An “Inconceivable” Conversation With Dr. Pete Cornwell on Simple vs. Agentic AI

Dr. Pete Cornwell, Director of Contact Center, offers a fresh perspective on customer care and is sharing his wealth of knowledge at Customer Contact Week in Las Vegas. With over 35 years of experience spanning information systems, design, architecture, and consulting for industry leaders like Terazo and Blue Cross Blue Shield North Carolina, his expertise runs deep. Add to that a decade as a professor and chair of Engineering and Information Sciences at DeVry University, and it’s clear that Dr. Cornwell has plenty to say about the ever-evolving world of digital transformation.

Before heading west for the conference, I sat down with him to glean some insights he’ll be sharing with attendees and partners alike.

Take a Seat, Class is in Session

Our conversation was set to focus on AI and its applications in the contact center, but as I launched into my questions, Dr. Cornwell first asked me to examine a meme.

If you’re unfamiliar with the 1987 movie The Princess Bride, you’re not only missing out on a cherished piece of nostalgia, but you’ll also need a bit of background to understand his analogy. In the film, the protagonist delivers a famous line to a villain who repeatedly uses the word “inconceivable”, even when things are clearly very conceivable.

Pete followed up the meme by saying that if there’s any term that makes him grind his teeth more than Digital Transformation, it’s Agentic AI. It’s tossed around daily as a flashy, vague placeholder for everything from artificial intelligence and large language models (LLMs) to integrations and machine learning (ML)-driven workflows. This misuse is particularly troubling in the contact center space, where it has become a buzzword applied to almost anything.

Pete is intent on drawing a clear distinction between AI and Agentic AI from a customer contact perspective. Both are critical components of today’s AI-driven customer care, but Agentic AI is poised to unlock a wealth of future opportunities in this space.

“Simple” AI

While LLM-based models are incredibly complex, AI is often used for relatively simple applications in customer service, such as self-service, agent deflection, or assistance. Many companies begin their AI journey by deploying it to deflect calls from human agents, handling straightforward tasks like providing business hours, account balances, or credit card activation.

Additionally, voice and text-based chatbots can support intelligent routing, allowing customers to bypass frustrating IVR menus and connect directly through an Intelligent Virtual Agent (IVA). Yet, despite these capabilities, this is still not Agentic AI, these functions serve as filters between customers and human agents, managing deflection or routing rather than true autonomous decision-making.

The space continues to evolve as new AI-driven capabilities are added to CCaaS (Contact Center as a Service) offerings each year. AI-powered agent prompting, coaching, and even translation are all part of what’s possible. While Pete admits his skills as a clairvoyant aren’t highly rated, he predicts that these capabilities will become commoditized within five years, standard features in the arsenal of any CCaaS vendor. He invites anyone to call him out if he’s wrong, he’ll be waiting.

What Is Agentic AI?

If you want a simple definition of what agentic means, it’s actually embedded in the term. Agentic AI is used to describe LLM-driven software that can execute sufficiently complex workflows that it could replace an agent. This typically means that like a human agent, agentic AI will need to make decisions in a highly variable data environment, sourced from both the customer and via integration.

From an implementation perspective, like its human equivalent we want to give each unique agent type a well-defined set of responsibilities to achieve advantages of understandability, maintainability, error management and observability. Similarly, many of the communication and business metrics we use to measure the performance of human agents can apply to their agentic counterparts. For example, a credit card company could conceivably use agentic AI for everything from general service enquiries to fraud reporting, and even customer satisfaction surveys.

What we can draw from this example is that we open the possibility for collaboration with agents coordinating activity to fulfill extensive tasks that would often require multiple human representatives and frustrating delays as the customer is transferred between departments.

Agentic AI in Action

Pete expanded on this concept using a lost credit card scenario, illustrating how Agentic AI can streamline customer service. This process involves four AI components:

1. A simple AI IVA chatbot that will provide voice-based routing for a customer. The following agentic components (“bots”) that will use natural language processing and output to speak to the customer.
2. General Customer Service – agentic AI designed to handle a range of customer service scenarios from simple balance inquiries, new card verification and of course then going on to lose it and requiring a replacement.
3. Fraud Handling – a bot designed to establish and open a fraud investigation.
4. Customer Survey – a bot that will craft an optional customer satisfaction survey based on the workflow delivered to the customer.

These agentic bots, coupled with the aforementioned IVA have the capability to provide a seamless flow of interaction with the customer. Consider the following voice flow:

1. An anxious customer calls the customer service number, after a prompt, the caller simply says “I’ve lost my card.” The Simple AI IVA routes the call to the General Customer Service Bot.
2. The General Customer Service bot first validates the customers identity and the missing card number [omitted to save trees and your time]. The customer reciprocates with the correct information.
3. The General Customer Service bot then asks the customer when they believe they lost the card. “Sometime last week, I don’t use it much and when I looked in my wallet I couldn’t find it” the customer answers.
4. There are recent transactions on the card, so the card account is passed to the Fraud Handling bot. Meanwhile the General Customer Service verifies the address on file with the customer and calls back-end services to print and dispatch a new card.
5. The Fraud Handling bot then asks the customer to verify a predetermined number of recent transactions from the account.
6. The customer replies “I’ve never been to Cancún” when presented with a specific transaction involving a Mexican resort and spa. The Fraud Handling bot opens a fraud case, again using a back-end integration.
7. The General Customer Service bot gives the customer a claim fraud case number and a delivery time for the card.
8. Finally, with assent from the customer (hopefully now less stressed) a Customer Survey bot draws a set of questions drawn banks associated with the bots they interacted with. The customer responds, the data is logged and the call ends.

Pete recalled a similar experience that required three different human agents, each with a long wait time between transfers. This AI-driven approach achieves the same outcome but with major advantages:

• No human were required for the workflow.
• The customer perceives interacting with a single entity throughout the call.
• Reduced wait times and minimized anxiety about disconnections.

To Recap

Agentic AI has the potential to completely replace an agent through the provision of complex workflows driven by complex inputs and integrations. Simple AI provides simple self-service for deflection purposes and supports a live agent and/or can provide natural language driven call routing.  Finally both have a critical role to play in building an AI-driven contact center self-service experience.

Source: Read More