Kirstin Burke:
We’re really excited about today’s topic. Were talking again about AI and we’re talking about a different aspect of AI, which it seems kind of crazy that we’re talking about the evolution of AI with as young as it is. But this is a technology that is, or capability that is moving at a very rapid clip. And actually we bring up these topics as we hear our customers and prospects talking about what they’re confused about in the market or what they’d like advice for. And so really we’re bringing the things that we’re sharing with our customers and prospects to you.
And today we’re going to be talking about something around agentic AI and kind of this progression from generative AI to agentic AI. And this is something that Gartner has recognized as the top strategic trend for 2025. And in fact saying that agentic AI is going to make up 15% of daily work decisions by 2028. Sounds like it’s big, sounds like it’s something that’s growing. And something that we want to talk about and bring to the table because it’s very different.
These two types of AI, while very different, can be very complementary. And so we thought we’d unpack this a little bit, talk about each, talk about where they fit in an organization, and then we’ll let the conversation go from there.
Shahin, let’s start out, about two years ago or maybe it was December of the year prior, AI really hit the scene. ChatGPT hit the scene, blew everything up. And all of a sudden we were all creating new content and doing all sorts of new cool things really largely with generative AI. Can you kind of talk to us about generative AI, where its strengths are and really now this emergence of agentic AI?
Shahin Pirooz:
Yeah, so we did come out, as you said, and talked about AI when it was starting to blossom and bloom, and why is all of a sudden AI becoming so mainstream and so relevant? And it was, that transformational step, it was kind of like the Web 2.0 and Web 3.0. As we started to do more and more things that, you know, the web didn’t get recreated but there was an evolution to how we interacted with it.
And AI was the same thing. We had gone, what we were calling AI before that was machine learning and deep learning and it was analytics based, it was culling data and self learning from the data that it was culling and presenting business decisions or decisions in general. But it was options. It was like here’s, you know, 32 different things you can think about.
Fast forward to Generative AI’s inception and launch, and now we’re having conversations with this AI model. And the reason that we were able to do that is that the large language models, which is a huge repository of training data, that is used to train it. Just like you would train your kid. You’re going to teach your kid. This is “A” and Apple starts with A. And you know, same thing. You’re teaching the generative AI the constructs of language and understanding all of the semantics of how language is put together to have a conversation so that it can have it, and how to interpret is probably one of the biggest things, the intent that somebody is requesting in a question, in a conversation.
Tonality is something that’ll eventually become a thing so you can not only interpret what you think their meaning is, but based on their tone. I may be asking a question and it sounds sincere in text, but if you hear my tone you’ll know that I’m being a complete putz and being rude and sarcastic. And so that tonality becomes something that’ll be an evolution of generative AIs that are conversational, they’re used for content creation they’re largely replacing search. I can’t think of the last time I did just a regular Internet search because I don’t want to cull through 72 different results. I want to look at what one result is and then look at the research that that result came from.
So to me, I would say that was the big step up where we’re at that inflection point again where it’s yet another evolution of AI constructs, and it’s this agentic AI. I would say emergence is probably the best way. We’ve been talking about what I’ve been calling autonomous coworkers for a long time. And I’ve been on a mission.
I remember I spoke to a company, I don’t know in 2018, that was creating virtual coworkers. Basically, they were creating these models that were able to take, they had modeled the human senses. They were using cameras for eyes. They were using sensors for smell and temperature and touch. They were using other sensors for being able to detect something is wet from a touch perspective. They were using microphones for ears. And so they build these human-like constructs around microphone speakers, cameras, to replicate how we take inputs and interpret those inputs and make decisions from them. And they built some really cool coworkers.
They built a coworker that was monitoring a lab environment with tens of thousands of mice. And it was so important to keep the mice alive. But the minute that their hay, their bed gets wet, they die very quickly, like within 24 hours. So they literally had lab assistants walking and opening these 10,000 cages and sticking their finger in the hay to figure out if it’s wet or not. So put in this autonomous coworker. And now this coworker is analyzing all 10,000. And as soon as there’s some moisture, it notifies and says, hey, cage such and such is wet, you need to go change the hay. And so now they’ve gone from one poor or maybe ten poor lab assistants who are sticking their hand in hay to just going and focusing on what the problem is. And I think that that concept, although it was so early, was really the budding concept of agentic AI.
