E-commerce ready for AI agents

How to prepare your platform for purchases delegated to arificial intelligence

Until recently, the development of e-commerce focused primarily on how a user moves through an online store. We analyzed the shopping journey, category structure, search, product pages, checkout, payment methods and the quality of the mobile experience. The entire architecture of online sales was designed with a human in mind – a person who clicks, compares, reads, filters, adds a product to the cart and makes the purchasing decision independently.

This model is not disappearing, but it is beginning to change. It is becoming increasingly clear that the future of e-commerce will not depend solely on whether the store is convenient for the end user. It will also become increasingly important whether the sales platform is understandable, accessible and reliable for AI systems that will support customers in searching for products, comparing offers, analyzing parameters and, in some scenarios, also completing the purchase.

In practice, this means that e-commerce is entering a stage in which an online store will no longer communicate only with a human. It will also have to communicate with AI agents, shopping assistants, recommendation systems, conversational search engines and external platforms that will interpret the store’s offer on behalf of the customer. IBM describes agentic AI as systems capable of achieving specific goals with limited human supervision, which means not only answering questions, but also planning actions and performing the next steps within a process.

For retail companies, this means a very specific change. Visibility in Google, a good product page and an efficient checkout will still be important, but they will no longer be enough. If a customer starts asking their AI assistant about the best product for a specific use case, the most advantageous B2B offer, product availability on a specific date or a technical substitute matching their needs, the winners will not only be the companies with good marketing. The winners will be those whose data, architecture and processes are prepared to be correctly read, interpreted and handled by a new type of purchasing intermediary.

Why this topic is becoming important right now

Agentic commerce is no longer a distant technological concept. In January 2026, Google announced the Universal Commerce Protocol as an open standard for agentic commerce, covering the entire purchasing process – from product discovery, through purchase, to post-purchase support. Google also points out that agentic commerce means a situation in which AI performs tasks on behalf of users, and that systems themselves must be able to cooperate with various agents, platforms and payment providers. 

At the same time, standards related to payments carried out by AI agents are developing. Google Cloud presented the Agent Payments Protocol, an open protocol intended to support secure transactions carried out by agents on behalf of users, with an emphasis on greater interoperability, trust and control over authorization. 

For the e-commerce industry, it is also important that this direction is beginning to be addressed directly by commerce platforms. In 2025, Shopware already wrote that agentic commerce is no longer a distant vision, and that AI agents will discover, compare and buy products on behalf of customers. In the same material, Shopware emphasizes the importance of open standards, interoperability, merchant data sovereignty and control over processes.

An even stronger signal is the Shopware 6.7.10.0 update from May 2026, which introduced a new type of sales channel: Agentic Commerce. According to Shopware, it is intended to serve as a central entry point for AI-based product distribution, enabling products to be shared with platforms such as ChatGPT through a compliant JSONL feed and allowing the impact of AI-generated traffic on business to be tracked.

This is a very important change because it shows that the topic no longer concerns only forecasts, conferences and experiments by large technology companies. It is beginning to enter specific e-commerce platform features. If a sales platform has a separate channel for agentic commerce, it means that companies should start thinking about AI agents in the same way they previously thought about marketplaces, price comparison engines, Google Shopping, mobile applications or B2B channels. Not as a curiosity, but as another environment in which the offer must be correctly available, understandable and controlled.

The problem does not start with AI, but with data quality

The biggest mistake in the discussion about agentic commerce is assuming that it is enough to connect the store to a new AI channel, generate a product feed and wait for sales. This way of thinking is very similar to the approach that for years led companies into problems with marketplaces, ERP integrations, PIM, price comparison engines and marketing automation. The channel itself does not solve problems if data, processes and business logic are inconsistent.

An AI agent does not look at a store the way a human does. It will not be impressed by a banner, it will not guess a missing parameter and it will not interpret an incomplete product description the way an experienced sales representative would. If the system is supposed to compare products, match them to the user’s needs, check availability, price, delivery terms and purchasing options, it must work on data that is consistent, complete and machine-processable.

In practice, this means that many companies will have to return to the foundations that have been postponed for years. To the quality of product attributes. To category consistency. To proper variant mapping. To logical relationships between products. To current inventory levels. To the correct handling of individual prices, promotions, discounts, limits and commercial terms. To technical data that is not stored only in PDFs, marketing descriptions or Excel spreadsheets scattered across the organization.

In B2C, the consequence of poor data may be lower product visibility, an incorrect recommendation or losing to a competitor that described its offer better. In B2B, the consequences are even more serious, because a purchasing agent may operate in an environment with individual price lists, budget limits, user permissions, order history, contractual terms and specific purchasing logic. If the system cannot clearly answer what price applies to a given customer, whether the product is available, whether the order requires approval and whether a given user has the right to place it, the AI agent will not solve the problem. It will only reveal it faster.

