When a customer finally reaches out for help, are they meeting a shield or a bridge?
Most contact centers are answering that question by accident. The default move with AI has been deflection — keep volume away from live agents, lower the cost to serve, and count the tickets that never reached a human as a win. But deflection measures how many people you kept out. It says nothing about how many you actually helped. That gap, between AI that’s been deployed and AI that demonstrably improves the experience, is where the real strategy sits.
To figure out where autonomous AI belongs in a customer interaction — and where it absolutely doesn’t — we put the question to Chang Chang, Senior Director and Head of Product for Webex CX. His framing reorders the whole debate. The dividing line, he argues, isn’t the task. It’s the stakes.
The front door, not the back office
Ask Chang where autonomy earns its place, and he starts by widening the aperture. “While many autonomous AI agents are deployed in the high-volume, low-complexity interactions where customers want an immediate answer, we see a massive opportunity where autonomous AI acts as an AI Concierge, serving as the front door to a brand, handling intent recognition, routine fulfillment, and orchestration of data across systems to resolve customer needs.”
That’s a bigger role than the usual FAQ bot. The concierge isn’t there to make customers go away. It’s there to meet them, understand what they actually want, and pull the right data from the right systems to get it done.
But a front door still needs a lock. “Human supervision remains non-negotiable,” Chang says, “as the accountability of these autonomous AI agents ultimately still lies with the brand.” His sharpest line is also his simplest: “Autonomy without accountability is a liability.”
So how should a CX leader decide where the line sits? Not by sorting tasks into “AI can” and “AI can’t.” As he puts it, “the line isn’t drawn by the task, but by the stakes.” Drawing it well takes security, observability, and manageability strong enough to keep humans in the loop at the right moments across the entire agent lifecycle. The point isn’t to hover. It’s to be able to step in, in Chang’s words, “before a service failure becomes a brand failure.”
Stop counting deflections
Here’s the uncomfortable part for anyone reporting AI wins by ticket volume. Asked what evidence shows AI is raising CX rather than just cutting cost, Chang didn’t reach for a containment rate. He pointed somewhere else entirely.
“The evidence is found when you stop looking at deflection rates and start looking at resolution quality, customer effort, and customer sentiment,” he says. Then comes the metaphor that anchors his whole philosophy: “We are seeing a clear distinction between brands that use AI as a shield to keep customers away and those that use it as a bridge to get them to a resolution faster.”
The proof, in his telling, shows up in what he calls loyalty moments — the times AI anticipates a need or remembers a hyper-personalized detail from a past interaction and turns a touchpoint that could’ve been frustrating into one that resolves almost without effort. That’s a small thing to engineer and a hard thing to fake. It’s also the difference between a customer who tolerates your AI and one who’s quietly impressed by it.
The rising value of a human
If AI is absorbing the routine, what happens to the people? The intuition that human agents get less valuable turns out to be backwards.
“The ‘human premium’ is not just rising,” Chang says. “It is becoming the primary differentiator for premium brands.” As AI commoditizes the routine and the front door, the ability to handle complexity and emotion becomes the highest-value skill on the floor. He describes where this leads as Connected Intelligence: a hybrid workforce where AI agents and human agents work the same problem together.
That shift breaks the old playbook for staffing and training. You can’t keep drilling agents on rigid scripts for routine queries that AI already handles faster. Training has to pivot toward emotional intelligence, critical thinking, and the skill of managing AI-augmented work. Chang’s image for the new role is worth sitting with: human agents trained “as pilots of AI,” leaning on real-time guidance and suggested responses to handle the technical side of a call so they can put their full attention on the person on the other end.
The measurement has to move too. Average Handle Time — the metric that rewarded getting people off the phone — gives way to resolution quality and sentiment shift. And the staffing math changes in a way that’s genuinely new: AI agents offer something human teams never could, effectively infinite capacity, while human agents hold onto the high-touch, high-emotion work. “The goal,” Chang says, “is to empower them to deliver a high-impact, high-empathy experience that AI simply cannot replicate.”
Where the line sits
Chang’s argument lands on a quiet reversal. The contact center’s AI story has mostly been told as a cost story. He’s telling it as an experience story — one where autonomy handles the front door, accountability stays with the brand, and the human premium climbs precisely because the machines got good.
The brands that win this won’t be the ones that deployed AI first or deflected the most. They’ll be the ones who knew, interaction by interaction, whether they were building a shield or a bridge.
