Last week I asked ChatGPT to find a vacuum cleaner for me. Not through Google, not through an online store. Just in a chat window. I typed: "I'm looking for a cordless stick vacuum, quiet, for a small apartment, budget up to 200 euros." Within three minutes I had a buying guide with four options, including prices, reviews and a clear comparison. I didn't need to visit a single webshop.
That sounds like a neat trick. But it's the beginning of something bigger. Much bigger.
What is agentic commerce exactly?
Let me start with a definition everyone can understand. Agentic commerce means an AI agent shops on your behalf. You describe what you need. The AI figures it out, compares options, and in some cases can even complete the purchase. You no longer need to search, filter and click yourself.
McKinsey describes it as AI that predicts consumer needs, explores options, negotiates deals and executes transactions. Morgan Stanley estimates that agentic commerce could account for 190 to 385 billion dollars in e-commerce spending in the US alone by 2030. That's 10 to 20 percent of all online sales.
Let that sink in: we're talking about a market that is virtually zero today.
The 'Smart Shopping' feature in ChatGPT
If you want to see what this already looks like in practice, you don't need to look far. OpenAI launched the "shopping research" feature in ChatGPT at the end of 2025. It works simply. You describe what you're looking for. ChatGPT asks follow-up questions about your budget, preferences and situation. In the background it searches the internet, compares prices, checks reviews and delivery times. After a few minutes you get a personal buying guide with the best options side by side.
The remarkable thing: those recommendations are not advertisements. OpenAI says product selection is independent and not influenced by commercial partnerships. If you have memory enabled in ChatGPT, suggestions are even tailored to your previous conversations. Does ChatGPT know you're a gamer? Then it factors that in when advising on a laptop.
At ShopTalk 2026, OpenAI announced further expansions: richer visual results, side-by-side product comparisons, and image-based search. The direction is clear. ChatGPT wants to be not just your knowledge base, but also your personal shopper.
There is one caveat, though. OpenAI's first attempt to have checkout happen inside ChatGPT, Instant Checkout, has been scaled back. Walmart reported that conversion was three times lower for purchases made directly in ChatGPT compared to when customers were redirected to the retailer's own website. Forrester analyst Emily Pfeiffer summed it up well: the entire ecosystem is still young, and nobody has really figured this out yet. But that the direction is set, nobody doubts anymore.
Google throws an open standard into the mix
Google isn't standing still. At the National Retail Federation conference in January 2026, CEO Sundar Pichai presented the Universal Commerce Protocol, abbreviated UCP. This is an open standard that enables AI agents to communicate with webshops. Browse product catalogs, check inventory, compare prices, and ultimately complete checkout.
The consortium behind UCP reads like a who's who of retail: Shopify, Walmart, Target, Etsy, Wayfair, Zalando, Mastercard, Visa, Stripe, Adyen, Best Buy. They've all signed on. This isn't just a technical standard. It's an agreement about what the future shopping experience will look like.
Google simultaneously launched the Business Agent: an AI assistant that allows shoppers to chat directly with brands on Google Search, as if they were in a store. And Merchant Center gained dozens of new data attributes, specifically designed for conversational commerce.
Google's message is clear. The search results page as we know it, with ten blue links, is being replaced by something fundamentally different. An AI that chooses for you.
Bol and Coolblue: what are the Dutch doing?
The big question for Dutch e-commerce is of course: how are our own platforms responding?
Bol announced a store-wide Shopper Agent at the Webwinkel Vakdagen. CEO Maite Zubiaurre presented "Billie", an AI assistant that will help customers across all product categories. You'll soon be able to tell Billie: "Order Starbucks capsules for me", and the agent handles it. Zubiaurre was candid: assortment, price and delivery time are being made radically transparent by AI. Context becomes more important than ever, because consumers search for products that fit a specific situation.
Coolblue is also investing heavily. In 2025, CEO Pieter Zwart announced that the company would put 150 AI agents live for product advice. That fits Coolblue's DNA: everything revolves around helping customers make the right choice. But there's a paradox here too. Coolblue has built its entire brand on user experience. The beautiful website. The personal delivery service. The reviews with a smile. But an AI agent doesn't see that smile. It sees product attributes, delivery times and return policies.
The shift nobody sees coming
And here lies the heart of the story. The shift of agentic commerce doesn't play out on the screen you see. It plays out in the layers beneath it.
