Customer experience must serve AI agents and people

Customer experience must also serve agents acting for customers

A better customer experience could mean the customer never uses the interface a business spent years improving. If a customer delegates a purchase or booking to an AI agent, the business must be understandable and accessible to that agent. Its screens still matter to people, but winning business could also depend on supporting software that evaluates offers and completes authorized transactions.

Delegating that work requires an AI agent: a system that uses models, tools and information to pursue a task on someone’s behalf. A customer agent represents the person seeking a product or service; an enterprise agent acts for the business providing it. These agents could interact directly as AI moves from generating information and recommendations to reasoning, coordinating work and taking action. Customer experience, or CX, would then extend to how those interactions support commerce, personalization, loyalty, search and interface design.

A delegated task can bypass the journey a business designed

Delegation changes the customer’s role from operating a process to specifying an outcome. Generative AI lets people interact with information through natural language, while agentic systems connect AI models to tools, data and workflows so they can act on an objective. The difference is the responsibility the system accepts: a conversational response can help someone decide what to do, while delegated execution can carry out the resulting task.

Business travel makes this change concrete because completing one trip can require several separate systems. A traveler may move among an airline app, a hotel website, loyalty programs, a calendar and a corporate booking platform. Each handles part of the task. The traveler must connect the information and decisions, so even when better interfaces reduce effort, coordinating the trip remains their job.

In a hypothetical delegated journey, the traveler gives an agent a single instruction: “Get me to Chicago for my meetings next week. Stay within policy, use my preferred airline when it makes sense and book a hotel close to the office.” Before acting, the agent would need to interpret the destination, meeting timing, company policy, airline preference and hotel-location requirement. Those constraints would guide how it evaluates schedules and prices alongside the traveler’s preferences and loyalty information.

To turn that evaluation into a booking, the customer’s agent could query enterprise agents for availability, compare options and potentially negotiate within defined parameters. It could complete the transaction once an option met the relevant constraints. The traveler would avoid the usual sequence of apps, websites and booking platforms, even though the business would still have to support every underlying decision and action.

The design objective therefore becomes successful, authorized completion of the task. Teams can assess existing customer journeys by identifying which an agent could potentially complete from the initial objective through to a transaction, then mapping the information and actions each requires. For the Chicago trip, this would expose the connections the traveler currently makes between separate systems. Better navigation would continue to help people using those systems directly, while delegated access would need to support the same task through software.

Agents need usable offers and authorized transaction access

Supporting that task changes what a business must make accessible. Decades of CX work have focused on reducing effort for people, including whether navigation takes three clicks or five. An agent instead needs product information it can interpret, clear prices, current inventory and a secure way to act. Click counts reveal little about whether those capabilities exist.

Product information comes first because an agent must understand an offer before comparing it. Machine-readable information is organized so software can retrieve and interpret it, giving the agent a basis for judging whether a product or service meets the customer’s objective. Pricing lets the agent assess the option against the customer’s budget, while current inventory establishes whether it is available. Unclear or inaccessible inputs can therefore limit consideration before a transaction begins.

Because these inputs affect consideration, their quality belongs in CX investment decisions. An audit should examine whether an external agent can retrieve and interpret product details, pricing and inventory, and whether each is suitable for the decision being made. A website makes information available to a visitor, but software also needs a usable way to access it. Product data and product pages therefore need separate usability assessments.

Once an agent can evaluate the offer, it needs access to the business actions required to complete the task. An application programming interface, or API, gives software a defined way to request information or ask another system to perform an operation. In a delegated journey, APIs could let the customer’s agent check availability and complete an allowed transaction through enterprise systems. Their value to CX depends on whether they support the actions the customer’s objective requires.

Access to those actions also determines whether a delegated journey can finish. A business may offer an attractive human-facing experience and enough information for an agent to compare its offer, yet leave the agent unable to transact. The underlying systems could consequently become as important as the visible interface. CX and engineering teams need to assess the connection between understanding an offer and acting on it.

Allowing an agent to act raises another question: who is asking, and what may that requester do? Identity establishes which agent is interacting with the business, and authentication checks that identity. Authorization determines whether the agent has permission to perform the requested action for the customer. Recognizing an agent and accepting its authority are separate decisions, and both affect whether the task can proceed.

Permission establishes whether the transaction is legitimate, while trustworthy information supports the customer’s decision. Customers need confidence in the information their agents use to compare options, and enterprises need confidence that requested actions fall within the agents’ authority. Accurate data and verified permission serve these distinct needs. Interface design alone cannot establish either form of trust.