So the key differences between generative AI and agentic AI is that generative AI is used to generate things, generate content, generate conversation. Agentic AIs are autonomous. They take actions based on content. So they’re more adapting to the inputs that are coming in and making a decision based on that adaptation and learning from past experiences. So their models are more about learning behavior as opposed to learning language. So rather than semantic interaction, it is now, let’s call it tangible interactions.
Kirstin Burke:
What would you say, you know, when Gartner puts it out there that this is the top strategic trend for 2025? I think the last couple years every single organization is trying to explain to their board, their investors, themselves, how is it that we’re going to take advantage of AI? Which is almost like saying, how are you going to take advantage of cloud, right? It’s so big and the use cases are so broad. When you look at generative AI versus agentic, and hearing that agentic is considered the most strategic focus point this year, what is the difference and why is agentic so much more strategic and why is the focus there?
Shahin Pirooz:
So I think we have to back up.
Kirstin Burke:
Or do you agree? Let’s start with that.
Shahin Pirooz:
So it’s the word strategic that I agree with. And I don’t want anybody to jump and think that they’re saying you need to make some tactical decisions. So planning for it, I agree with. Doing something in ’25, I think we need to get past the first step which is generative AI.
There’s so many companies that are still scared to use generative AI because of security risks or my data is leaking or whatever, you name it. Any number of things that are concerns around generative AI. And those that are adapting it, are seeing vast improvements in productivity.
We internally, as you know, leverage the generative AI capabilities of Microsoft’s Copilot throughout multiple of our stacks and the data is within the Microsoft ecosystem wrapped and secured with the Microsoft permissions tied to our tenant. So there’s some very interesting benefits that we’ve achieved and gained by just using that generative AI that’s foundational in our business. Now we make, we don’t put, we’re very careful, we have policies in terms of how to use it. You don’t put any PII into it, you don’t put any customer data into it, you don’t put anything that is identifiable in any way. But you can use it to ask questions about things like industry trends, things like what’s happening in the market today. And they’re able to add a lot of value because these systems, the large language models that they have developed and learned from, people are sampling with those types of questions. But I think what is a more, and I hate using the word tactical because I don’t want people to go and start digging trenches, but the more tactical thing to happen today needs to be companies figuring out how to use generative AI internally and take advantage of it.
And there’s two factors here. One of them is to consume generative AI to use it to get productivity and do more for your business. The other is to expose generative AI to your customers so that they can get advantage of some of the information you have. I get the hesitation towards the second part. But that internal piece, companies that are not integrating AI into their work life modeling, and their plan for how they’re going to be tactically dealing with workload are going to fall behind.
It’s like 20 years ago we used to say, IT doesn’t differentiate you as a company, but if you don’t do IT well, you will be negatively differentiated against your competitors. It’s the same thing today with generative AI. If you’re not leveraging generative AI to get the productivity gains and your competitors are, you will be behind. You will have some implications in terms of your direct competitive landscape.
Now, why generative AI versus agentic AI is the focus is because generative AI is here and it’s easy, really getting to the point where people are understanding and using it in their daily lives. And like I said, I’m not doing search anymore. I look at the results and the references that generative AIs are presenting back to me and go look at the content.
So, if I take, let’s just break down what a agentic AI does versus what a generative AI does by using a scenario. So the first thing is that generative AIs, we’ve talked a lot about, they’re using generative models to help generate images, text, conversations, you name it. And they’re training on large data sets to be able to learn statistical properties of those data sets. So they’re basically grabbing all this knowledge from massive amounts of data that we’re feeding into them. Just like, as if you were to take your kid and stick them in a library and say you can’t come home until you read every book.
That’s effectively what we’re doing with generative AI. And this generates new content based on the learning patterns from what they’ve read, from the data sets we give them. So put bad data sets in, you get an uneducated generative AI. Now agentic AI, on the other hand, works on the concept of rational agents. So we’re creating multiple little agents that are doing things and they’re rational because they’re making decisions based on content. They use reinforcement learning. So they reinforce the positive outcomes, positive behaviors, and learn more about that as opposed to the behaviors that get negative reactions. And they do integrate with large language models, which is the critical piece that I’m about to tie together. And they’re more focused on the more tangible workforce optimization, workflow optimization as opposed to productivity gain. Productivity gain for an individual versus a coworker that is doing menial tasks that an individual doesn’t have to do.