That is why preparing e-commerce for AI agents is not a marketing project. It is a data, architecture and integration project.

E-commerce for people and e-commerce for machines

For years, companies have designed stores to be visually attractive, intuitive and convenient for the user. This is still important, but in the era of agentic commerce there is a second dimension of design: machine-readable commerce, meaning a sales environment readable by external systems.

The point is not to replace the store interface with data files. The point is for the offer, purchasing logic and transaction terms to be available both to a human and to a system acting on their behalf. This means greater importance of APIs, product feeds, structured data, consistent identifiers, proper variant labeling, correct product semantics and integrations with source systems.

In the context of B2B solutions for 2026, Shopware describes agentic commerce as a trend in which e-commerce platforms are increasingly operated by autonomous AI agents, while software must provide machine-readable interfaces and structured product data so that purchasing agents can predict demand, evaluate offers and place orders within defined budgets.

This statement shows the scale of the change well. Agentic commerce is not limited to a user asking a chatbot for a product recommendation. In more advanced scenarios, an agent may operate within defined budget frameworks, analyze availability, check alternatives, evaluate purchasing conditions and initiate a transaction. If such a future is to work safely and predictably, the e-commerce platform must be prepared to communicate with systems that need not only content, but also a precise data structure.

In practice, this means that companies should start asking themselves new questions. Is our offer understandable without the visual context of the store? Are all important product features stored as data, and not only as text descriptions? Are variants, bundles, substitutes and complementary products logically connected? Will the price presented in the AI channel be consistent with the price in the store, ERP and sales system? Will a B2B customer receive the correct terms if their purchase is supported by an external agent? Are we able to measure traffic and orders coming from AI channels?

If the answer is “we don’t know”, the company probably does not yet have an AI problem. It has a problem with the readiness of its e-commerce architecture for a new method of sales distribution.

Where companies most often make mistakes

The most common mistake is treating AI as another layer placed on top of existing chaos. A company has inconsistent product data, complicated pricing exceptions, several sources of truth, integrations that work partly manually, unstable inventory synchronization and processes that are not fully reflected in the system. Despite this, it expects AI to improve customer experience, increase sales and relieve the team.

In reality, AI does not fix disorganized e-commerce. AI accelerates action based on what already exists. If the data is good, it can help with better offer matching, process automation and faster guidance of the customer through the purchase. If the data is poor, it can generate incorrect answers, wrong recommendations and a false sense of automation faster.

This is particularly visible in B2B companies that for years developed digital sales as an addition to the traditional work of sales representatives. In such a model, many rules still function outside the platform. A salesperson knows which customer can receive a discount, which product can be replaced with another, when it is possible to exceed a standard limit, who has to approve an order and how to interpret non-standard contractual terms. The problem begins when the company wants to move this process into e-commerce and then additionally make it available to AI assistants or purchasing agents.

If business rules are stored in people’s heads, emails, spreadsheets or undocumented exceptions, they cannot be safely automated. An AI agent cannot function properly in an environment where sales logic has not previously been translated into the system. That is why implementing agentic commerce requires not only technology, but also process organization, responsibility for data and a clear definition of what serves as the source of truth in the company.

The second mistake is thinking only about product visibility, not the entire purchasing process. The product feed itself may increase the presence of the offer in AI channels, but it is not enough if the customer or agent cannot correctly go through the next steps: checking price, availability, delivery, payment, return terms, product configuration, order approval or post-purchase service. Google, when describing the Universal Commerce Protocol, points out that the standard is intended to cover the entire shopping journey, from discovery and buying to post-purchase support, not only the moment of product presentation.

This is very important because the future of AI in e-commerce will not be limited to answering the question: “which product should I choose?”. The increasingly important question will be: “can the entire purchasing process be correctly handled by an ecosystem involving a human, an AI agent, a commerce platform, a payment system, ERP, PIM, CRM and logistics?”.

Consequences for sales, operations and costs

Companies that do not prepare their platforms for agentic commerce may begin to lose visibility in new customer touchpoints. Until now, the fight for user attention took place in search engines, social media, marketplaces, paid campaigns and directly in the store. In the agentic commerce model, part of this fight moves into environments where the user does not browse ten stores, but asks one question and expects a specific recommendation.

If an AI agent compares offers, data quality will become extremely important. A product with an incomplete description, inconsistent parameters, missing availability information or an unclear variant structure may be omitted, even if it looks attractive in a traditional store. In this sense, the SEO of the future will not only concern content and links. It will also concern whether product data is complete, up to date, consistent and machine-interpretable.

The operational consequences will be just as important. If AI agents begin to generate traffic, inquiries and orders, companies will have to measure their impact on sales, conversion, service costs and lead quality. The Shopware 6.7.10.0 update explicitly indicates the possibility of tracking the impact of AI-generated traffic using the existing affiliate infrastructure. This shows that agentic commerce will require not only a new channel for data distribution, but also a new approach to analytics.

