When AI agents act on behalf of customers: How customer service must prepare

What Does It Mean When an AI Agent Acts on Behalf of a Customer?

An AI agent acting on behalf of a customer can initiate, manage, or complete part of a buying or service process for a person, household, or organization. It can understand a goal and collect the information needed to pursue it. With the right access and authority, it can communicate with companies, request service, trigger workflows, or complete transactions.

The customer is usually still human. The AI agent becomes a representative that can carry out some of the work the customer would otherwise do personally. This is different from the AI a company uses inside its own customer service operation. 

The idea is related to what Gartner calls a machine customer, although the terms are not quite interchangeable. They define machine customers as “nonhuman economic actors that purchase goods or services on behalf of people or organizations.” 

A connected car scheduling its own maintenance fits easily into the machine-customer category. A customer’s AI agent contacting a provider on someone’s behalf is closer to a representative carrying out part of the customer journey.

What makes these agents different from simpler forms of automation is their ability to interpret intent and work within boundaries set by the customer. 

This introduces a new participant into the customer journey. The next shift is from human-driven journeys to increasingly agent-mediated journeys.

How Are AI Agents Already Changing Customer Service? 

Customers are already using AI assistants before they reach a company’s own service channels. Tools such as ChatGPT, Gemini, or Claude can research an issue and make sense of information that once required moving across several websites. They can also compare options against the customer’s own circumstances. Increasingly, the question becomes more personal: “What is the best option for me?” 

Customer service is beginning to see the same change. Gartner reported in July 2026 that customers are approximately three times more likely to use third-party generative AI tools than company-provided chatbots during service interactions. Among customers who use GenAI, 58% said they had used it to complete a task on their behalf. In B2B environments, the figure rises to 74%.

For customer service, the important distinction is between a customer using AI to understand an issue and authorizing AI to pursue its resolution. A customer may research a charge with an AI assistant and then ask it to act on their behalf by contacting the provider. At that point, the service organization has to understand both the request and the authority behind it. What has the customer asked the agent to do? Which decisions still require the customer’s involvement?

How Do AI Agents Acting for Customers Change the Customer Relationship?

As humans, we are always striving for convenience. Most of us enjoy buying things for our hobbies or presents for our children. We are less enthusiastic about replenishing household supplies, waiting in a contact center queue, disputing a charge, or arranging a product replacement. AI agents acting on behalf of customers could relieve people of more of that repetitive work.

That convenience also changes the relationship between companies and customers. As people delegate routine interactions, companies may increasingly communicate with an AI agent rather than directly with the person they serve. The agent can become a gatekeeper between the company and its customer.

Companies wanting to stay in direct touch with their customers will need to offer meaningful experiences that people want to participate in themselves. In my view, those experiences should focus on human needs for knowledge, achievement, gratification, and self-actualization. A relationship built only around transactions gives customers fewer reasons to engage personally.

What Changes When Customer Service Interacts with an AI Agent?

This forces us to rethink some familiar CX questions. How do we organize customer service for an AI agent? What does a good experience mean when the participant completing the task is a machine acting for a person?

Traditional customer experience is designed around human perception. It includes simplicity, empathy, tone of voice, visual design, emotional reassurance, trust, speed, and the feeling a person has after an interaction.

Machines do not experience brands in that way. An autonomous AI agent does not care about beautiful design. It does not respond to emotional advertising. It is not impressed by storytelling. It does not reward a company for sounding friendly if the data is incomplete, the API is unreliable, or the policy is ambiguous.

For an AI agent, good customer experience means having accurate data, structured information, clear rules, machine-readable product specifications, transparent pricing, real-time availability, reliable APIs, low latency, strong identity verification, clear authorization, predictable workflows, explainable decisions, secure access, good error handling, and auditable outcomes.

For a human, poor CX may mean having to repeat the same issue multiple times. For an autonomous agent, poor CX may mean that pricing cannot be programmatically calculated, terms are hidden in PDF documents, service status is not exposed through a reliable interface, product data is outdated, authentication is unclear, or there is no structured way to escalate an exception.

This means customer experience must expand. The key question is no longer only, “Was the customer satisfied?” It also becomes, “Could the customer’s authorized agent successfully, safely, and accurately complete the customer’s intent?”

That is a new category of CX.

Although an AI agent does not experience empathy or emotional reassurance as a human customer does, the person behind it still may.Companies need to support the agent’s task without losing sight of the customer’s needs and preferences.

In my view, transparency, security, reliability, consistency, and predictability become especially important in these interactions. The agent needs usable information about products, services, policies, and the conditions attached to an action. Companies should examine whether that information is sufficiently detailed and structured for an AI agent to work with.

Journey mapping remains essential. Companies still need to understand what the customer is trying to accomplish and remove unnecessary friction. The difference is that part of the journey may now be carried out by the customer’s AI agent. Mapping should include where the agent gathers information, requests an action, encounters an exception, and returns to the customer for a decision.

