing page or dedicated subpage on a store containing a single window that covers the entire search, comparison, and purchase process (keeping in mind how checkout gets resolved, whether it requires an account or not).
The store will still remain the operational hub - that's where the data on products, prices, and logistics will live. AI will act more as a "layer" that makes it easier to navigate that system.
Instead of one breakthrough moment, we'll see a series of smaller changes. Each will take over a fragment of the purchase process, until at some point the whole thing looks completely different from how it looks today.
Summary
Agentic AI in e-commerce isn't a question of "if," but rather "how and when" - even though the news on this front doesn't exactly build a lot of confidence in the idea.
We won't see a single moment when AI starts buying on users' behalf. Instead, we'll witness a gradual evolution - from recommendations, through partial automation, to more advanced scenarios.
From a business perspective, though, the most important takeaway is different. The winner won't be whoever implements AI first. It will be whoever has organized data, well-integrated systems, and infrastructure ready for new technology. Because no matter how advanced AI becomes, one thing won't change: it won't work in chaos.
You should know, though, that you can already use AI in your e-commerce business today - from a smart chatbot to support for customer service and product listing management. Click the button below or fill out the contact form and find out how AI can strengthen your store and help it stand out from the competition.
