Jason Leung grew up in the South, where hospitality is a way of life. Now a director of product management at ServiceNow, he's spent years building products used by people all over the world. He brings his Southern hospitality to every situation.
"Whether I'm traveling the world or building global products, I always try to make experiences feel warm, approachable, and human," he says. In an industry that often prizes speed and scale above all else, that's a quietly radical commitment.
It’s also shaped his approach to one of the most reliably frustrating experiences in modern life—calling a help desk—and led to his development of an AI voice agent.
Most people know the feeling. You dial in with a straightforward question and end up navigating a phone tree, waiting on hold, and explaining your issue two or three times before reaching someone who can actually help. When you finally do, they're often fielding the same questions they've answered dozens of times that day—a poor experience on both ends of the line.
The underlying problem is structural. Even sophisticated help desks are built around routing and retrieval: Get the caller to the right person, pull up the right record, resolve the ticket. They don't adapt to how people actually talk or the subtle variations in how someone might describe the same problem. And they scale poorly; more volume means more agents, more time on hold, more friction.
For employees calling in about IT issues, HR questions, or account problems, that friction adds up. And for the agents on the other end, the repetition of handling the same uncomplicated requests day after day is its own kind of drain. That’s time that could be spent on the complex, nuanced problems that require a human.
When agentic AI arrived, Jason saw an opening. "I knew we could do better," he says. The challenge was proving it.
Not everyone was convinced the technology was ready. Skepticism about AI voice agents runs deep. Anyone who’s yelled at an automated phone system knows the gap between what these tools promise and what they deliver. Jason's bosses wanted evidence, not assurances. So he did what any good engineer does when faced with doubt: He built a prototype.
The pivotal moment came early in testing. A customer called in and simply started talking—naturally, conversationally, the way you'd speak to a person. The issue was handled. The conversation flowed. And at the end, the customer paused and asked: "Wait, was that a bot?"
That question was the answer. It meant the interaction had felt human enough that the customer hadn’t thought to ask until it was over.
Jason is quick to point out what his AI voice agent is not. "This isn't just a phone tree with better branding," he says. The agent can troubleshoot technical issues, place orders, pull up a caller's paid time off balance, and walk someone through a multistep process—all in a single, continuous conversation. It understands context, handles follow-up questions, and doesn't lose the thread when a caller changes direction mid-sentence.
What makes the AI agent genuinely useful, rather than just impressive in a demo, is how it handles the full texture of real human communication. People don't call help desks and recite their problems in clean, structured sentences. They start in the middle. They use shorthand. They say that "the thing keeps freezing" instead of naming the application.
A traditional system—even a well-designed one—tends to falter at exactly these moments, forcing the caller to start over and rephrase until their words fit the system's expectations.
Jason's AI agent is built to meet people where they are. And when a request falls outside what it's experienced, it flags the gap and learns from it. Over time, the agent develops a richer model.
Jason compares the process to dialing in a great espresso shot. The mechanics look simple from the outside: grind, tamp, pull. But the difference between a good shot and a great one comes down to dozens of small, precise adjustments to certain parameters—grind size, water temperature, extraction time, pressure—that compound into something noticeably better. Get any one of them wrong, and the whole thing suffers.
Get them all right, and the result feels almost effortless. "Small adjustments can completely change the outcome," he says. "That's true for espresso, and it's true for AI."
Jason’s building toward an AI agent that treats callers the way he learned to treat people while growing up in the South: with warmth, patience, and genuine attention to what they need.
As he says, "This is what happens when you take human warmth seriously and let AI do the heavy lifting.”
Find out how ServiceNow can help you put AI to work for people.
