One of the great joys of being an analyst is having the time to look at new buzzy technologies and form opinions on whether they will be useful for contact centers. Then I get to spend time talking to folks about how the technology will make a difference in the contact center.
The latest hot new thing is agentic artificial intelligence. This is a real and exciting technology, bringing the ability to act to AI. This is a hot, nascent, and poorly defined technology, so hot that it was the talk of this year's Enterprise Connect conference in March, so nascent and poorly defined that I saw many vendors bending its definition to accommodate whatever their products do. This bending of reality doesn't help anyone, so I thought I'd share some information on agentic AI and some thoughts on what it will mean to us over time.>
I'm an analyst at Forrester Research; other analysts here are looking at agentic AI from a variety of angles, and there is a lot of excitement for the potential Back in August, Forrester published a report that works sort of as a primer for agentic AI, what it is, and what it could mean. The report identifies six traits that agentic AI brings to the table:
- Reflection — The software improves itself. You have seen those little thumbs up and thumbs down responses to output. Agentic AI software can take that and reprogram itself to behave differently next time without help from a human. More unnervingly, the software can also analyze interactions and make changes based on signals that something went wrong.
- Memory — Generative AI talks and interacts. Much like my 3-year-old grandson, it lives in the moment, interacting without thinking about what is next. If you are going to automate a process you need to understand all the steps in the process and be sure that you are walking through them properly. At runtime you need to remember your state and act accordingly.
- Planning — Planning is the difference between giving software goals to achieve vs. programming every step in the program. The interactive voice response systems I built in the 1990s automated processes, but you had to explicitly teach them every step of the process and build out all the error handling. It was laborious and brittle. Much of what passes for robotic process automation today is no better than what I was building 30 years ago. Agentic is different; it works at a higher level and is more flexible and much easier to program.
- Tool use — Agentic systems can interact with other software processes, access data, and interact with external systems to do the work it needs to do. All of this is required if you want to have your chatbot open tickets for your customers or make changes to flight reservations.
- Multiagent collaboration — This is what most people identify as the key to agentic applications. I can have one piece of software call others to do specific things. This is not exactly a new paradigm, but combined with the above you are unleashing software that has a new level of power and capability.
- Autonomy — Software with the above characteristics can run on its own without the sort of human intervention we are used to providing.
So What?
In terms of what agentic AI can do for the contact center, I like to think of it as the yin to generative AI's yang. Think about building a chatbot; generative AI makes it so much easier to talk to customers. Historically, building a chatbot required a six-week process for every customer question, requiring developers to identify all the possible ways someone might ask a question, laboriously define data slots and error handling, and then stay on top of how the questions might change over time. Generative AI understands what people say and knows how to react; you just tell it to get this information from the customer and you are kinda done. Oh, and the conversations are far more personal and enjoyable.
With agentic AI the benefit is very similar, but the focus is on the other end of the interaction. Once you have all the information from the customer, you can use that information to submit a new support ticket. You need to tell it where the data is, give it instructions on the data format and availability, and send it on its way. Imagine, not needing to format every API call or having to explicitly build out individual steps to ensure every bit of the process is followed in the proper order with the proper permissions and the like.&
The above is an oversimplification; there is work to get this to scale beyond prototypes, concerns of hallucinations, and many other issues around hardening sufficiently to be ready for customer service. In fact, at this moment there are scant few examples of generative AI based self-service applications in production, much less any true agentic AI.
Hopefully this gives you a sense of the potential of agentic AI for your contact center. This isn't going to happen overnight, but it's coming, and it's going to change customer service.
Max Ball is a principal analyst at Forrester Research.