What an AI agent evaluates when selecting a product on your behalf is completely different from what you as a human look at. An agent doesn't look at your homepage design or your campaign images. It looks at structured product data: specifications, inventory status, delivery time, return policies, reviews. If that information is missing or not machine-readable, you simply don't exist in the agent's world.
That's a major shift. UX moves from interface to information. Not how you present it, but what you make available determines your success.
And it goes beyond data alone. The classic marketing funnel, from orientation through comparison to conversion, collapses. There are fewer moments where you as a brand have influence. Consumers no longer come to your website to browse. The AI does the browsing, and only presents the end result.
From ranking to selection
This may be the most important shift to understand. In the world of Google, everything revolved around ranking. Position 1, 2, 3. You paid for clicks, optimized for keywords, fought for the best spot on the page.
In the world of agentic commerce, everything revolves around selection. The AI doesn't present ten options. It presents one, two, at most three. That selection is based on multiple factors simultaneously: relevance to the situation, reliability of your offering, completeness of your data, and your track record on delivery and returns.
You're no longer competing on one dimension. Price alone won't cut it. Position won't either. You compete on everything simultaneously. And small errors in your data can result in being left out of the selection entirely.
OpenAI runs into reality
I also want to be honest about the state of affairs. Because agentic commerce is not a finished story. It's a promise under construction, and some promises are already being revised.
OpenAI's Instant Checkout was the showpiece: buying products without leaving ChatGPT. But it didn't go as planned. Of the million Shopify merchants that were supposed to connect, only about thirty were available after months. Product information was regularly outdated. Prices were wrong. Inventory wasn't current. The complexity of real-time e-commerce (taxes, returns, fraud prevention) turned out to be enormous.
OpenAI has now switched to a model where you're redirected from ChatGPT to retailer apps. Less sexy, but more workable. And honestly: it aligns better with reality. Consumers happily use AI for research, but still want to handle checkout themselves on a trusted site.
That doesn't change the direction. It only tells us the road there is bumpy.
How this will develop
Let me give my own assessment of how this will unfold over the next two to three years.
In the short term, say this year and next, I expect AI agents to take over mainly the research part of the customer journey. Product comparison, summarizing reviews, choosing the best option. The actual checkout will remain on the retailer's website for now.
In the medium term, toward 2028, I expect standards like UCP to mature. Then agents will actually be able to complete transactions, first for simple repeat purchases and later for more complex categories. The category matters: you'll trust an agent with a pack of detergent sooner than with a new bank account.
In the long term, 2030 and beyond, the complete architecture of online retail changes. The webshop becomes less a place where customers come, and more a back-end infrastructure that serves agents. Your homepage becomes less important than your API. Your brand becomes less important than your data.
That's also the paradox for platforms like Bol and Coolblue. Their strength lies in the customer relationship. But if the first interaction happens via ChatGPT or Google's AI, they become dependent on an intermediary layer they don't control. First-party data — your own customer data, your own purchase history — becomes your strategic anchor.
What should you do differently tomorrow?
If you run a webshop, or if you're responsible for the digital strategy of a retailer, there are a few things you can do right now.
First: treat your product data as your most important asset. Not your campaigns, not your design. Your data. Are your product descriptions complete? Is your inventory data current? Are your delivery times reliable? Do you have structured data (schema.org markup) on your product pages?
Second: let the AI crawlers in. Check your robots.txt and give OAI-SearchBot and Googlebot access to your product pages. If you block them, you don't exist.
Third: think in situations, not keywords. Don't optimize your product content for "best vacuum cleaner 2026" but for "quiet cordless stick vacuum for small apartment". That's the kind of question someone asks an AI agent.
And finally: accept that you're losing control over part of the customer journey, and invest even more in the part you do control. Your own channels, your own customer data, your own post-purchase service.
The bigger picture
We're moving from a world where you needed to be visible, to a world where you need to be selected. That's not the same thing. Visibility could be bought with ads. Selection must be earned with data, reliability and relevance.
In the coming years, a new layer will emerge between the consumer and the webshop. A layer of AI agents making decisions on behalf of the customer. Those who understand this layer and anticipate it will have an advantage. Those who wait until it's mainstream will find that the agents have already passed them by.
And it starts today. With a simple question to ChatGPT: "Find me a vacuum cleaner."