For all the noise in the CX market, few organisations are solving the right problem, I’ve seen plenty of “next big things” come and go and right now, “Agentic AI” is the phrase on everyone’s lips. Every vendor claims to have it. Every board member wants to know when you’re deploying it. Yet, for many leaders I speak to, the term feels less like a strategy and more like a gamble.
If you are tired of the hype cycle, you are not alone. The reality of Agentic AI isn’t about flipping a switch and walking away. It is about understanding the nuance between helpful automation and dangerous overreach. To get to the bottom of this, I sat down with Jonathan Rosenberg, CTO at Five9. We stripped away the marketing gloss to explore what “agentic” really means when you put it in front of live customers.
The Dial of Trust: Defining Agentic AI
The disconnect between technical definitions and business reality is often where projects fail. Academics might define Agentic AI based on reasoning capabilities, but for a CX leader, the definition needs to be operational.
Rosenberg describes it simply. It is a system that works autonomously to handle consumer issues with minimal human intervention. But the key isn’t just autonomy. It is autonomy with guardrails:
“Our view is that there’s no one right answer here. What the right answer actually is depends on the goals of the business, depends on the risk-reward trade-off that brands have when building these systems.”
He introduces the concept of a “dial of trust.” On one end, you have full autonomy—giving an AI agent instructions and letting it figure out the steps, reason through problems, and invoke APIs on its own. On the other end, you have flow-based approaches. These are more scripted but still operate without human intervention. This concept should be adopted as a practical decision framework by CX leaders.
The goal isn’t to crank the dial to eleven immediately. It is to find the operating point where you achieve self-service goals without introducing unnecessary risk.
Start Simple or Risk the ‘Black Hole’
There is a temptation to throw AI at your most complex, expensive problems first. You look at that 30-minute phone call that costs you a fortune and think, “Let’s automate that.”
Rosenberg warns against this approach. When you give bleeding-edge models high autonomy on complex cases with little direction, they hallucinate. They make mistakes. They harm the customer experience.
Instead, the recipe for success is often boringly practical.
According to Rosenberg:
“One of the customers that we’ve seen a lot of success with is Wyndham… they had great success using our AI agents for some really basic use cases of password reset. It’s one of my favorite, like start here. And it’s still shocking the amounts of brands that still go to a human for a password reset.”
Wyndham automates 40,000 password resets monthly. They also automated 80% of booking cancellations with less than a 1% abandon rate. These aren’t glamorous use cases. But they are high-volume, low-risk, and they work.
Governance Day Zero: A Design Principle
If you treat governance as something to bolt on after you build the system, you are already too late. Rosenberg argues that control and transparency must exist from “day zero”—the design phase; and this is non-negotiable.
You need the capability to test prompts on your actual data before a single customer interacts with the AI. You need to tune those prompts until accuracy hits your benchmarks.
Once you deploy, the work doesn’t stop. You need real-time dashboards that monitor for hallucinations or prompt injection attacks.
“You have to have these dashboards and metrics that work in real time that can alert you when something looks like it’s going awry… This whole thing is a loop and it’s not something that can be edited afterwards. This has to be in your product from day one.”
Without this, you are effectively flying blind.
The Human Element: Copilot, Not Replacement
We have all seen the headlines about companies planning to replace their entire support staff with AI. Some, like Klarna, made big public splashes about this pivot. But for most brands, the “replace everyone” strategy is a trap.
Rosenberg notes that while AI handles the mundane—like the password resets no agent enjoys—it cannot replicate human empathy:
“AI can’t provide empathy, it just can’t, because it’s not a person. Only people have empathy. And so there are times you want that live connection, there are times you need it for business reasons.”
Sales is a prime example. You likely don’t want an AI handling the final negotiation of a high-value deal. You want the AI to filter the leads, qualify the opportunity, and then hand it to a human closer to seal the deal.
This brings us to the safest entry point for skeptical leaders: agent assistance. Tools like post-call summarization are “no-regret” moves. An AI listens to the call, drafts a summary, and the human agent reviews and approves it. It saves time, improves data quality, and keeps a human in the loop.
Awareness Is No Longer the Challenge. Alignment Is.
We often look for technology to solve our problems, but Agentic AI forces us to ask a harder question: How much do we trust our own processes? The technology is ready to act on our behalf. The real challenge for 2026 isn’t buying the AI; it is building the governance and the confidence to let it work. Are you ready to let go of the wheel, or do you just need a better navigator?
Get ahead of the next wave of CX transformation. Join Jonathan and other senior leaders from Five9 and NCO Europe to review 2025 and unpack the five trends that will shape EMEA contact centres in 2026.