So for the last two years, everybody’s been saying, am I going to lose my job? And the answer is, no, you’re not going to lose your job. But are there things that you do that, you know, there’s so many people that are buried every day and they don’t get to some very important things because they had all the little minutiae to take care of. What if that minutia, which most of the time is repeatable, can be automated and you can create a worker that specifically understands that minutia? An agentic AI, an agent, for that specific task. Now how are you going to teach this agentic AI to do what it does? It’s behavioral and reinforcement learning. So it’s teaching itself. As it does something, you say, no, that’s not right, you need to do it this way. It adapts and learns and says, okay, this is the way I need to do it. Takes advantage of a generative AI as its foundational knowledge. So we’ve now developed, if you’re not using generative AI, what knowledge base are you going to leverage for your agentic AIs to do stuff with?
So when you integrate the two together, they become powerful because you can use a generative AI to, let’s say, for example, create an email that you want to send out for a marketing blast. Then the agentic AI takes that email, integrates it with your email blast mechanism, whatever that is, and emails it out to your target audience. And it knows that that particular campaign was designed for customers of this classification. So it’s going to do those steps, automate those steps of taking the email, creating the campaign in your CRM, attaching that campaign to the email, and sending the email to all the people that are in that campaign. That’s where agentic AI becomes valuable. Now that it has something creating the content, because it’s not going to create the content, it can now take that content and do something with it.
Kirstin Burke:
That’s interesting because I think when you think about it, or read about it, or whatever. It doesn’t necessarily come across as a step function, that you really have to lock in on the generative AI to even be ready to think about agentic AI. But that makes a lot of sense because you’ve got to have your source of truth or your source of content for that agentic AI to do what it needs to do best.
Shahin Pirooz:
Yeah, I was doing a exercise when we first started turning on Copilot in our environment. And I needed, I wanted to test writing a service specification for one of our products. And so, without any input other than make me a service specification for this. And, in a Word document, I tapped on Word Copilot and I said, make me a service specification for this product. And within 30 seconds I had a seven page service specification. Was it perfect? No. But it was close enough that it saved me a ton of time developing the framework. Now I can go and edit the content to be more specific and then I can use that content to teach it and say, I was looking for this when I asked for a service specification, not what you sent me.
Another example is, I follow and I have for a long time, I came up with a previous mentor of mine, this concept of what we call SPIN engineering. And SPIN is a sales methodology that stands for situation, implication, I’m sorry, situation, problem, implication and need. And the idea is around being able to understand what the situation is that the individual is in. What are the implications if they do nothing? What’s the problem that’s being solved? And what do you need to do in order to move to the next step? And I was struggling with teaching specific decision making to my staff at that time, and operational staff, and SPIN became the mantra and it still is, at every company I’ve been at we implement SPIN.
And so I taught Copilot. I went in and I said, hey, here is this concept of SPIN engineering. Here’s what it means, here’s how I want you to structure data. So I was teaching it and then I said, when I tell you to SPIN something for me, this is the format I want you to do it in. And from that moment on, if there’s an article or a security incident, or not an incident, but security incident on the web, not our incident, so we’re seeing IOCs and things like that. I literally take the article and paste it in to Copilot and I say SPIN this for me.
And it does a SPIN summary of what the situation is, what the problem is, what are the implications of this particular attack, and what needs to be done to address it, including the IOCs and all that. So that’s an example of how I save myself hours and hours of reading and translating so that we can create advisories to our clients. And it’s a very simple coworker without having the agentic AI. Now, with the agentic AI, I can tell it go find articles on security incidents and feed them to Copilot and tell Copilot to SPIN it, and then add an advisory, and then send it out to our clients.
Kirstin Burke:
That was my next question. That you’re training it to generate, you know, the generative side, you’re teaching it SPIN, you’re teaching it all that. So my next question was how do you then move that to an agentic model, you know, that is repeatable and that is ongoing and that really serves as that individual for you to get that task done.
Shahin Pirooz:
Exactly.
Kirstin Burke:
Well, it’s interesting another use case that we have just started with, but similar. You look at business development, the business development process, right? You look at things, you know, how do you move something from top of the funnel to mid funnel. And you know that that has always been an unthankable job. It has always been something that is always a hard nut to crack, whether it’s your business or whether you’re trying to build it. But we’re starting to find that AI, both generative and not quite at agentic yet, that there are so many things that you can train AI to do in that BDR role. It’s doing all that work from the content to the emailing to the follow up to the booking the meeting.