In B2B, the consequences may be even deeper. If a business customer starts using a purchasing agent that is supposed to compare suppliers, control the budget, ensure compliance of purchases with company policy and automate repeat orders, the sales platform must be able to answer far more questions than a classic B2C store. It must know who is buying, on behalf of which organization, under what terms, with what limits, at what permission level and according to what approval rules.

If these elements are not standardized, the company begins to bear the cost of exceptions. Every non-standard discount, manual process, ambiguous order status or lack of synchronization between systems becomes an obstacle to automation. The more such exceptions there are, the harder it becomes to safely make the platform available to new AI channels.

What the right approach should look like

Preparing e-commerce for AI agents should begin with a platform readiness audit, not with choosing a tool. The first step should be to check whether the current architecture allows controlled sharing of product, pricing, inventory and transaction data with external systems. If each channel uses a different set of data, and the timeliness of information depends on manual team activities, the company should first organize its sources of truth.

The second step is product data analysis. In the era of agentic commerce, a marketing description is not enough. Attributes, relationships, classifications, identifiers, variants, compatibility, substitutes, units of measure, technical data, certificates, availability and rules for presenting products in different channels become important. In many companies, this means the need for stronger integration of e-commerce with PIM or organizing the way product data flows between ERP, PIM, the sales platform and external channels.

The third step is translating business logic into the system. In B2B, individual prices, shopping lists, budgets, user roles, approval processes, order history, delivery terms, product availability for selected customers and relationships between accounts in the organizational structure are particularly important. If an AI agent is to support a customer in purchasing, it must operate within the same rules that apply in the platform, ERP and sales team.

The fourth step is preparing the integration layer. Agentic commerce will not function reliably if the sales platform does not have efficient connections with systems that store the most important data. In one of its materials on B2B 2026, Shopware points out that consistent data, pricing logic, customer service rules and smooth transitions between self-service and sales supported by a sales representative create the foundation for AI- and agent-supported workflows.

The fifth step is governance, meaning control over what the agent can do, what data it can read, what actions it can initiate and at which point a human decision is required. This element is particularly important because the agentic AI market is developing quickly, but not all projects deliver real value. Reuters, citing Gartner, reported in June 2025 that more than 40% of agentic AI projects may be cancelled by the end of 2027 because of rising costs and unclear business value.

This forecast does not mean that agentic commerce should be ignored. Rather, it means that companies should approach it maturely. They should not implement AI just for the effect of novelty, but build foundations that will allow them to use this channel when it becomes a real source of sales and customer service.

Why e-commerce architecture is becoming more important than a single feature

In the traditional approach, companies often assessed an e-commerce platform through the lens of a feature list. Does it have a cart? Does it support promotions? Can payments be connected? Does the search work? Is there a B2B module? Can a landing page be created? This way of assessment was understandable, but it is increasingly proving insufficient.

In the agentic commerce model, the most important question is different: can the platform safely, consistently and scalably share data and processes with various channels, interfaces and external systems?

This shifts the focus from features to architecture. API-first, modularity, the ability to work headless, integrations with source systems, central channel management, permission control, automation mechanisms and the method of data storage are gaining importance. That is why companies that for years have built e-commerce as a collection of custom workarounds may have a greater problem entering agentic commerce than companies that previously invested in organized architecture.

An AI agent will not patiently work around the limitations of an old system. Nor should it be given access to random data just because the company wants to appear quickly in a new channel. If the platform is to communicate with agents, it must have clearly defined boundaries, rules, data formats, authorization, action audit and the ability to track business results.

That is why preparation for agentic commerce should be treated as part of a broader e-commerce development strategy, not as a separate experiment. The same applies to AI-first development, integrations, PIM, automation, analytics and the development of sales channels. All these elements have one common denominator: without good architecture, a company will increasingly produce additional layers of complexity faster and faster.

Why Shopware responds well to this direction

Shopware is a particularly interesting platform in the context of agentic commerce because it has been developing for years toward openness, flexibility and API-first. This matters because the future of online sales will depend less and less on one classic storefront, and more and more on the ability to support many channels, interfaces and purchasing scenarios from one consistent environment.

In practice, Shopware allows companies to build an architecture in which the e-commerce platform is not a closed store, but the central element of the sales ecosystem. It can support different sales channels, integrate with ERP, PIM, WMS, CRM and payment systems, while also enabling frontend development in a headless or composable model. This is important for companies that want to develop B2B, B2C, cross-border sales, marketplaces, mobile applications, a customer portal and, in the future, also AI-based channels.

