Agentic AI in E-Commerce: The New Era of Autonomous Shopping

What is agentic AI, and what can AI agents do?

Agentic AI - systems of autonomous agents capable of making decisions and taking action without user involvement - is the next step after generative AI in e-commerce. When it comes to making purchases in particular, agents seem like a dream solution: they search for products on their own, compare prices, analyze preferences, and order whatever the customer needs. All of this happens very quickly - which, in some examples, sounds almost too good to be true.

Picture this: you forget about a gift for a party and don't have time to pick one out. You tell an AI your ideas, and the agent selects and buys the gift, taking the burden off your shoulders. Another example: you're stuck in traffic and give an AI your shopping list. You pull up to the stores and pick up your finished order. Sounds interesting, doesn't it?

We might still be getting a little ahead of ourselves, but agentic commerce is already making it possible for AI to handle the entire shopping process for the customer - from ideas, to payment, to delivery.

Where is agentic commerce already working? Real-world rollouts and how they operate

Amazon's "Buy for Me" - a shopping agent built into your Prime account

Amazon, one of the pioneers of e-commerce in many respects, has rolled out a feature called Buy for Me, available in select countries (currently mainly the US), while Rufus - a different AI solution from Amazon - is increasingly active in Europe.

Buy For Me works as follows:

• The user describes a need (e.g. "I want a garden set with a grill and chairs, under $1,500").
• The AI agent searches the catalog, analyzes reviews, and picks the best sets based on reviews, data, and what it knows about the user.
• It then completes the purchase without requiring the user's approval.

Buy for Me requires prior authorization for the agent to make decisions and access to a payment account. Amazon's goal is for the agent to operate without needing to use the store interface at all - the customer just "gets the package", meaning the purchase journey is shortened to the absolute minimum. In Poland, Buy for Me isn't available yet (nor is the Rufus agent), but it's worth keeping an eye on - Amazon tests new markets very aggressively, so we can expect these AI solutions to eventually appear on the Polish Amazon site.

"Buy with Pro" on Perplexity - a shopping agent built into search results

Perplexity AI, a tool competing with ChatGPT in semantic search, launched a feature called Buy with Pro in 2024:

• The user asks a question: "Which smartwatch works best for swimming?"
• Perplexity analyzes product data, reviews, and rankings - and suggests a specific purchase, which can be completed directly from within the search tool.
• Integration with Amazon and other e-commerce platforms (including individual stores) allows for an instant purchase without leaving the Perplexity window.
  

Buy with Pro is only available to holders of a paid Pro subscription. Perplexity does operate in Poland, but the shopping feature isn't available locally yet - it requires a foreign account or a VPN.

Shopping in Google with AI Mode - a search engine with a shopping mode

In 2025, Google introduced an experimental feature within the AI Mode of its search engine, available in the US, UK, and Canada:

• Searching for something like "boho-style wedding outfits" takes the user into AI Shopping mode.
• The system analyzes preferences (style, color, price, size) and recommends products.
• Users can enable an automatic purchase mode if they trust the algorithm.

One remarkably interesting feature of AI Mode that sets it apart from the other solutions described here is a virtual fitting room powered by AI - you upload a photo and get back a generated image of yourself wearing the clothes you're browsing. And if you're not sure, or want a second opinion, you can easily save the images and send them to friends or anyone else.

Watch Google's video showing what shopping with AI Mode can look like:

Shopping with Google's AI Mode sounds very interesting, but just like the solutions above, it's not available in Poland yet. As with Google's AI Overview, we'll probably have to wait a little longer for it to roll out here. That said, it does have an upside if you're thinking about tapping into this feature's potential down the line - we get the chance to see (and study the data on) how AI Mode shopping performs in the real world, and how to properly prepare an online store for it.

What potential does agentic AI hold for e-commerce?

Agentic AI in e-commerce simplifies the customer's purchase journey to the absolute minimum - and according to as many as 93% of respondents in a Cisco study, this technology, regardless of the scale at which it's applied, will make purchase journeys more predictable and smoother.

Personalization and shopping automation

Agentic AI shopping lets customers buy products with almost no involvement of their own - the customer communicates their needs, and the system handles the rest. Agentic AI can know a user's purchase history (across many stores), their style, preferences, even their mood (based on their calendar, emails, and social activity), and use that to anticipate needs before they even arise.

This fits what customers want: speed, autonomy, and minimal involvement - though it's worth noting that in this scenario, the brand or the specific store becomes less relevant in the purchase process.

Profit and operational efficiency

Agentic AI can also automate customer service, inventory management, marketing campaigns, and internal support. One example: Walmart in the US is rolling out "super agents" - Sparky for customers, Marty for advertisers, Associate for employees - aiming for a 50% increase in online sales over the next five years, on annual revenue reaching $648 billion.

Recurring and subscription purchases - fully automated

Here's a very interesting example of using agentic AI alongside a broader system, such as IoT (the Internet of Things). By monitoring how quickly products get used up (like laundry detergent, coffee, or pet food), it can remind you or even automatically order the next delivery. If prices, availability, or reviews change - or if a new product appears that might interest the user - it can dynamically switch the product and/or supplier. This way, instead of you having to remember to buy something, AI could do it for you and manage household purchases on your behalf.

The risks of agentic AI for store owners

For now, the idea of AI agents doing the shopping for customers sounds great - but anyone who works with customers, or in e-commerce, knows that a purchase isn't a straightforward process and doesn't always end the way we intend. If we're thinking about AI doing the shopping, we immediately have to think about... mistakes and problems. Not just with customers, but with the products themselves or the funnels you use in your store. So it's worth listing the risk factors that could come with the spread of agentic AI in e-commerce:

Potential errors and complaints

If an agent buys the wrong product (the wrong size, an incorrect spec), the customer will still return it to you. The problem is that:

• you have no influence over the purchase decision,
• you may bear the cost of handling returns, even though the customer wasn't involved in the decision,
• it will be harder to resolve a complaint, since the customer doesn't know the details of the purchase.