There’s all of these things that can happen that at some point get handed off to your very good, trained, established rep who can then go do what they do very well. So, you know, just two use cases that we look at in a business to say, well, where is it that we can take menial things that we need to make sure get done, that we need to make sure get done in a repeatable fashion, in a high quality way, but then can also hand off to our team where their talent set is. So that we use the talent that we’ve got in the way that it’s most productive as possible.
Shahin Pirooz:
Agreed. Yeah, I would say, coming back to your original question which sparked this whole dialogue, it was your follow up question to the question, which was do you agree with Gartner’s decision that it is strategic? And I think the answer is yes. It kind of has to be. If we think about what we just talked about with agentic AI, focusing more on autonomous decision making and actions, while generative AI is specializing in content creation based on learning patterns, then if you take a look at those two, agentic AI has to be part of the strategic business planning for business operations. Like the things you just described.
Like how do you do the pipeline management and how can you train an agent AI to go do that for you so the sales reps are just not having to go and like change, you know, I got all this feedback from this call, does that mean that there’s buy decisions or know they’re moving off to a competitor. If it’s a competitor, can we get competitive data on that competitor and understand what happened? And, you know, the bigger thing that we’re not seeing yet is when you integrate the synergies between these two AI modelings, the innovations that are going to come out of that are going to be crazy. Like the workflow that we’re implementing today in our sales process, business development process, if there was a platform that focused on just creating agents for inside sales, for example. Or content creation for outbound marketing, or you name it.
Any number of these use cases, these are now evolutions and innovations that will start coming up where people are mixing the power of generative AI content creation with the actions of an agentic AI that’s autonomous. And of course there are security implications and all that, but there always are. Those things don’t go away.
Kirstin Burke:
For sure. Well, I think what’s interesting about this, too, and if I’m listening to this right, obviously these are strategic business decisions and business imperatives. To your earlier point, if you’re not thinking about this, if you’re not starting to do it, it’s going to affect your business because your competitors are, right? So, definitely board level, CEO level priority, yet each business unit owner knows their business best. And so, there are a lot of business… your business unit, my business unit, right? We’ve gone in there and we’ve been like okay, how can we use this? We’re figuring it out and we’re doing it.
Now if there’s a leadership mandate for each team to do that, is that the way you advise companies to do this? Or is a lot of this stuff happening kind of bottom up? Because it seems hard for a leader to dictate that this needs to happen not knowing all of the intricacies of processes and even those areas that are the most high probability or that will have the biggest bang. Like as a leader you may not know that, but your people do. And how do you get the buy in and the creativity to really start making this happen?
Shahin Pirooz:
Yeah. It’s hard to do anything from the bottom up because it takes forever. But I’ll give you a corollary conversation I just had yesterday, which was I was talking to a partner who said, you know, I’ve got some customers who are taking advantage of generative AI to move some work processes, accelerate them. But I’ve got these verticals which I don’t know how to even get started on. I don’t know what conversation to begin. Like, how do I get them excited or interested? They, you know, most of the clients that I don’t have some ideas for how to use generative AI end up saying it’s not for me, we don’t need it, there’s a data risk or whatever.
So it’s a lot like when security was, you know, it was at the firewall. And most companies would say, I got a firewall, security is not for me, nobody’s targeting me. It’s the same dialogue right now. We’re at the cusp of this new evolution of technology consumption. So, you have to start and make the business recognize the potential of leveraging these AI models in their business. The potential productivity and profitability gains associated with them.
Scalability becomes so much easier when you’re not having to do a stair step function of, you know, instead of creating agents, having to hire an individual and that individual now only can do a fraction of what you need and they have to be busied up with something else. Whereas you can create this teeny tiny little agent, and they’re not small, but you get the idea to go do that little fractional task.
So having that dialogue at an executive level with the executive team to get them excited about it and get them thinking, oh, maybe there is some opportunity here. Then, going into the departmental and understanding the business and what are the potential workloads that can be optimized? Is it just content creation? Is it content creation and action? What fits best in their business? Do they expose some of the content creation so customers can ask questions? Think of a generative AI chat that customers can interact with to say, what’s the status of my shipment, what’s the status of my order, what’s the status of my medications, you name it.