The touchpoints deserve attention, too. Companies should examine how customers’ AI agents can interact with their service channels and where conversational interfaces or appropriately controlled APIs may be needed.

What Do Enterprises Underestimate About Autonomous AI? 

The first thing I think enterprise teams underestimate is authority. More precisely, delegated authority under uncertainty. Identity and integration are difficult enterprise problems, and data introduces its own challenges. Companies have years of experience dealing with all of them. Autonomy creates a different question. What has the customer authorized the AI agent to do, when its understanding may be incomplete or wrong? 

Delegated authority under uncertainty

An AI agent may understand most of a customer’s request and still misunderstand an important detail. Permission to communicate on someone’s behalf does not necessarily include permission to make every decision involved in resolving the issue. 

Exception handling and reversibility

The second issue I see underestimated is exception handling and reversibility. Enterprises tend to design the successful journey. Much less attention may go to what happens when the AI acts on outdated information or misunderstands an ambiguous request. Conflicting policies can create another problem. A workflow can also fail after the AI has already completed several earlier steps.

This is where autonomous AI becomes much more complicated than producing the right answer. Once AI begins taking action, enterprises have to design for the unsuccessful journey too.

Why Can a Simple Customer Request Become Complicated for AI? 

Imagine a customer authorizes an AI agent to contact their broadband provider and request cancellation. The AI agent brings the request; the provider remains responsible for applying its verification requirements, business rules, and cancellation process. 

A good example is, “Please cancel my broadband service.”

Before anything happens, the broadband provider has to establish whether the customer’s AI agent is authorized to request cancellation and exactly which service should be cancelled. Broadband may be part of a bundle with mobile or television service. Removing one service may affect pricing or discounts on the others. Contractual fees, notice periods, or regulatory requirements may also apply. The AI agent may also need answers to questions the customer never mentioned in the original request:

• Is there an outstanding order or engineer appointment?
• Does the customer need to retain a telephone number?
• Are there unpaid charges or pending payments?
• Does equipment need to be returned?
• Is a retention offer allowed in this situation?
• Is the cancellation connected to an unresolved service fault or financial difficulty?

The process may then span CRM, billing, provisioning, payment, and logistics systems. Those systems have to stay synchronized as the cancellation moves forward. 

Now imagine the broadband provider’s cancellation workflow successfully completes seven steps and fails on the eighth. The broadband service could be disconnected while billing continues. Another service in the bundle could be affected by mistake. 

This is one of the main challenges I see with autonomous AI services. Language understanding is the easier part. The difficult work is executing a business process while respecting identity, policy, and customer rights. 

The customer’s AI agent needs to understand the conditions attached to the request. The provider remains responsible for applying its business rules and completing the cancellation correctly. A request arriving through AI does not remove those obligations. 

How Should Customer Service Handle an AI Agent’s Delegated Authority?  

Customer service needs to distinguish what a customer’s AI agent can request from what the enterprise can authorize and execute.

The customer’s instructions may set conditions around price, timing, or the changes the agent can accept. The enterprise has its own requirements around account access, approval limits, business rules, and customer rights. Completing the journey requires respecting both.

When a request moves beyond the authority established for the agent, the process needs a clear point for returning to the customer. An AI agent’s ability to communicate or supply account information should not be treated as unlimited authority to act.

Where Will AI Agents Acting for Customers Appear First? 

I think retail and e-commerce will see AI agents acting for customers first at a larger scale because many purchases and post-purchase service tasks are routine and relatively easy to complete. Product catalogues, prices, availability, and delivery options are already largely structured. Outcomes are easy to verify because the correct product was either ordered at the correct price or it was not. 

Many mistakes can also be reversed through cancellation, return, or refund. Customers have a strong reason to delegate repetitive work such as replenishment, price comparison, or order tracking. 

The first interactions will probably not be spectacular examples of full autonomy. They will be ordinary requests. A customer could tell an AI agent:

“Find my usual cat food, buy it if the total price is below 40 EUR, arrange delivery before Friday, and ask me only if those conditions cannot be met.”

The same agent could later check the shipment or change the delivery date. 

I believe telecom, travel, and financial services will follow. Their journeys are more complex, and the customer effort being delegated to AI agents is much greater, so I expect adoption to move more slowly. 

What Do AI Agents Mean for Enterprise Customer Experience Architecture? 

Readiness requires more than an AI chatbot. An enterprise needs to connect customer intent and context with the data, business rules, workflows, and backend systems required to actually complete the work. The incoming AI agent acts on behalf of the customer. The workflows it requests access to remain under enterprise control. The architecture needs to support that boundary while carrying the request through the systems and people needed to resolve it. 

Intent Alone is Not Enough

An AI agent may arrive with a very clear goal. The enterprise still needs to know which information is relevant and which workflow should be triggered. It needs to know which systems are involved and what actions are permitted. There also needs to be a clear point where a human is brought into the loop.