These trust requirements bring security and governance into journey design. Governance sets the rules under which agents may act, while audit policies define how their actions are recorded and reviewed. CX leaders should work with technology teams to define authentication, permissions and audit policies alongside the journey’s information and transaction requirements. As customers delegate manual steps, those controls become essential to making the remaining decisions and actions trustworthy.

The customer’s agent may bring context no single brand holds

With access and authority established, the next question is what information will guide the agent’s choice. Companies have spent years building customer profiles from transactions, browsing behavior, loyalty data and other signals. These help a business predict what someone may want within the relationship it can observe. A personal AI agent acting across an individual’s activities could bring a broader set of inputs to the decision.

With the customer’s permission, those inputs might include a calendar, budget, travel preferences, subscriptions and personal priorities. Together, they connect needs across parts of the customer’s life that an individual brand is unlikely to see in full. In the travel scenario, the agent could therefore understand requirements beyond the customer’s history with an airline or hotel. Delegating the task would still require specific permissions to access personal information.

Broader context could change the sequence of personalization because the agent could determine the customer’s needs before searching for a suitable business. Selection would begin with the customer’s own information, shared with permission, and each offer would be assessed against those needs. A business would need to make its offering clear and actionable enough for the agent to judge the fit. Its own predictions would remain useful within that wider decision process.

Company-held profiles could therefore retain value while their relative advantage changes. A brand’s detailed knowledge of its customer relationship differs from an agent’s potential understanding across the customer’s activities. The business can respond by making its offers straightforward for an external agent to interpret and use. It could then serve the customer’s needs while the customer’s agent holds the broader personal context.

Businesses must compete for agent selection as well as human attention

When an agent uses that context to choose which businesses to consider, discoverability becomes a machine-facing concern. People respond to brand, design, familiarity, advertising and habit. Agents could weigh price, availability, preferences, loyalty benefits and policies differently, evaluating these factors together. People also care about price, and an agent can account for an expressed brand preference, so the change concerns how these influences enter the decision.

A business therefore gains another selection process to design for. Its offering needs to be discoverable and credible to the software comparing options for the customer. Making offers, terms and loyalty benefits machine-discoverable gives the agent information for that comparison. Human attention remains commercially valuable, while delegated software adds another route through which the business may be understood and selected.

Loyalty benefits can influence this selection when the agent can interpret their value. An agent could weigh them alongside price, availability and policy while applying the customer’s preferences. The Chicago instruction allows that judgment by asking for the preferred airline when appropriate. A business therefore needs to express its loyalty benefits clearly enough for the agent to include them in the comparison.

Existing channels and delegated access will need to coexist

Designing for agent selection can proceed alongside continued investment in the channels customers already use. Websites should remain available, and mobile apps should continue serving their users. The technology and standards needed for a fully agent-to-agent economy are still developing. Preparation can therefore expand access while preserving familiar ways for customers to interact with the business.

Supporting both forms of access requires interoperability: the relevant systems and agents must be able to work together. Journey planning should bring CX and engineering teams together to examine how data architecture, APIs, identity, permissions and security support interactions across those systems. Can each connection let the task proceed under the customer’s authority? Where delegation is appropriate, teams need to establish which information and actions an authorized agent can access across the participating systems.

Key takeaways for leaders

• Design for delegated tasks: CX teams need to identify customer journeys an AI agent could complete from instruction to transaction. Map the information, decisions and actions required across systems to expose gaps in delegated access.
• Connect usable data with authorized actions: CX, engineering and security teams need to assess whether agents can interpret product data, retrieve current prices and inventory, and complete permitted transactions. Define authentication, permissions and audit policies alongside those capabilities.
• Prepare for customer-controlled context: A personal agent could use permissioned access to calendars, budgets and preferences to understand needs across a customer’s activities. Product and CX teams need to make offers clear enough for agents to assess against that broader context.
• Make commercial value machine-discoverable: Marketing and commerce teams need to express offers, terms and loyalty benefits in forms agents can retrieve and compare. Agent selection could depend on how well those benefits meet the customer’s preferences, constraints and objectives.
• Build delegated access alongside existing channels: Websites and apps remain important while agent-to-agent standards develop. CX and engineering teams need to coordinate APIs, data, identity and permissions so authorized agents can work across systems.