Agentic commerce – AI agents in e-commerce

How we've shopped over the years

Commerce has always sought to shorten the distance between a need and purchase. We started with stationary stores, where the interface was the man behind the counter. Then came the era of telephone sales, which gave us convenience, but took away the visual aspect of the product. In the end, we were dominated by e-commerce — online catalogs, in which we, the customers, had to learn how to use filters, search engines and complicated checkout processes.

Today we are on the threshold of the fourth shift. The graphical interface where you click on the tiles begins to give way to Agentic Commerce. This is the moment when you stop “operating the store” and start “ordering purchases”.

From chatbots to AI agents with permissions

We remember the first wave of chatbots. They were supposed to support customer service, but in practice they most often annoyed by serving rigid answers from the FAQ base. The real breakthrough came with the development of Generative AI and what we have already written about in the context of AI agents.

The difference is fundamental: the chatbot only talks. The AI agent is working. In the traditional model, you are looking for running shoes. You enter the site, check the filters: “forest”, “soft sole”, price “up to EUR 100”. You browse 20 pairs, you read the descriptions. In the Agentic Commerce model, you write one sentence in a chat or say to the phone: “Find me shoes for running in the forest, with a soft sole up to EUR 100, which have the best reviews for grip on mud.” The agent not only finds them, but analyzes the technical characteristics and compares them with your previous purchases to hit the size. The agent becomes an expert who leads until the product is added to the cart.

A new era: shopping without entering the platform (ACP and Instant Checkout)

The real paradigm shift, however, it's the moment you realize that your sales can take place entirely outside the online store — and we're not referring to eBay, Amazon or Google. In 2026, consumer behavior is evolving towards conversation. Increasingly, instead of typing phrases into a search engine, users spend time inside interfaces such as ChatGPT, Gemini or Perplexity, treating them as their personal shopping assistants. This is where the decision-making process takes place. The customer is no longer looking for a link to the store; he expects the assistant in the form of artificial intelligence, with whom he is just talking about planning a trip to the mountains, to immediately select the right backpack for him and go through the finalization of the transaction.

This is where comes the idea of Instant Checkout. It is a mechanism in which the purchase process is shortened as much as possible: from the intention (“I need it”) to the confirmation of payment, without having to go through the cart, create an account or fill in the delivery forms on the seller's website. Everything happens inside the chat. This is no longer a vision of the future — it is a reality driven by open initiatives such as the Agentic Commerce Protocol (ACP).

What is the Agentic Commerce Protocol (ACP)?

Thanks to the collaboration of technology giants (such as OpenAI and Stripe), ChatGPT has gained a communication standard that allows it to interact directly with the payment and warehouse infrastructure of e-shops. The AI agent checks availability, calculates delivery costs in real time by negotiating with carrier APIs (e.g. FedEx, DHL) and initiates payment using secure user data. As a result, protocols like ACP can become for commerce what HTTP has become for web pages faster than we think.

ACP is not the only initiative in this field, as Google and Shopify are currently also creating an open specification dedicated to Agentic Commerce — Universal Commerce Protocol. Therefore, we can expect similar solutions in other popular LLMs soon.

In the agent world, traditional SEO (keyword optimization) is giving way to AIO (AI Optimization), where semantics and real-time data delivery rather than static descriptions matter. However, it should be borne in mind that in this AI commerce approach, marketing activities change to providing structured data (e.g. via API or advanced product files e.g. in JSON-LD format), so that AI agents can “read” your store's offer. If the agent does not “see” you in the protocol, for the client you do not exist. Therefore, the question is: is your e-commerce technologically ready to “pitch” its offer to the world of third-party AI agents, or does it still assume that the customer will always land on your homepage on their own?

Agentic commerce is not just a “chat” — it's the concept of mlutichannel commerce

It would be a mistake to limit agent technology only to websites. The true power of Agentic Commerce manifests itself where painstaking, manual human labor has been required up to now: in phones and email support.

• Voicebots and automatic phone orders: In companies where speed and 24/7 availability matter, AI voicebots in e-commerce are becoming full-fledged merchants. The client calls a familiar number and, instead of waiting for a call to the consultant, places an order with the bot: “I need five bags of mortar X and delivery for tomorrow for the construction of Y.” Voicebot verifies statuses, reserves goods and completes the transaction in real time.
• Phone calls with the “background assistant”: there are industries — such as specialistic goods or B2B — where the relationship with the customer is sacred. The trader must have time to joke, ask about the weather or maintain a relationship with a long-term contractor. Here voicebot would be a mistake, but AI commerce can act as an “invisible helper”. While people are talking, the agent listens and generates a transcript (that is, it converts speech to text in real time - you can read about how we do this in our article on AI transcriptions: https://www.sagiton.pl/en/blog/how-to-use-ai-transcriptions). On this basis, e-commerce automation captures key information, checks the availability of products in the ERP and prepares an order. The trader does not have to manually enter anything into the system — after putting down the handset, he just glances at the finished sketch and sends it to the customer via WhatsApp or email for confirmation with one click. In critical industries, such as the sale of medical equipment or aerospace components, where a mistake for a single symbol can be disastrous in consequences, such a mechanism of human verification before shipping an order is a key fuse.‍
• Smart Email Support: mailboxes in commercial departments are places where processes often get bogged down in attachments. Agentic Commerce in this edition is an AI that not only reads the content of the message, but analyzes the submitted specifications, lists of requirements in PDF or statements in Excel. Artificial intelligence automatically extracts data from these documents, verifies them with our assortment and prepares a complete quote or draft order. Thanks to this, the response time to a request for quotation is reduced from a few hours to a few seconds. This is one of the key elements of modern sales automation, which allows traders to focus on closing transactions rather than rewriting data from email to the system.

