AI agents change the risk profile of customer service when they move from supporting employees to taking action across enterprise systems.
A summarization tool can help an agent work faster. A workflow agent that accesses customer data, applies a discount rule, triggers an approval, or changes a service process carries a different level of responsibility.
For Daniel Wilks, Vice President of ServiceNow CRM EMEA, the enterprise decision is no longer only where AI can help. CX, IT, operations, and security leaders also need to decide where AI agents can act, what they can access, and when human authority returns. Wilks framed the shift as a move from assistance to operational accountability:
“AI is no longer simply summarising information or recommending a next step; it can access customer data, interact with different systems and take action on a customer’s behalf.”
That creates a practical governance question for CX leaders. Faster automation can reduce delay, but uncontrolled action can create new customer, compliance, and operational risks.
When AI Moves From Assistance To Action
Enterprise AI has often started with lower-risk support tasks: case summaries, knowledge retrieval, response drafting, and next-best-action prompts.
Those use cases still leave a person responsible for the final decision. The risk changes when an AI agent begins to initiate work across CRM, billing, inventory, Configure, Price, Quote, service management, or fulfillment systems.
A customer may experience the result as a faster quote, a case update, a scheduled repair, or a changed order. Internal teams need to understand how the decision was made, which system supplied the data, and whether the action followed approved rules. Wilks argued that visibility becomes central once AI starts acting across workflows:
“Organizations need much greater visibility into what AI is doing, what it can access and who is accountable for its actions.”
That visibility cannot sit after the event as a reporting exercise. Agentic AI needs an operating model that defines access, records activity, and gives teams a way to intervene before a poor decision reaches the customer.
Customers may not see the governance layer, but they will feel the outcome if a workflow acts on incomplete context.
The Controls AI Agents Need Before They Scale
Wilks identified three foundations for trusted AI agents: visibility, clear boundaries, and human intervention.
Visibility gives teams an audit trail of agent activity. Boundaries define the data, systems, actions, and decisions available to the agent. Human intervention points return authority to a person when risk, judgment, or exception handling requires it.
Asked what needs to be in place before enterprises scale AI agents, Wilks put the control model simply:
“Three foundations are essential: visibility, clear boundaries and human intervention.”
Those controls become more important when customer work crosses functions.
A service workflow may begin in CRM, check an entitlement, review inventory, trigger a field-service task, request an approval, and notify the customer. Each step needs permissioned access and clear limits.
Poorly governed automation can reproduce the same fragmentation that CX teams are trying to remove. An AI agent working from incomplete data may act too early. An agent with broad access but weak boundaries may perform actions that should require approval. Wilks warned that autonomy needs controls that match the level of action: