Agentic AI: Behind the 2025 Top Tech Trend
17:36 Share

Are you trying to navigate the hype of A I now is the time to capitalize investment in A I has reached a new high with three quarters of cees already personally using A I tools gardeners here to help providing executive leaders with expert insights, strategic guidance and actionable advice to make informed decisions and stay ahead of industry trends. Visit gartner dot com or click the link in the show notes to download our A I action plan and learn how to build the right A I strategy for your organization.

Welcome to garden thing casts. I'm caren stock. Lock her in today or diving back into gardener's top strategic technology trends for twenty twenty five and specifically agented ai.

As you read through the electron's research, you may notice that each trend features one of gardeners expert analysts. And to kick off the first trend, a entia I, we're joined by tom kha, a senior director analyst with gardener's technical services providers division. Tom, welcome in.

Thank you very much again. And these are very exciting times in the world of agents. A, I.

They are, and we are ready to hear about them. So you can help us set the stage a little bit. Can you tell us how are we defining agented ai? Why and why is this the very first trend in gardeners top ten tech tron report?

Well, I think we're defining the gentile I is autonomously, you know, systems that can plan economists sly and take actions to meet goals. And right now, I can tell you that my day, you know, A I agents are a subset of agents. A I and I spend pretty much all day, every day talking about A I agents and that, like the definition of A I agents, is that they are autonomists or semi autonomous software entities.

They use A I techniques to perceive, make decisions and take actions and cheap goals in their digital or physical environment. And there's a couple of keys about band is that is that the A I agent is the software energy is not like the large language model that I might be using to plan. So I think that a that's a thing of gardener that we like to keep separate is that an AI agent is a is a software entity and IT uses like a large language model to make plans, but IT isn't the model that has other software components to IT, and that's which drives its ability to take action.

Can give us a quick example maybe?

yeah. So if you think about them, let's say you're trying to process a bunch of really complex, uh, invoices and you need to read these invoices and then you need to make a plan for where you're gonna put every line on those invoices. So you would use the software maybe to bring the information to the large language model where you're going to create the plan to go and then put every line item in the right place in the counting software. When you go put those line items into the software by, you're using tooling, which is part of A I agents, to put the line ims in the right place and say, the accounting software. So I I think that's a good example of if you track where i'm going there.

that is a good one. And I think some listeners like me may have also heard the term AI agents or even guardian agents. How is this different from agented? ai?

yeah. So I think guardi an agents are a type of agent. One of the big chAllenges here with A I agents is that as we roll them out and we have more and more of them, question is, how are you gonna cure them? You know, these are things that make their own plans. By talking to an a large language model or not using some other A I technique, they make their own plan.

So the question is, six months from now, how do you know that A I agent is behaving correctly? So there's this idea of a multi agent system where you're breaking down the test into very specific roles for each agent and that maybe one of those agents might be a guardian agent that is doing security check that maybe making sure that all the agents are still on track with the ultimate goal were trying to achieve. But also just just the whole idea of you know in the platforms that we see coming caring， there's there's there's like an orchestration level where you can sort of see all the AI agents you have working. And I think there's some question whether or not you can secure a multi agent system good enough from the orchestration level. So maybe as part of that multi agent system, you put in a guardian agent to keep an eye on things.

okay. I know you gave us the example already with the invoice, but i'm wondering if there's some more examples of A I agents that we're already seeing today to just help you bring me to life even more.

Yeah, I think I think like you see a lot of activity going and like insurance claims processing, right, which is a very you know if you process insurance claims slowly, you know IT cost the insurance market a lot of money. You know it's it's they call IT like leakage. You're something like that, right? It's it's so speeding up those processes that can be automated just with code, you need to be able to make a plan.

You need to do a little bit more there. Customer service, of course, is a place where A I agents are gonna have a big impact. You know, planning out a loyalty program based on your personalized experience where you just you know you called into the context center and now in a AI agent is gonna go make a whole plan for you to be part of this loyalty program, but it's gonna personalized and the AI agent might even take the actions of that look like sending you an email and three months and things like that.

I think another big place, of course, is code generation. General A I in general has already had in an impact on code generation. And I think more sophisticated assistance that can really come up with their own plans, which is what would be the difference between an A I assistant and A I agent. I think A I agents are definitely gonna know encode generation will definitely play a role.

Okay, some turns claims, customer service code generation. So some good current examples. How do you see IT expanding next year and maybe even beyond that? yes.

So I guess we should be careful here like the deployment of actual AI agents that can make their own plan and take action with or without human review.

I mean, they're very fun far between, right? A lot of this stuff right now is proof concept um but I do see I mean, one of the things about A I agents is that many of them today are using a large language of model forward to help them sort to create their plan well as these large language models gets smarter. So we are A I agents.

So I do think that's a kind of an interesting twist in something for everybody to keep an iron. We see some of the real big large language models adding actions that almost seem like they're part of the model. So there's this kind of boring of who's gonna doing the action. And as the eight, as the models get smarter at math, figuring out puzzles, they're gona make our AI agents even stronger, right? Because the AI agents use them as a two.

okay? So learning as they go, we're gonna take a quick break. We heard a little bit already about some really good examples. What in A I agent really is when we come back, we will talk a little bit more about some of the chAllenges and some insights on where some of the leaders are are focused or really should be taking some action right now.

Are you in a new role? Chances are you're not alone at any given time. A quarter of executives are in transition, with forty percent under performing and meeting their objectives. A new role requires a new plan, and gartner provides the insight, guidance and tools you need to make an impact now and throughout your tenure. Is IT gardener that com or clicking the link on the showing tes to downadup exclusive action plan and speed up your impact?

So now we are back, and tom would love to hear some more about what are some of the chAllenges that organizations and executive leaders are already seeing with the rise of A I agents.