AI as an operational assistant: eliminating processes that cost rather than earn

Agentic Commerce is not just a “buy button”. The real revolution takes place where AI takes over the operation of processes that generate only costs, without bringing any direct profit. Speaking of administrative “back-office”, which in any trading company is a huge financial burden.

A typical example is a request to change data on an invoice or issue a correction. For an employee, this is a few minutes of tedious clicking, for which no one pays him an additional margin. For an AI Agent, it's a matter of seconds. The customer writes: “I have mistaken the VAT ID in the order from an hour ago, please correct it on X.” The agent enters the accounting system, verifies the permissions, applies a correction and sends the document back. The company saves the time of a specialist who can deal with the actual closing of the sale.

The same goes for return logistics and complaints. These are “cost-generating” processes — every minute spent on them is a real loss to your margin. Thanks to the use of AI in e-commerce, the automation will independently guide the customer through the return process: it will verify the legitimacy of the complaint on the basis of the submitted photos (visual analysis), generate a courier label and reserve the goods for exchange.

In the service area, the role of e-commerce automation is shifting from “ordering” to intelligent management. Although recurring services follow a schedule, their coordination can be a logistical nightmare. The AI agent not only keeps an eye on deadlines, but actively manages incidents. When a customer reports a device malfunction, the agent analyzes the free slots of technicians in the calendar and immediately “agrees” a date for the visit that suits both parties. It's about reducing the role of customer service to pure, cost-free automation.

Fear of automation and AI agents — how to overcome distrust?

Of course, every technological revolution initially arouses resistance — it is a natural defense mechanism against the unknown. Today, many customers declare distrust of AI, which is supposed to make purchasing decisions for them or manage their portfolio in case of e-commerce owners. However, it is worth looking at it from a historical perspective: until a dozen years ago, it was considered crazy to buy clothes online without trying them on first. Today it is a standard that drives billions in turnover. So it was with payments — out of fear of “stealing the signal” from the card, we went into the agenda over paying with a phone, a watch, a smart ring, and even eye biometrics. What yesterday was an impassable barrier, tomorrow becomes a transparent element of everyday life.

Convenience always ultimately wins with uncertainty, as long as the user trusts that the process is safe. At Agentic Commerce, we build that trust through layered authorization, tailored to where the conversation takes place:

1. Transaction Approval (Human-in-the-Loop): An agent is never an “arbiter.” Regardless of the channel, it ends the process with a request for acceptance, ensuring the so-called non-repudiation
   
   • In the online store: The AI agent in e-commerce generates an interactive card with a summary and a “Pay” button inside the conversation window. Clicking opens a secure, native payment window (such as Stripe), where you approve the amount with one swipe.
   • In ChatGPT/External Assistants: The agent sends an authorization request to your digital wallet. You receive a push notification on the device you currently have with you and confirm the transaction with biometrics.
2. Smart Limits and Permissions: The user defines the “level of freedom” of the system. You can allow him to order consumables (e.g. printer toner or coffee for the office) for up to EUR 25, while any unusual or larger transaction requires your clear “yes”.
3. Transparency of decisions: Trust builds understanding. An agent can't be a black box. To each question, it must be able to answer: “I chose this product because it is X-certified, fits your device model and is available with delivery in 24 hours”.
4. Data security (tokenization): thanks to standards such as ACP, AI commerce does not “see” your full card number. It operates on secure tokens that are assigned to a specific purchase. Even if the conversation with the agent were intercepted, the financial data remains intact in the secure vault of the payment processor.
5. Identity and Digital Credentials (W3C): the agent does not need to know your store passwords. Thanks to the Verifiable Credentials standards (from W3C), it is certified with a digital credential. The store receives cryptographic proof that the agent has the right to act on your behalf, but your login details and identity remain under your sole control.

Summary: From Trend to Market Necessity

Until recently, Agentic Commerce was treated as a curiosity on the verge of science fiction. Today, it is a real competitive advantage tool that allows the fastest players in the market to take over a customer where the competition does not even display their advertising. However, looking at the pace of adaptation of LLM models and protocols such as ACP, tomorrow Agentic Commerce will simply become the market standard and a necessity to stay in the game for the attention of new customers.

However, it is worth remembering that in 2026 we earn double in e-commerce. Not only on sales itself and increasing conversions, but above all on radically reducing unnecessary operating costs. The implementation of the automation of tedious administrative processes, which Agentic Commerce guarantees, allows you to convert expensive working hours of the service department into automatic operations performed by algorithms. In an era of rising customer acquisition costs and pressure on margins, it is this process efficiency that can prove to be your biggest profit.