We are Sorry, All Agentic AI Agents are Busy NowWe are Sorry, All Agentic AI Agents are Busy Now
The underlying issue before anyone can embrace agentic AI is trust, which has to be earned -- both by customers and by the enterprise.
April 2, 2025
If there was a word cloud from the recent presentations at Enterprise Connect, the biggest, boldest words would be Agentic AI. Yet there was a large discrepancy between the Agentic AI capabilities I saw at Enterprise Connect and my personal experiences with automated self-service. It’s hard not to be jaded (or deflected) with the frustrating reality of the automated self-service solutions we regularly encounter.
Just the week before Enterprise Connect, I had two urgent customer service needs associated with my travel. Both had customer service chatbots, and neither bot was able to assist me with my problem. The first time was from the backseat of a ride-share heading to the wrong destination: I needed to correct the destination, but the app wouldn’t allow it once the ride had been booked. The driver was tried unsuccessfully too. There was no live assistance, and the customer service bot just couldn’t comprehend my situation.
Then came my awkward hotel check-out. The hotel insisted I owed the full payment even though it was prepaid at the time of booking. The travel gnome had a bot for service that truly wanted to assist me, but did so by urging me to change my problem to one it could solve.
Most self-service or automated customer service efforts are frustrating, so forgive me for being cautious of this agentic hype. Can this new AI really be that much better? Kinda! Agentic AI is here, and where it works, it works wonders. For example, Cisco showed in its keynote how an AI agent helped a customer order flowers. He didn’t know the name of the flower but could describe how it looked. The agent sent the customer a photo of a likely flower and completed the order. AI agents reimagine self-service with dynamic interactions. But there are a lot of situations where AI agents are unlikely to be used.
First, we need to clarify that along with the technology, the terminology is changing. The issue is the word “agent.” The contact center industry has claimed the word. A brochure or website that featured a photo of a smiling human wearing a headset was essentially a stamp that said “contact center content.”
There are, of course, lots of different kinds of agents and many have nothing to do with the contact center -- travel agents, secret agents, cleaning agents, etc. While context-sensitive definitions for the word “agent” are not new, it’s become confusing in the contact center industry because the industry is embracing the broader customer experience (CX). Suddenly, we have “agents” and “CX” in conversations that have nothing to do with contact centers. For example, grocery store cashiers are CX agents. Almost every company and sector is or will soon be talking about AI agents, and most of those conversations will have nothing to do with contact centers or the people who staff them.
AI Agents may or may not be related to a contact center operation. An AI Agent is an AI-powered software that can perform a series of tasks. It’s easy to underestimate how complex “a series of tasks” can be. Back to the grocery cashier: scanning items is a part of their job, but they also deal with products that don’t scan, they verify the buyer’s age for some products, collect and verify coupons, process different forms of payment, are a part of a larger theft protection system, and fix/replace damaged items, to name a few. Many self-checkout initiatives failed because automating or delegating all these tasks is very difficult.
Agentic AI involves a lot of new technologies, including various forms of AI and ML, natural language processing, and computer vision. They use modern generative AI for natural conversations to determine customer intent. Agentic AI systems will typically require access to multiple systems to do their tasks. These might be internal and external systems. Some will require integrations, preferably with modern APIs when available. In other cases, it might require speech-to-text conversions or screen scraping. All of these have their own set of issues. We also need to design for reliability and scalability.
While those complications are legitimate concerns, the bigger issue is that most of these Agentic AI systems need human supervision. As a result, most agentic solutions are implemented as personal assistants or internal-facing use cases. Enterprises are going to be rightfully cautious about giving AI agents the ability to refund, discount, or set prices. The underlying issue here is trust. Enterprises need to trust these AI agents, and trust has to be earned. In addition to hallucinations, enterprises commonly have inconsistencies in their documentation and workflows.
One firm that is making exceptional progress with customer-facing workflows is Cognigy. The week prior to Enterprise Connect was its Nexus event, where customers shared large-scale deployment successes. Cognigy specializes in self-service and assistive solutions in enterprise CX.
In his opening keynote, CEO Philipp Heltewig presented the concept of Artificial Capable Intelligence (ACI), a concept popularized in The Coming Wave, by DeepMind cofounder and current Microsoft AI CEO Mustafa Suleyman. ACI bridges the gap between weak or Deterministic AI (also known as Narrow AI) and Artificial General Intelligence (AGI). ACI makes agentic AI practical. It involves handling a collection of different but related tasks within a category.
Heltewig explained to me that generative AI has a limited ability to reason. It doesn't have the design or training for reasoning, but does reason. This is because the large language models convey reasoning; for example, when we say, “Have a safe flight,” we can reason that flying can be dangerous. Cognigy has succeeded by combining generative AI for natural conversations with deterministic AI for task completion. Cognigy’s rebooking agents can’t drive a car or beat world champions at Go, but combines its Deterministic Agents to accomplish related tasks.
Most of the AI examples at EC25 were about personal productivity, such as documenting, coaching, analytics, and other internal use cases. Several Agentic AI examples help employees become more productive, an example being the calendar management feature in Zoom that assists with scheduling meetings.
AI solutions that directly interact with customers will likely be kept on a leash until the vendors better address trust, likely via improved training and AI model improvements. We are starting to see agentic AI solutions handling simple customer service inquiries: Zoom showed its AI could perform some simple account changes and even upsell the customer, and at Reinvent in December, AWS demonstrated a travel concierge that leveraged customer information and travel details to suggest activities during the customer’s planned trip. If the customer accepts the recommendations, the AI then purchased tickets and made reservations. As we build more trust, perhaps we can skip the manual approval process in the future.
AI is changing every aspect of the contact center at breakneck speed. Agentic AI is exciting, but the vast majority of the use cases will be internally focused. External use cases such as making recommendations or implementing simple changes will be safe. It might be a while until these agents can issue refunds or solve unusual glitches – the very problems that will cause people to call and ask to speak to a contact center agent in the first place.
Dave Michels is a contributing editor and analyst at TalkingPointz.
About the Author
You May Also Like