Well, there's a lot of decisions to be pain. I mean, where are you gonna d your AI agents? Are you gonna create an environment within your own stack? Are you going to use one of your hyper scale like you know maybe IT makes a lot of sense to build your AI agents where you have um all your data in the cloud, for instance. Or maybe IT makes more sense to go to a start up who's doing something interesting and can provide you your own platform to to build A I agents.

And there's a whole big questioner how well those platforms connected into your data because A I agents like the rest of general AI, but they aren't rounded with good data. Well, they're not going to work and they're not going to be affected. And I think another chAllenges, it's interesting when I talk to these tech leaders, they have trouble gaining some other function leaders to think beyond general I assistance.

Like I think we become used to this idea. I can go to this chat feature and ask a question and journ the way I will give me an answer. But A I agents are supposed to do more than that, right? They're supposed to you tell that the goal and makes a plan and IT goes and does things right. So I kind of like trying get people thinking about, you know, I have I have a workflow that I do every day or twice a week. And if I had an AI agent to help me before I start that task while i'm doing the task, and then what what I wanted to do after the task, sort of this like before, with an after concept, to help people go beyond A I assistance and start thinking about how do I really get more productivity out of A I and .

who do you see as being involved in some of those decisions IT within IT or even beyond?

Yeah, no. Actually interesting thing here, right? I mean, I think you know our C T S R C I O S are gna decide where we're going to build our AI agents, what data can be connected, how are we going to protect the data so that you know, users can go build an A I agent to allows them to see data that they don't have the rights to see, like no salary, something like that.

So I think that's a big part of what happening right now is making in these decisions. But IT really requires the involvement of of line of business owners, right? If you want to build an AI agent to help your hr recruiters, you need to be talking to those recruiters about what they need, what are they looking for, what positions are they looking for in. So I think it's it's really one of these situations where where we need to put the power of A I agents in the hands of business leaders while still keeping IT secure, still keeping good governance and still keeping IT orchestrated .

make sense. And when you talk about protecting the data at an example of A A guardian agent.

yeah IT could that could be one example. But I mean, keeping mind that the biggest companies in the startups, they're all coming out with these platforms and they are going on this platform where you can go and build your AI agents. You're gonna connect your data sources, you're gona build an agent and they're going to provide you a some form of user security, a gateway to protect the agent so people can't pink them from outside.

I mean, most of these platforms are going to have some flavor of security. The question is, is, is going to is IT going to work good enough? Because these aren't robotic process automation, right? We had this space where we rolled out a bunch of R, P. A out there, and then people forgot what some of them did. But I didn't really matter that much because they had predetermined inputs and predetermined outputs.

But A I agents are .

going to be flexible. They are going to make their own plans, and that means the orchestration and governance of them matters a lot.

Make sense. So maybe you can share then what are some of the steps that I OS and executive leaders can be taking right now or should be taking in the near future as they think about this?

Yeah well, like I said, I mean, I wouldn't ignore the startups. Um you know nobody wants to have you know two years from now have thirty different providers of A I agents, but the startups are doing interesting things. So I would take a look at them, whatever cloud provider, I have my data and I would see what they're doing.

I take a look at what might data analytics companies are doing. And I think I think you need to experiment and you need to pick up A P, O, C, to work on where you have a problem of scale. And I I, I think this is so critical, we don't want to roll out A I agents to augment somebody y's work when they didn't have a problem of scale in the first place.

You know, you want to look at something where you need more of IT. Like if you have twenty people working on something and you had all the money in the world, would you up that to sixty people? Because more would just be Better.

You know that's that's a scale that's a that's a scale issue to go after. That's gonna you more productive. It's gona make those pointing people Better at their work.

So I think that's important when we roll out we've seen people roll out gena. I and I really didn't impact anybody's productivity and they sort of fall into the zone of indifference like I didn't matter. We don't want to do that with A I agents. We're looking. We're looking for places where we need more scale.

Got IT is a great tip. And I guess the time running out so quickly, if we wrap up, is there one main take away you'd love to leave all the listeners with.

oh, well, this is this is the big thing i'm really following. And that is this idea of long term memory in A I agents, where were going to be able to use their actions and their success rate of achieving goals and then either go back and train the agent to do Better next time, or maybe go and train, uh, fine tune the model. So it'll develop a Better plan with the A I agent in the future.

This is where we get into adaptability, where agents are learning, where they're getting Better and Better every time. And that even lead to this idea of a federal agents that every time you ask an agent to do something, a new version of IT gets spun up. And when IT gets spon up, it's based on the latest, greatest information about how to accomplish that goal.

And so now you have these agents that are getting Better learning over time, it's kinds scary. So we're going to have to have really good security and governance on these things. But that's what i'm excited about seeing who's gonna be able to take those actions and then turn them into improvements.

exciting times before lots of opportunities been enabled and multipliers. Tom, really appreciate your time.

Thanks for joining me today. yeah. Was wonderful being in her.

Very nice to talk to you. Thanks for listening to this latest episode. Thin casts features gardener, senior director, analyst tom, if you like to learn more about our topic today, does IT garden I come and click the links in the shower notes think cast, i'll be back in two weeks to break down more of the top ten tech trans. In the meantime, please read, review, subscribe and share with a colleague, so neither of you we will miss IT thin .

cast is a production of garden. This podcast may not be reproduced or distributed in any form without gardener's permission. IT consists of the opinions of gardener's research organization, which should not be construed as statements.

Effect content provided by other speakers is expressly the views of the speaker and or their organization. While the information contained in this podcast has been obtained from sources believed to be reliable, gartner all warranties as to the accuracy, complete ness or adequacy of such information. Although gartner research may address rebel and financial issues, gartner does not provide legal or investment advice, and its research should not be construed or used as such.