AI agents are moving quickly from experimentation into customer experience (CX) operations, supporting service automation, agent assistance, proactive engagement and workflow orchestration.

That shift creates significant opportunity to improve customer experiences while increasing efficiency. It also means organizations need answers not just to the question of what AI agents can do, but also what they are allowed to do.

This question is central to the new World Economic Forum (WEF) playbook, “AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling.” Genesys contributed to WEF’s Frontier AI and Capabilities workstream within WEF’s Center for AI Excellence, whose new report series explores how organizations can define, assess and responsibly scale AI agents.

For CX leaders, the issue is especially urgent. Customer experience has long been a proving ground for AI because the use cases are immediate, measurable and closely tied to business outcomes. Agentic AI introduces systems that can reason across context, use tools, coordinate multiple steps and take action across workflows.

That matters because customer journeys rarely happen in one system. A single issue may involve customer history, billing data, policy rules, escalation paths and follow-up actions across multiple systems and teams. Agentic AI can connect those pieces more intelligently — but once AI agents can act across the journey, raising important questions like: What should agents be allowed to do? Under what conditions? With what oversight?

 

Why Agentic AI Changes CX

Agentic AI is especially relevant to CX because customer journeys typically depend on context, timing and action across multiple systems. When those systems are disconnected, customers feel the friction — repeating information, waiting through handoffs or wondering whether the company will follow through. AI agents can help close that gap by connecting context and taking action across the journey.

For instance, a service-focused agent could gather context, check policy and prepare a recommended next step for a billing issue. An agent-assist experience could summarize customer history and prompt escalation when sentiment changes. A proactive engagement agent could identify a customer at risk, while a workflow orchestration agent could coordinate follow-up across CRM, contact center, billing and field service systems.

The common thread is that agentic AI is not just answering questions; it is helping work move across the enterprise. That’s what makes it so promising for CX: It can help organizations respond faster, reduce friction and deliver more connected experiences while giving employees more time for empathy, judgment and expertise.

But as AI agents move from limited pilots into live CX environments, their actions can have real-world consequences. They may interact with sensitive customer data, influence customer outcomes or trigger financial, operational, compliance or reputational risk. An agent might recommend a refund, escalate a complaint, update an account or send proactive customer communications.

That is why scaling agentic AI requires more than confidence in the model. It requires a clear methodology for defining what an agent can do independently, where human approval is required, when escalation should happen and how agent behavior should be monitored over time.

 

ACAP: From Capability to Authorization

The WEF playbook introduces the Agent Capability and Authorization Profile, or ACAP, as a practical way to close the gap between what an agent can do and what it is authorized to do.

In simple terms, an ACAP is a living authorization record for an AI agent. It defines the agent’s role, the workflow it supports, the systems and data it can access, the actions it can and cannot take and the points where human oversight or approval is required.

For CX leaders, that structure is essential. The question is not simply whether an agent can complete a task; it’s whether it can complete that task safely, consistently and within clearly defined authority.

In practice, an ACAP-style approach can help CX organizations answer questions such as: Can this agent only recommend a refund, or can it issue one? Can it update a customer record, or only retrieve information from it? Can it send a proactive message, or does a human need to approve it first? When should it escalate, and how will its behavior be monitored over time?

This becomes more important as organizations deploy multiple AI agents across journeys, channels and functions. A virtual agent supporting customer service, an internal agent assisting employees and an orchestration agent coordinating back-office workflows may all rely on shared data, tools or models. But each deployment may require different permissions, oversight and escalation rules.

In CX, context matters. The same capability can carry different levels of risk depending on the journey, geography, data involved and business action being taken. ACAP helps make those differences explicit.

 

What Governed Agentic AI Looks Like in CX

Governed agentic AI does not necessarily mean slowing innovation. Done well, governance can create the confidence organizations need to scale. It helps accelerate innovation, rather than slowing it down.

That starts with focused use cases where value is clear and risk is manageable. Early deployments should build operational confidence while keeping exposure contained, often beginning with actions that are lower risk, reversible or subject to human review before expanding autonomy.

It also means designing human oversight into the operating model from the beginning. “Human in the loop” only works if the loop is meaningful. In high-volume CX environments, oversight can become a cursory, check-the-box exercise if supervisors are asked to review too many actions, too quickly or without enough context.

For customer-facing workflows, meaningful oversight often includes clear escalation paths, confidence thresholds, audit trails and human review for consequential moments such as financial decisions, sensitive complaints, account changes, policy exceptions or emotionally charged interactions.

Monitoring also needs to continue after launch. CX leaders need to understand how agents are performing, where they are escalating, when customers are dissatisfied, whether behavior is drifting and whether the agent is staying within its authorized boundaries.

At Genesys, we think the future of CX is not a choice between automation and human service. It is a coordinated model in which AI agents help orchestrate work while people remain accountable for trust, empathy, judgment and business outcomes.

 

Scaling Agentic AI Responsibly

As agentic AI starts to become part of CX operations, organizations need more than standalone AI tools. They need an approach that connects customer context, workflow automation, knowledge, analytics, human oversight and operational controls — enabling AI to act with context, but also within boundaries.

We see this as central to the next phase of AI-powered customer experience. The goal is not simply to automate more interactions. It is to help organizations orchestrate better experiences across customers, employees and systems — with governance, observability and human judgment built into the way AI operates.

That is also the direction reflected in innovations such as the Agentic Virtual Agent capability of Genesys Cloud™. As agentic AI becomes more embedded in CX workflows, organizations need solutions that combine intelligence with context, orchestration and control. The promise is not just that AI agents can take action; it is that they can help CX operations become more adaptive, proactive and connected while preserving customer trust.

For CX leaders, the path forward is to define authority clearly, sequence adoption carefully, keep humans involved where consequences are high and expand autonomy based on evidence.

The organizations best positioned to succeed will treat governance as an accelerator, not a constraint. By defining what AI agents are allowed to do and how their actions are supervised, CX leaders can move from experimentation to production with greater confidence — and create experiences that feel more connected, responsive and human.

 

Read the World Economic Forum playbook, “AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling,” to learn more about the ACAP framework and how organizations can prepare for trusted agentic AI at scale.