As we explored in “Beyond CSAT and AHT: The New Metric Stack for AI-Driven CX”, AI is changing the shape of service work. But the pressure on human agents does not come from automation alone. More often, it comes from the underlying operational friction that AI is meant to address: disconnected systems, poor visibility, and the constant effort of piecing together context across multiple tools.
When those issues persist, even the best AI strategy will fall short. When they are addressed, AI can help create a better rhythm of work for agents by reducing manual effort, surfacing context faster, and making complex customer interactions easier to manage.
For Nuri Gocay, Director of Platform Architecture for Zendesk Contact Center the priority is clear:
“A recent survey by Insignia Resource found that 87% of call center agents report high workplace stress. 74% of them are considered an ongoing risk of burnout. The work is heavier because they no longer have that break in between calls where they’re just doing a simple password reset.”
AI-driven CX can remove repetition. But it can also remove the light-weight interactions that used to act as breathing room.
When AI Is Implemented Poorly, Stress Moves From Customers To Agents
Leaders often frame AI rollouts as a way to streamline workloads and workflows . That is true when AI is integrated, trustworthy, and measured against outcomes. It is less true when AI is bolted on and pushed into production without fixing the underlying service design.
In that case, customers still struggle with issue resolution, and they escalate to human agents. Agents then inherit the frustration, plus the complexity, plus the burden of cleaning up AI mistakes. Gocay warned: If AI is taking everything that’s easy… everything that’s reaching a human agent is complex. It’s nuanced. Maybe it’s a broken process that an agent will have to chase down.
That is a structural shift, and it changes how teams should think about staffing, coaching, and escalation.
Why ‘More AI Tools’ Can Increase Cognitive Load
Many CX teams assume adding an AI assistant reduces effort. It can, but only if it reduces fragmentation. If agents still have to jump between systems, AI becomes another tab and another mental task.
Gocay described this as the modern version of the swivel-chair problem, and the consequences show up as slower resolution, more mistakes, and higher burnout.
“If you’re implementing AI as a standalone tool, instead of deeply integrated… agents have to toggle between those things. They’re toggling between their CRM, their AI knowledge base, their ticketing system, their phone.”
As we explored in our previous article, “Beyond CSAT and AHT: The New Metric Stack for AI-Driven CX”, outdated measurement frameworks can make this problem worse. If leaders continue to overweight AHT while agents are forced into constant context switching, they risk penalising people for system design failures.
Knowledge Retrieval Is Still A Hidden Tax On Agents
Even strong teams underestimate how much time gets burned searching for answers. That search time often becomes hold time. It also becomes stress, because the customer is waiting while the agent tries to stitch together context.
Gocay pointed to a concrete benchmark that should be uncomfortable for most leaders. Knowledge retrieval adds an average of 2.7 minutes per call. So that’s the agent putting people on hold to go find things.
If leaders want always-on service without always-burned-out teams, the operating model must has to reduce that retrieval burden through integration, and through better knowledge hygiene.
The Most Common Failure: Data Blindness That Forces Agents To Start From Zero
There’s a moment in almost every support interaction that reveals whether the system is designed for customer progress. It is the first question.
From a CX perspective, the question sounds polite. From a systems perspective, it can signal that context has failed to travel. Looking ahead, Gocay argued:
“The phrase that we train all of our agents to say, ‘how can I help you,’ is actually one of the biggest insults to our data and systems and design processes. Because that means something has failed.”
This connects directly to our earlier Zendesk conversation, ‘From Reactive to Agentic: What an AI‑Native Contact Centre Actually Looks Like’. AI-native, agentic models depend on context, memory, and action. If your agents are still starting cold, you are not getting the benefit.
What Good Looks Like: Protecting Humans With Smarter Routing And Better Feedback Loops
AI-driven CX leaders are going to have to treat agent wellbeing as an operational metric, not a culture slide. That starts with workload design.
One practical lever is routing by cognitive load, not only by skill. The goal is to avoid stacking emotionally heavy interactions back-to-back, especially in an always-on environment.
Gocay framed the move as a practical shift:
“The concept of routing by cognitive load and not just by skill… Maybe they just got off three difficult calls in a row where the customer was swearing and screaming… Maybe it’s time that they get a password reset.”
Another lever is giving agents a simple way to flag low-quality AI answers. That creates ownership and continuous improvement, and it also reduces the feeling that AI is something happening to them instead of something they are shaping.
A Simple Step Leaders Can Take This Quarter
If a CX leader wants progress fast, start with one change that creates breathing room and builds trust.
Pick a single friction point that repeatedly forces agents into search and context switching. Fix the integration path, and then measure the effect on repeat contact, escalation, and agent sentiment. Keep it narrow enough to ship quickly, and broad enough to matter.
The deeper lesson is that AI-driven CX will be judged by whether customers make progress, and whether humans are still capable of delivering empathy at the moments that matter.
Gocay’s warning is blunt, and it is strategic. If empathy is the last meaningful human differentiator, leaders can’t afford to drain it from the system.
