How Agentic Commerce Is Changing Enterprise Customer Interactions

When you think of online shopping, you’ll likely imagine customers navigating to a website, searching for the product they want, comparing options and completing purchases with no outside help.

This process is beginning to shift with the advent of AI agents by taking on some of these tasks on a customer’s behalf. Instead of the customer interacting directly with a business, a customer might increasingly rely on an AI agent to find a product, make a booking or initiate a transaction.

This now creates a new challenge for enterprises. Customer-facing systems must be able to recognise when an AI agent is acting on someone’s behalf, establish what it’s permitted to do and ensure that transactions are handled appropriately. This is the point at which agentic commerce moves beyond just automated shopping and becomes a question of how businesses can properly manage customer relationships when systems act on their behalf.

What Happens When the Customer Is a Machine?

A visitor arrives on a company's website. They check product details, compare prices and ask whether a particular item is available. But the visitor in question isn't a person. It's an AI agent acting on a customer’s behalf.

This is one of the changes happening behind agentic commerce. Instead of navigating websites and completing every step themselves, customers can potentially use AI agents to find products, compare options and, where supported and authorised to do so, make purchases.

The same concept could extend beyond shopping. An agent might help a customer manage an online booking, request information from a service provider or make changes to an existing agreement.

These possibilities are still developing, and naturally not every service can be handled this way today. But they do raise an important question for enterprises: what happens when a customer interacts with a business through software rather than personally through its website or application?

The challenge goes beyond providing another way to shop. Businesses need to recognise which agents are legitimate, establish which customer they’re representing and determine what they’re allowed to do.

Their existing systems must also be able to process requests correctly with these AI agents in control and provide support when something goes wrong.

How Does Agentic Commerce Change the Customer Journey?

Traditional ecommerce is a customer going to a website, choosing a product from a variety of options, looking through the details to find one that fits their needs and checking out.

AI shopping assistants are already helping with some aspects of the journey, such as quickly finding products and comparing relevant options. Agentic commerce takes the idea further by allowing an agent to carry out tasks on the customer's behalf, including completing a transaction when the relevant systems and permissions allow it.

For example, a customer could ask an AI agent to find a laptop that meets certain requirements, costs less than £1,000 and can be delivered within three days.

The agent in question could compare products, check availability and prepare an order. If the merchant does support agent-led checkout, the agent could complete the purchase within the permissions granted by the customer. In this scenario, the customer may never need to visit the retailer's website but still receive the product they’re looking for.

This changes how businesses receive customer requests. Rather than relying solely on people interacting with their websites and applications, they may need to support requests from external AI platforms and agents.

Google's Universal Commerce Protocol (UCP), which was introduced in January 2026, aims to allow AI platforms, merchants and payment providers to talk to each other with a common set of rules. It enables commerce activities such as product discovery, checkout, and order management while allowing businesses to use their existing infrastructure as is.

OpenAI's Agentic Commerce Protocol is another example of this. Its documentation describes how participating merchants can support checkout within ChatGPT while retaining control of order validation, payment processing and fulfilment through their existing systems. It should be noted that this checkout experience is currently limited to approved partners.

These developments show that the infrastructure for agent-mediated transactions is taking shape. They don’t mean that all online purchases are going to become autonomous or that every merchant will support these interactions.

Taking this into consideration, enterprise leaders should now take the opportunity to understand whether their current customer-facing systems can accommodate this new way of receiving requests.

Knowing Who the Agent Represents

When a person visits a website and signs into an account, the business can use its existing authentication processes to identify the customer.

An AI agent introduces another layer. The business may need to establish two things: whether the agent is legitimate and which customer it represents.

A recognised agent might be permitted to browse products but not to make purchases. It might be authorised to buy an item up to a certain price but cannot exceed a certain limit or substitute for a more expensive product if the preferred option isn’t available.

Businesses therefore need to distinguish between the following three things:

• Agent identity: Is the software making the request recognised and trusted?
• Customer identity: Which customer is the agent acting for?
• Permission: What actions has that customer authorised?

Businesses may use different methods to recognise AI agents and establish who they are representing in the transaction. One example of this is Visa's Trusted Agent Protocol. It’s been designed to describe how merchants can verify the signed messages associated with an agent’s activity as well as receive information that helps identify the customer.

However, adoption of this protocol is optional. It should be noted that recognising an agent doesn’t automatically prove that a specific transaction has been authorised, so businesses must still check the permissions associated with each request.

The point isn’t to replace every existing identity system currently in place. It’s to ensure those systems can support legitimate customer interactions when an agent acts as an intermediary.

Without that distinction, a business could mistake a legitimate agent for unwanted automated traffic or accept a request without establishing whether the agent has the necessary authority.

Permission Matters as Much as Identity

Identifying an agent doesn’t automatically authorise it to carry out every action available to the customer it represents.

If someone asks an AI agent to compare insurance policies, that doesn’t necessarily give it permission to purchase a policy, change existing cover or submit a claim.

The same principle applies to retail purchases. Permission to find a product isn’t equivalent to the permission to buy it. Also, approval for one transaction doesn’t automatically extend to a different purchase. This is where transaction authorisation becomes important.

Google's Agent Payments Protocol (AP2) addresses what needs to happen to record and verify payment intent and authorisation. Its specification describes mechanisms such as payment mandates and supporting records that connect transactions to the authority granted by the customer.

The broader lesson is quite straightforward: businesses need to know not only who initiated a request, but also whether the customer approved the action and agreed to its terms.

This could involve spending limits, restrictions on particular products or services, approval requirements or other conditions. If an agent attempts to act outside those limits, the business should be able to request further approval or reject the transaction.

These controls are especially important when an action involves a financial commitment, changes to an account or other decisions with lasting consequences.

Agentic commerce should make customer interactions more convenient. It shouldn’t turn limited permission into unrestricted authority.

Existing Enterprise Systems Still Do the Work

Even though agentic commerce changes how customers reach a business, it doesn’t remove the systems or infrastructure needed to serve them.

A retailer still needs accurate product information, current prices, reliable stock records, payment processing and order management. If an agent receives incorrect information, it may make a poor decision or attempt to place an order the business simply cannot fulfil.

For example, an agent could select a product based on the customer’s requirements but is relying on outdated pricing or availability records. The final order might exceed the customer’s budget or fail because the item is out of stock. Problems can arise from inaccurate enterprise data, an agent misinterpreting accurate information or making an incorrect selection. Reliable interactions depend on trustworthy data and systems that process requests correctly.

This makes integration a key factor to consider when assessing agent readiness. Businesses may have to modify their commerce APIs, product feeds and commerce platforms to enable approved agents to access accurate information and use the supported transaction processes.

UCP aims to simplify the connections between commerce platforms and various consumer-facing services. OpenAI's Agentic Commerce Protocol provides a more specific example of how a merchant can share product data, manage checkout sessions and communicate order updates.

Regardless, the merchant's underlying, existing systems still remain important. For example, the merchant is still in control of order validation and payment processing when using OpenAI's documented checkout flow.

Enterprises don’t necessarily need to replace their existing infrastructure to support agentic commerce. Instead, they need to understand which capabilities are already available to them, which ones require changes and how important information can efficiently move between the systems involved.

At the end of the day, an agent can only act effectively based on the information it receives so accurate data and reliable integration are essential components.

Keeping Transactions Accountable

When an agent acts for a customer, several parties may be involved: the customer, the agent provider, the merchant and the payment provider. If a transaction goes wrong, the business needs to establish what link in the chain caused the issue.

Did the customer authorise the purchase? What information did the agent receive? Did the transaction stay within the agreed limits? Was the order processed correctly? Keeping clear records helps businesses investigate problems, respond to customer complaints and resolve disputes.

AP2's approach to recording payment intent and supporting transaction evidence is one example of how emerging protocols can help establish what was authorised.

However, technical records don’t automatically settle every question of legal responsibility or replace applicable consumer protection, payment or data protection requirements. Businesses still have the responsibility to understand their own responsibilities and ensure customers can obtain help when an agent makes a mistake or a transaction doesn’t meet expectations.

This also affects customer support. If a customer contacts the business directly about an order or service request initiated by an AI agent then support teams need access to the relevant history of interactions. This history should include what the customer asked the agent to do on their behalf, what information it used, what permissions were granted and what action was taken.

Without all of the context and background information, the customer may have to explain the situation from scratch while staff may struggle to identify where the problem occurred. Businesses should make these records available to authorised support teams so they can properly investigate issues, correct errors and explain outcomes. The business must be able to trace all machine-initiated customer interactions and activity back to the relevant customer and service.

The objective is to preserve accountability even when the customer is not directly operating the interface.

How Can Enterprises Prepare for Agentic Commerce?

Businesses don’t need to suddenly assume that every customer will soon use an AI agent or adopt every emerging protocol. It’s still a work in progress. But, they can start by assessing whether their existing systems have the ability to support agent-initiated requests. This can be done by asking questions like:

• Can we recognise legitimate agents? See whether existing systems can distinguish approved AI agents from normal automated traffic and suspicious requests.
• Can we identify the customer behind an agent? Establish how an agent can be linked to the correct customer account.
• Can we enforce permissions? Determine whether systems can distinguish between browsing, retrieving information, preparing an order and completing a transaction.
• Is our data reliable? Check whether product details, prices, availability and service information are accurate across the systems that supply them.
• Can we trace and resolve transactions? Ensure the business can pull relevant records, investigate failures, and explain results to customers.
• So where does human approval fit in? Mark actions that require explicit confirmation, further checks or human review before performing them.

The answers will be determined by the organisation, its existing technology and the type of customer interactions it wants to support. Some companies may be able to adapt existing systems, others will need new integrations or added capabilities.

The aim isn’t to automate every interaction. It’s to establish where agent-mediated services can be useful and how to properly support them without losing control of the customer relationship.

Keep the Customer Relationship Intact

Agentic commerce introduces a different way for customers to interact with businesses. An AI agent may find a product, request a service or initiate a transaction, but the business must still know which customer’s interests are being represented and what that customer authorised in the first place.

That requires more than a transaction record. Businesses need to connect agent activity with the customer’s permissions, order or service history and support processes, while making it clear how customers can get help and correct mistakes.

The challenge ultimately is to ensure that customers receive consistent, reliable service no matter if their journey started by interacting directly or through an AI agent, with clear accountability for actions taken throughout.

As customer interactions increasingly involve AI agents, keep on top of the opportunities and challenges that come along with it through EM360Tech’s exploration of emerging technologies.