The new Agentic Commerce sales channel in Shopware 6.7.10.0 is a good example of this direction. It is not only about the technical ability to generate a feed. It is about AI channels beginning to be treated as a separate category of product distribution that can be configured, developed, monitored and integrated with the rest of the platform.

For B2B companies, it is particularly important that Shopware makes it possible to combine platform flexibility with the logic of business processes. If an organization needs individual price lists, customer structures, different permissions, quotation processes, ERP integrations, availability rules and support for multiple channels, the platform architecture must be prepared for complexity. Agentic commerce does not remove this complexity. It makes it necessary to organize it even better.

From our perspective, the most important thing is that Shopware does not lock a company into one way of selling. It allows e-commerce to be designed as a system that can develop together with the market. Today this may mean B2B, B2C and cross-border sales. Tomorrow it may mean integration with AI agents, conversational channels, new payment models, purchasing automation and more advanced personalization.

How CREHLER helps prepare e-commerce for this sales model

At CREHLER, we look at agentic commerce not as a single feature, but as a natural consequence of what we have been saying for a long time: modern e-commerce must be designed architecture-first. If a platform is to be ready for new channels, AI, automation, integrations, international development and more complex sales processes, it cannot be built only as a store with a set of features. It must be designed as a scalable business environment.

That is why in e-commerce projects we analyze not only the appearance of the store and the list of functional requirements, but also data, processes, integrations, sources of truth, pricing logic, customer structure, product management and possibilities for further development. In the context of agentic commerce, this approach becomes even more important, because every error in data or process can be multiplied in new channels faster than before.

Preparing a platform for AI agents should cover several areas. First, it is necessary to check whether product data is complete, consistent and ready to be shared in formats readable by external systems. Second, it is necessary to analyze whether pricing, availability and purchasing logic are clearly reflected in the system. Third, it is necessary to assess whether integrations with ERP, PIM, WMS, CRM and payment systems ensure data timeliness. Fourth, analytics must be prepared to measure the impact of AI channels on traffic, conversion, orders and profitability.

In projects implemented on Shopware, we can combine these elements into a coherent architecture. Shopware provides a solid technological foundation, but only the right implementation determines whether the platform truly supports the company’s development. For this reason, it is highly important to work with a partner who understands not only the technology itself, but also sales processes, integrations, B2B, B2C, product data, UX, automation and the long-term maintenance of the platform.

Agentic commerce will require greater maturity from companies. It is not enough to have an online store. You need an organized sales ecosystem that can communicate with people, systems, external channels and AI agents. Companies that start preparing earlier will not have to react chaotically when new purchasing models become the standard. They will be able to use them as an advantage.

What companies should do now

The most reasonable first step is not to implement an AI agent, but to assess platform readiness. It is worth checking whether the current e-commerce has stable foundations that will allow it to safely develop sales through AI channels in the future.

At the strategic level, this means answering several questions. Do we know which data is the source of truth? Is the product described in a way that can be understood by both a human and a system? Are prices, availability and commercial terms consistent across all channels? Does the platform have an API-first architecture and can it communicate with external systems? Are we able to distinguish AI-generated traffic from traditional user traffic? Do we know which actions an agent can perform independently and which require human approval?

These questions may seem technical, but in reality they are highly business-oriented. They concern offer visibility, service costs, data quality, transaction security, control over sales and the company’s ability to respond quickly to market changes. Agentic commerce will not be a separate world disconnected from current e-commerce. It will be another layer that uses what the company already has – good or bad.

That is why it is worth starting preparations with architecture. With organizing data. With integrations. With processes. With responsibility for product and commercial information. With checking whether the platform that supports sales today will also be able to support the channels that are only beginning to gain importance.

Companies that organize e-commerce earlier will use the new sales channel faster

Agentic commerce does not mean that the classic online store will stop mattering. Rather, it means that the store will become part of a larger ecosystem in which purchasing decisions will increasingly be supported by AI. The customer will still expect a good offer, efficient service, transparent terms and trust in the brand. What will change, however, is the way they reach that offer and who helps them evaluate it.

For retail companies, this is an important moment. Agentic commerce can be treated as another fashionable slogan, and companies can wait until the market forces a reaction. It can also be treated as a signal that e-commerce must be designed with greater maturity: as a system of data, processes, integrations and channels, not only as a visible frontend.

From our perspective, the greatest advantage will be gained by organizations that begin organizing their foundations now. Not because every store must immediately support purchases made by AI agents. But because the same foundations are necessary for scaling B2B sales, cross-border development, process automation, personalization, ERP and PIM integrations, efficient analytics and the further development of the platform.

If you want to check whether your e-commerce platform is ready for the next stage of online sales, it is worth starting with a conversation about architecture, data and processes. At CREHLER, we help companies design and develop scalable e-commerce platforms based on Shopware – platforms that respond to today’s business needs, but do not close the door to future channels, technologies and sales models.