Orchestration Becomes the Decisive Capability

When a customer’s AI agent requests an action, the enterprise needs to carry that request through its own controlled workflows. 

Avaya Infinity is a unified customer experience platform that brings systems, data, AI, and workflows into one enterprise-controlled environment. It provides an orchestration layer where customer context can stay with the interaction while enterprise systems are brought into the process as needed. 

A customer may contact the service organization with one request, while the information needed to resolve it sits somewhere else. Account details may be in CRM. Billing information may live in another system. 

The enterprise still needs reliable connections between its service workflows and systems such as CRM, billing, and knowledge management, regardless of how the customer’s request arrives. The orchestration layer connects those systems with the interaction. 

MCP provides a standard way for AI to access approved enterprise tools and context, rather than rebuilding those connections for every new use case. These capabilities support the enterprise’s handling of the request; they do not, by themselves, establish what the customer has authorized their AI agent to do. 

Human Expertise Remains Part of the Flow 

The future customer journey will still require people. Some processes will remain ambiguous. When a person takes over, the context of what has already happened needs to move with the interaction. This is the idea behind tandem care in Avaya Infinity. AI and human agents operate as part of one continuous experience, with context preserved as responsibility moves between them. 

How Should Customer Service Prepare for AI Agents Acting on Behalf of Customers? 

I would begin with a specific service journey rather than assume every incoming customer AI agent should be able to complete every task. 

• Map the agent-mediated journey. Identify routine tasks customers may delegate and examine how their AI agents would move through the available service touchpoints. Look for friction in obtaining information, requesting actions, and resolving exceptions.
• Review information and access points. Examine whether product details, service policies, and transaction conditions are clear and usable by AI agents. Consider which existing channels can support these requests and where additional interfaces may be needed.
• Define the request and its limits. Identify what a customer’s AI agent can request, what authority needs to be established, and which decisions require the customer’s direct involvement.
• Examine the complete workflow. The broadband cancellation example shows why a simple request can involve several systems, contractual conditions, and consequences for other services. Those dependencies need to remain explicit when the request arrives through AI.
• Design for exceptions. Determine what happens when information is ambiguous or a process fails after earlier steps have changed the account. Reversibility and human involvement belong in the design.
• Preserve context when people take over. The customer or employee should be able to understand what was requested, what has already happened, and what remains unresolved.

AI agents can relieve customers of boring and repetitive tasks, leaving people more time for experiences they actually enjoy. Companies should start by identifying the services customers are likely to delegate. Those wanting to stay in direct touch with people will also need to offer something beyond a transaction: meaningful experiences that customers want to take part in themselves. 

Frequently Asked Questions About AI Agents and Customer Service 

What is an AI agent acting on behalf of a customer?

An AI agent acting on behalf of a customer can initiate, manage, or complete part of a buying or service process for a person, household, or organization. With the right access and authority, it can communicate with companies, request service, trigger workflows, or complete transactions. The customer is usually still human, while the AI agent carries out part of the journey on that customer’s behalf. 

Is an AI agent the same as a machine customer?

The concepts overlap. Gartner defines machine customers as nonhuman economic actors that purchase goods or services on behalf of people or organizations. A customer’s AI agent may perform that purchasing role, but it may also gather information or pursue a service request without making a purchase. 

What should a company establish before acting on a customer’s AI agent’s request? 

The company needs to establish whose behalf the AI agent is acting on and what the customer has authorized it to request or accept. That delegation must be considered alongside the company’s account-verification requirements, business rules, and approval limits. Permission to communicate on a customer’s behalf should not be treated as unlimited authority to change an account or complete a transaction. Requests beyond the established scope need direct customer confirmation. 

What happens when a service workflow initiated by a customer’s AI agent fails partway through a task?

The enterprise needs to understand what has already changed and what still needs to happen. A workflow may fail after several earlier steps have already been completed, which is why exception handling and reversibility need to be designed into the journey. Once AI begins taking action, enterprises also need to design for the unsuccessful journey. The service team should be able to identify completed actions, determine what can be reversed, and explain the unresolved steps to the customer or their AI agent. 

When should customer service involve a human in an AI-mediated request?

A human should become involved when the customer’s instructions or the AI agent’s authority are unclear, the request exceeds established permissions, or an exception requires additional judgment. Direct customer involvement may also be needed to confirm a decision. Whoever takes over needs the original request, the actions already completed, and the issue that remains unresolved. 

How should customer service organizations prepare for AI agents acting on behalf of customers?

Identify routine tasks customers are likely to delegate and map how their AI agents would interact with service channels. Review whether the information and touchpoints support those journeys. 

Start with a specific service journey and define which requests an AI agent can make, how delegated authority will be established, and which decisions require direct customer confirmation. Map the systems and business rules needed to complete the request. Design exception handling and human escalation into the workflow, preserving context when responsibility moves to a person.