When your customer is an AI agent
Your next customer may never visit your website. They may delegate product search, comparison of terms and order placement to their AI agent. And when that happens, many of the traditional advantages of online stores may no longer work the way they used to.

A customer who pays no attention to advertising
For the past twenty years, we have built e-commerce around people. We have created user-friendly catalogs, improved product pages, optimized search results, experimented with shopping cart design and increased conversion rates.
All of this infrastructure is designed for a customer who chooses a product themselves.
But what happens when customers hand over a substantial part of that work to artificial intelligence?
Imagine a straightforward task: someone needs a washing machine costing no more than 60,000 rubles. It must fit in a small bathroom, be delivered on Saturday and come with an official warranty.
Today, the customer visits several online stores, examines specifications, reads reviews, compares prices and checks delivery terms.
Tomorrow, they might simply assign that task to their AI agent.
The agent will gather offers, eliminate unsuitable options, check whether prices are current and products are in stock, and calculate the total cost, including delivery. It may even place the order itself once the customer confirms the purchase.
The customer may never see the store's home page at all.
A beautiful website no longer guarantees a sale
Suppose two stores sell the same washing machine.
The first has invested heavily in design, advertising and personalization. But customers only discover whether the product is actually in stock after placing an order, delivery costs are calculated at the final step, and warranty terms are hidden in a separate document.
The second has a much simpler website. However, it has up-to-date stock information, transparent terms, structured product specifications and a way to obtain an accurate delivery cost automatically.
Which offer will the agent choose?
If its task is to find the best option according to the given criteria, the second store has a significant advantage.
This does not mean that design and marketing will cease to matter. People will continue to choose products independently, explore brands and make emotional decisions.
But another channel of interaction is emerging, one in which the attractiveness of the storefront gives way to the quality of data and the accessibility of operations.
Two different purchasing processes
In traditional shopping, a person follows the entire journey, from advertising and search to product selection, the cart and payment. At every stage, they are influenced by visual imagery, interface usability, brand trust and their emotional perception of the offer.
In agent-mediated shopping, much of this activity shifts to a digital intermediary.
The customer defines the task and its constraints. The agent collects offers, compares parameters, checks terms and then recommends an option or initiates a purchase within the limits of its authorization.
For businesses, this is a fundamental change: persuading the person is no longer enough. They must also enable interaction with that person's digital assistant.

Three levels of business readiness
I would divide preparations for agent-mediated e-commerce into three areas.
The first is data.
How up to date is the information about products, prices, stock, specifications and delivery times? Can an external system obtain a consistent answer?
If the catalog shows a product as available but the warehouse ran out yesterday, the problem is not artificial intelligence. It is data management.
The second is operations.
Can a digital agent do more than simply read information? Can it take action?
Check availability, calculate delivery, reserve a product, create an order and obtain its status.
This requires reliable integration interfaces, clear business rules and controlled processes. APIs are one way to enable such interaction, but by no means the only one.
The third is trust and security.
If an agent makes a purchase on someone's behalf, the business needs to understand who authorized it, which operations are permitted, where additional confirmation is required and how to handle errors.
Without these mechanisms, automating purchases creates not only opportunities but also new risks.
Readiness first, GEO second
Companies are increasingly discussing GEO — generative engine optimization, or optimizing their presence in the answers and recommendations of generative search systems.
The idea is straightforward: if a customer asks an AI system a question, that system needs to know about your company, understand its offers accurately and be able to recommend its products.
But I see an important management trap here.
A company can invest in promotion for AI, secure mentions in recommendations and attract new prospective customers. Only then might it discover that the agent cannot check actual stock, obtain an accurate delivery cost or complete an order.
We have brought the customer to a beautiful storefront but forgotten to open the door.
I would therefore organize the process differently:
- Put product data and sales terms in order.
- Make key operations accessible and reliable.
- Test a real scenario of interaction with an AI agent.
- Then scale up GEO promotion and the acquisition of agent-driven traffic.
This does not mean ignoring GEO until the fourth stage. Companies should already be improving basic brand visibility, the accuracy of information and the accessibility of content.
However, significant investment in attracting customers through AI should be synchronized with the actual readiness of the business to serve them.
GEO answers the question, “How will the agent find out about us?” Operational readiness answers, “Will it be able to buy from us?” The second question is no less important than the first.

Where should a business leader start?
I would not begin with a large-scale project to create a dedicated platform for AI agents.
It is far more useful to choose one specific scenario. For example, automatically checking product availability and calculating the total cost of a purchase.
Then answer several questions honestly.
Is our data good enough? How many systems are involved in obtaining the answer? Can we guarantee that it is current? Which actions can be performed automatically? Who is responsible for the result?
Such an experiment will reveal not only the company's technical readiness but also the real limitations of its operating model.
And it may turn out that, before working with AI agents, the company needs to address far more familiar problems: putting record-keeping, integrations and business process management in order.
The biggest change
In my view, the most interesting consequence of agent-mediated commerce is a shift in competitive advantages.
Previously, a company competed to bring a customer to its own website.
Now there is an additional task: making its offer accessible, understandable, reliable and attractive to a digital intermediary acting in the customer's interests.
In the new e-commerce landscape, it is important not only to appear in AI recommendations but also to be ready to fulfill the order that the AI agent brings.
Preparing for that should begin not with the purchase of another AI product, but with putting the company's digital foundation in order.