So those are all things I think that when we look at something like a ChatGPT, for the first time, and we’re sitting there and you’re looking at this prompt, you’re like, what do I ask it? And that is, visionary is the wrong word, but that is that visionary aspirational dialogue you need to have with executives in a company to say, what if you could do these things? Would that be valuable to you?
Kirstin Burke:
Right. Well, it’s almost like a workshop where you’re kind of looking at two areas of the business. What are the things that slow us down, what are the things we can’t afford to staff? Or what are the things that if we could only get this out of the way, we could do this? So it’s like, I don’t want to say it’s the negative side, but it’s like what’s the drag? But then it’s the art of the possible. What if we could do this? What if we could tie these things together? Our customers would love it.
And so it’s kind of looking at those two sides of the business to say, how do we be faster, better, more cost effective? How can we be more value add? I’d love for you to share the example you gave the partner yesterday because it’s that kind of getting out of your head and saying, well, what if these things could happen? To think of how you might be able to use it.
Shahin Pirooz:
Yeah, the specific example was, the vertical was healthcare. I’ve got all these clinicians and they think that AI isn’t really going to be able to do much for them with their customers. And I said, what if the doctor had a recording, like we do on our webinars and things like that, that was recording the appointment with that individual and capturing the notes so the doctor didn’t have to. But then on the back end could also take a look at the notes and evaluate symptoms and see that this individual has been talking about the same symptoms for the last six appointments. And it’s something that the doctor would have had to gone back and look at all six appointments to see, look at the notes from each one, when usually they don’t have time, they’re just looking at the last appointment.
But what if the doctor can now get a support worker, coworker, that is analyzing the history of this individual and saying, and I gave this example to you and you came back with a brilliant one, which is, here’s two symptoms. This one on its own is not a big deal. This other one on its own is not a big deal. But three visits ago they complained about that symbol, and today they’re complaining symptom and today they’re complaining about this symptom. And the two of those together are concerning. And now we need to investigate and see if there’s something here. So there’s things like that, drug indications, for example.
There’s so many different use cases, setting up appointments, being able to do that instead of, how about this date? How about that date? Being able to look at the doctor’s calendar and say, here’s the five dates I have available and times. Does any of those work for you? And being able to solve those problems without having to have you sit in there and wait for the nurse to come in to do this part and wait for the other person to come in to do that part. It’s at the fingertips of the doctor, so they don’t have to do all this, take the time to do the research before the meeting. It’s there for them, in front of them.
Kirstin Burke:
And at the end of the day, big idea, what if they could spend more quality time with their patients?
Shahin Pirooz:
That would be huge, wouldn’t it?
Kirstin Burke:
Well, so my takeaways from this, first of all, definitely the two are a pair, right? Salt and pepper, peanut butter and jelly, better together. But there really is a step function. That going in and all of a sudden thinking, oh, agentic AI is the next big thing, let’s go do. The answer is, well, no, let’s follow the flow. And to be successful with agentic, you really need to understand, embrace, and execute well on the generative AI and then kind of grow into those use cases for agentic. So definitely that’s a takeaway for me.
And then really having that space at either a leadership team level or even asking your area leaders, hey, where are the things? Kind of that, that workshop framework. Where are the areas that we’re struggling, that we can’t fund? The big idea, the what could we do? What could we do for our customers that we never thought we could do? You know, and really kind of trying to workshop that and look at the art of the possible and really then start skating to where that puck is headed, if you will, in terms of your efforts.
So I guess those would be the things that I would take away. I think the other one is AI is here to stay. And so there’s a reason that we keep talking about it, because as our prospects and our clients are looking at their technology refreshes, as they’re looking where to invest, whether it be storage or security or whatever, AI is at the front and center of so many of these folks thought process because it’s either in the front of their head or in the back of the head that this is either coming or it’s here. And I have to figure out how to accommodate it.
So, as DataEndure, we’re having these conversations whether they be super tactical or whether they be, you know, like Shahin did yesterday, helping someone just imagine what some use cases could be. And we’d love to help you. So, if this is an area you’re stuck, if this is an area you just kind of want to press on and get some other feedback, we’d absolutely love to join you in that conversation.