Talkdesk’s Munil Shah on Why “Humans vs. AI” Is the Wrong Fight

The Talkdesk product chief on agentic AI hype, why IVR is dying, why AHT is obsolete, and who’s accountable when AI gets a decision wrong.

Agentic AI has become the industry’s favorite word, claimed by nearly every customer experience vendor on the market. But the term’s ubiquity has made it harder, not easier, to tell where real transformation ends and marketing begins. Behind the buzz, a more consequential shift is underway: the systems that route, resolve, and escalate customer issues are being rebuilt from the ground up, and the metrics companies have relied on for decades to judge them no longer tell the whole story.

Few executives are closer to that shift than Munil Shah, Chief Product, Technology, and Customer Officer at Talkdesk, whose Customer Experience Automation platform sits at the center of the industry’s push toward autonomous resolution. In this conversation, Shah lays out why isolated AI agents aren’t enough, why he believes traditional interactive voice response is already obsolete, and why longtime CX benchmarks like average handle time may be actively misleading. He also addresses a harder question hanging over the entire industry: when an AI agent makes a wrong decision, who is actually responsible?

Full interview;

Every CX vendor claims to have agentic AI today. Has AI become the industry’s biggest buzzword — and how should buyers separate genuine innovation from clever marketing?

We can’t reduce agentic AI to only a buzzword. The technology is fundamentally changing customer experience (CX) by understanding context and nuance, coordinating multiple tasks, and helping resolve increasingly complex customer needs. 

The real differentiator, however, isn’t simply whether a vendor offers AI agents. Too many organizations still deploy isolated AI agents to solve individual problems across siloed parts of the business. The differentiator is whether agents can work together to understand the full scope and complexity of customer interactions and collaborate across departments and systems to resolve issues. The next stage of innovation is orchestrating specialized AI agents, enterprise systems, and human employees to automate entire customer journeys—from the moment an issue arises through resolution.

Of course, businesses also need to be confident in the AI agents they deploy within this more automated environment. That’s why Talkdesk considers the complete agent lifecycle in our Customer Experience Automation (CXA) platform. With the recently launched Agent Builder, along with evaluation and observability tooling, organizations can create, test, diagnose, and govern AI agents, helping enterprises protect trust while scaling automation confidently.

Talkdesk has predicted the death of IVR, yet most enterprises still run on legacy systems. What’s stopping the industry from moving on?

We’ve all groaned on the phone when we hear “Press 1 for…” Traditional IVR was designed to route customers to the right person. Customer Experience Automation changes that model by resolving many issues without even requiring a transfer. AI agents can work together across voice, messaging, and enterprise systems to complete tasks like loan pre-qualification, prescription refills, or managing insurance claims—without forcing you through phone menus and transfer queues, only to wait while a rep manually does the work.

That said, replacing legacy systems is not an overnight process, since enterprises and their workforces need time to adjust to new platforms and processes. But this transformation is already underway, as AI-first tools are becoming a staple for the CX industry.

Also Read: The Metric That Mattered Most This Prime Day Wasn’t Revenue

If AI agents can resolve issues autonomously, are we approaching a point where human contact center agents become the exception rather than the rule?

We need to move beyond the conversation of “humans versus AI.” The future is “humans with AI.” What we’re witnessing is the rise of the hybrid workforce, in which specialized AI agents and human employees operate side by side. Where that division of labor falls depends entirely on a company’s business rules, risk tolerance, and operational strategy.

For routine, high-volume requests, AI agents deliver instant, autonomous resolution. But when an interaction involves high value, or demands deep empathy and nuanced judgment, a human agent steps in. In those moments, AI doesn’t disappear. It shifts into a supporting role, acting as a real-time copilot that proactively assists the employee and automates tasks, allowing them to stay entirely focused on the customer.

While AI agents continue playing a larger role in the workplace, people remain an important part of the customer experience. What’s changing is how work is divided. 

CX has long been measured by CSAT and NPS. In an AI-first world, which legacy metric has become obsolete, and what should replace it?

Many traditional CX metrics no longer tell the full story, and Average Handle Time (AHT) is where that breakdown is most obvious. As AI absorbs routine tasks, human agents handle more complex and higher-stakes issues, naturally driving their handle times up. That’s not a drop in human productivity; it’s proof AI is working. Evaluating these performance streams in silos misinterprets higher AHT as poor performance, which undermines the AI ROI story.

Instead of measuring how fast an agent can end an interaction, organizations must measure end-to-end business outcomes. Resolution Rate is the metric that matters most now. Customers care that their task was completed entirely, accurately, and quickly. That’s what makes a great customer experience.

Most enterprises are racing to deploy AI across their customer service operations. Are they solving the right problems, or just automating broken processes faster?

For AI to deliver real value, organizations must start by asking themselves the right questions. Too many rush to use AI on top of an unready data foundation when they should first fix their data readiness and identify the specific problems they want to solve.

A strong data foundation and structured knowledge enable AI to act effectively. Without it, AI agents make errors on outdated or siloed information. Unlocking that data is what makes multi-agent orchestration possible—enabling specialized AI agents to collaborate and execute tasks across different systems, departments, and teams. 

Also Read: CMOs Shift Budgets Toward Digital Growth: Gartner

As AI starts deciding refunds, escalations, and customer interactions, who’s accountable when it gets the call wrong — the vendor or the enterprise?

AI is not foolproof. And accountability has to be shared.

Technology providers are responsible for building AI systems that are accurate, transparent, and designed with strong governance and testing capabilities. Enterprises are responsible for defining business policies, monitoring performance, and determining where human oversight is appropriate.

The goal shouldn’t be to remove accountability from either party. It should be to create AI systems that make decisions transparently, can be validated before deployment, evaluated in real time, and continuously self-improve. As AI becomes more autonomous, trust will depend on that shared responsibility.

If you were advising a Fortune 500 board today, would you tell them to invest more in AI or in rebuilding the data and governance it depends on? Why?

AI innovation and governance aren’t either-or decisions. They are both critical steps to succeeding with AI at scale that need to work in tandem. As AI usage increases, the demand for innovation to differentiate and safety guardrails also goes up—trust shouldn’t be a consideration after agents are already deployed, interacting with customer data, and delivering answers. Enterprises should invest in building a foundation that enables specialized AI to operate across customer journeys, enterprise systems, and business processes while maintaining security, compliance, and trust.