Combined with AI's tendency to hallucinate and over-interpret, this is a real problem - and, to put it skeptically, AI in its current state still isn't trustworthy enough to be authorized to make purchase decisions.

Dependence on platforms and ecosystems

Many store owners will have to adapt to the requirements of OpenAI, Google, Amazon, or Walmart - supplying APIs and product data in a specific format. We can also expect integration and maintenance costs for adapting systems to AI agents. That could mean smaller stores and businesses being pushed to the margins, since they don't have the budget for this kind of technology.

Loss of direct contact with the customer

One of the most serious consequences of agentic AI adoption in e-commerce - from a store owner's point of view - is the loss of the relationship with the customer. In the traditional model, the user visits your site, browses the offer, reads descriptions, clicks on products, sees promotional banners, and might sign up for a newsletter - in other words, they interact with your brand.

In the agent-driven model, none of that may happen at all.

An AI agent operates on its own. It receives a general instruction from the user (e.g. "find me a cheap phone with good battery life") and searches offers from many sources itself. If your store has the right product and data, the agent might choose you - but the customer doesn't even have to know exactly where the purchase was made. The consequences are fairly serious:

• No chance for upselling or cross-selling. Since the customer never sees your site, they won't see a pricier version of the product or related accessories either.
• No influence over the customer experience. You can't build a positive visual impression, showcase your site's design, your added value, or your brand's philosophy.
• No loyalty. The customer remembers that "AI handled it", not that they bought from you. They won't come back on their own - because they never registered you as the seller.

This is a fundamental shift in the logic of online selling. We're moving from the classic model of "attract - showcase - sell - retain" to a model of "get chosen by AI - fulfill the order". Brand-building efforts would then need to intensify in post-purchase touchpoints. Branded packaging, a flyer inside the box, maybe even a small gift, can help your store get remembered better by the user. In the same way, processes like customer service, returns, and complaints would need to be re-evaluated and better implemented to strengthen customer satisfaction and improve brand or supplier recall.

Risk of manipulation and abuse

Whoever controls the agent decides what the customer buys. That raises some questions:

• What will AI pay the most attention to? Price, quality, customer reviews?
• Does the agent choose the best product... or the one being promoted by a sponsor?
• How biased could an agent turn out to be?
• Does the algorithm act in the customer's interest, or the platform's?

For store owners, this creates a risk that unscrupulous platforms could manipulate rankings or extract fees for higher placement.

Agentic AI in practice - predictions and closing thoughts

Although agentic AI is one of the hottest topics in e-commerce, we're still at the stage of early tests and pilot rollouts. Amazon, Google, and Perplexity have shown what shopping driven by autonomous agents could look like, but most of these solutions aren't available globally yet - let alone in Poland. It's reasonable to expect some time will pass before they reach our market in full form. Even so, savvy online store owners can already start acting and preparing for the change ahead.

For stores running on popular SaaS platforms like WooCommerce, Shopify, or Shoper, the first step toward agentic AI is smart chatbots. This is a real, deployable-today solution that delivers noticeable value. A chatbot like this can understand a user's natural language, identify their needs (e.g. "I'm looking for a gift for a 10-year-old, under 100 PLN"), search the store's catalog, and carry the conversation all the way through to completing a purchase. It can also draw on existing data for returning customers.

For the customer, this is simpler, more convenient, and more intuitive than classic product filtering. For the store owner, it's a chance to increase conversion and stand out from the competition.

Sellers operating on Amazon will likely be the first to feel the impact of agentic AI on sales. Amazon is aggressively developing its "Buy for Me" service, which lets agents place orders directly on the user's behalf. It's worth noting, though, that this technology will likely be available at first mainly to sellers using the FBA model - those relying on Amazon's warehouses and Prime logistics. Amazon has an interest in having AI agents pick products that can be delivered quickly, cheaply, and without issues - and only its own infrastructure can guarantee that. So we can expect a dedicated order channel to emerge for handling agents, with products having to meet specific criteria - complete data, availability, high service quality.

In Google's case, data will matter most. A product feed connected to Google Merchant Center, complete and up-to-date product information, good photos, and accurate descriptions - all of this will influence whether a given product gets picked by a shopping agent within Google's AI Mode. The type of product will matter too - Google demonstrated the virtual fitting room using clothing, but it's safe to assume it will also apply to things like home décor and furniture, cosmetics, or even, going further, car rims or... roof tiles. As you can see, the range of applications is broad - but it won't cover every type of product on the market. So not everyone will be able to take full advantage of what AI Mode offers. It's also worth revisiting the topic of product advertising - who knows whether Google will end up enabling product promotion in a way that's completely different from how it works today.

There's a lot pointing toward a shift in e-commerce, where the battle for customers will be shaped not just by ad campaigns, but also by readiness to communicate with AI - through complete data, a clear structure, semantic descriptions, and technological readiness. It's worth acting now - clean up your data, review your product feed, and test out chatbots. Check how your site performs from an algorithm's point of view - and how much untapped potential it still has.

Let's talk about your e-commerce

You don't need to roll out full AI in your store right away, but it's worth checking now whether your sales process, order management, and data structure are ready for what's coming.

• we'll look at how your sales and logistics processes work,
• we'll assess how ready your store is to work with new technologies,
• we'll show you specific places where you can boost efficiency and better prepare to tap into the potential of agentic AI.

Get in touch with us - no strings attached - and get concrete recommendations.

Fill out the contact form