AI Agents Need a Reliable IT Foundation First

Hey, everybody. We're back at the ManageEngine User Conference in Orlando, and I'm talking with my friend, Romanus, about, well, where is all this AI stuff going? Because there's just a lot of agents now and models, and people don't know exactly what to use when.

So from your perspective, when do we use what type of model, and how does that get inserted into something like an IT system? Because sometimes I feel like the latest and greatest models may be a little overkill for what we need to do. True.

So how do you see this evolving? Yeah. As you mentioned in your question itself, people don't know what to use and where.

Right? That's where it's going to give more clarity. Right?

When you have your basic infrastructure properly designed, reliable you get, and then you have this AI layer. This is going to tell you, okay, for this security purpose, you don't have to see multiple dashboards to make a decision and feed the decision in terms of remediation to your tool so that it can implement. Instead, that's going to give you suggestions based on different dashboards from various telemetry data, which it collects from the entire network and gives you suggestions.

And based on human approval, it can go to the next step. It can take its decision on its own, but still having human in loop for the sake of business, for the sake of critical mission operations business, you need to have human in the loop. That's how it's going, more towards autonomy.

I feel like a lot of folks are diving into AI. There's a lot of enthusiasm, some intrepidation. But I wonder if their infrastructure in the IT environment is really ready for AI, and should they be thinking through what it is that they need to lay the foundation for to really get the benefits of it?

You mentioned telemetry data. Not everybody has that readily accessible. So what should folks be putting in place first before they leap?

Right. So unless and until you have your basic data structure available ready, hopping onto AI is not going to solve the basic problems. So it needs proper data flow.

It needs to understand what's happening at the grassroot level, and it will have the agents to implement what is required for you in terms of running your business. So it all depending on the infrastructure. Say, for example, if you're managing your endpoints, it could be mobile device, it could be your phone, or a tablet, or Mac, or Linux machine, or a server, or a Windows box, right?

But the requirement is you have to make sure that you have to secure them. So each has its own way of managing it, securing it. So you don't have to have tools or worry about the tools to manage your Mac, Windows, your mobile devices.

So at this layer, you can say there is a latest patch or there is a threat, and this threat vulnerability has to be patched across your network. And you say that you give a workflow to your system and it deploys. It deploys it differently for different operating systems.

You don't have to worry about that. Now, the underlying technology of collecting information from these different endpoints and then deploying it across your entire estate is automatically taken. Unless you have better structure inside, you'll not be able to achieve that.

Mm-hmm. So that's actually being automated, and the information is collected, and you are able to talk to the system. Is this ever gonna wind up being fully autonomous, or is it always gonna be some state of semi-autonomous because there is this human in the loop, or is the human over the loop, or where does the human fit?

Human in the loop is something that generally is guidelines given by the AI guardrails or people who are governing that. Because take example of healthcare, right? A patching is patching for standard patching for every organization, let it be retail, let it be banking, healthcare.

Now, but when you do a reboot, are you rebooting a critical pacemaker machine which is under operation? You cannot do that. Mm-hmm.

From the patch, from the IT perspective, from security perspective, you have to reboot that machine. But from a domain expert who's actually understanding the business will know that this cannot be rebooted at this time. So you have to take a separate downtime and do the management.

Now, that's where you need to have human in the loop to ensure that what is critical for the business, for the human value, you have to make sure that it's taken care better. To your earlier point, it's now possible through natural language to talk to the systems. Correct.

But the systems are also gonna talk back to us, no? Mm-hmm. And so how smart will all this get?

Because at some point, will I have an IT environment that just tells me that this thing is about to break and we should do something about it? Right. Yeah.

So you can define it. Say, for example, there's a critical malware alert, right? Now, you have escalation matrix.

Every IT teams would have that. So in case of emergency, this is your first contact. You have to contact him.

Now, if he's not responding within five minutes, it will go to the next one. So you have a hierarchy of people who can contact. And if everything else fails, nobody's picking up, then you make a decision of isolating that machine so that you can save the network.

Now, this is something that the way that machines can talk back to you in terms of putting the agency and letting you know that this is... Now, if you are going to attend that call, you can make a decision, otherwise it's going to make a decision. So that's the way that the AI is going to talk back to you, giving you instructions and level of people that it can reach for the problem.

Mm-hmm. As we kind of think about the way IT is managed in the AI era, there's gonna be, I don't know, hundreds of AI agents. Right.

At some point, how do we orchestrate that, and will we reach some sort of cognitive load where we can't manage all these AI agents, or will there be some mechanism of hierarchy of agents that I can wrap my head around? Yeah. So that's really critical.

That's the problem that in future you can expect that to happen. Now, that's where we have to have a proper identity for each AI and process whatever autonomously you do in terms of AI agents. And everything is monitored.

And so it has to ensure that it's reaching for the bound limited data which it is supposed to reach, not beyond that. If it is going beyond that, you have to have some control of that. And it has to take actions which is defined only for that agent.

Now, everything has to be monitored, like how you monitor your infrastructures, your surface performance, the access control, the permissions they have. Similarly, there is more scope in terms of how many AI agents you have. How are you controlling that?

What are its ACLs, and are they overreaching? Right? Are they doing something which is out of bound?

All these things have to be monitored, and so the system would be part of that monitoring also. There are some folks that would say, AI is leading us to the proverbial hell in a handbag kind of thing. But is there a way to do this safely?

Can we govern these things and kind of do this in a way where they don't wind up running amok and becoming rogue? Okay. Yes.

There's a lot of guardrails how people have to use AI. So the simple thing is, if it is making a decision, can it be explained? Can it be given within the set of instructions what you give?

Is it making decisions or the decisions what it makes can be explained better from the business point of view, from the values that you have set for that? Now, that's kind of guardrails that you should have with the system. So that ensures that it is actually within the limit.

It's not going beyond that. So continuous monitoring on the behavior and reach of the AI in terms of permissions, in terms of database access, in terms of the performance remediations that it does in your network. So how does the management of IT itself evolve?

Because today everybody's got 30 different tools that they're using, and they're swiveling around in their chairs getting whiplash. And does that all kind of flatten out eventually? Because now I'm just putting a layer of AI agents above the data sources and I'm not trying to manage and install 15 different tools.

Right. In one way, so as I mentioned earlier, you should have the basic reliability layer which actually does the actual management, actual security, actual data collection that's happening. So only based on that information, it's actually building it better and making it easier for you to just talk to it so that it can integrate, call it different data, different sources, and then make your life easier.

So the underlying thing is going to be the basic functional management configurations, data collection, telemetry. Only based on that, the AI is becoming very super useful for an enterprise. If it's not functional and if you don't get sufficient data, then the AI is not going to perform well.

That's right. So the model is not going to be really useful unless you have better essential infrastructure. One of the things I also feel like, every time somebody leaves an IT department, the knowledge goes with them.

Yeah. There's a lot of tribal knowledge. Is that going to be something we can now capture a little bit more so that we can bring in new folks and we don't have to start from zero every time?

And is that going to be one of the benefits of AI that we don't talk enough about? Yeah. In one way, yes.

The knowledge or the experience, what a seasoned technician who's managing the specific to one particular enterprise, subject knowledge, subject matter expertise can be trained into the system so that it's not going to start for every individual who comes into the organization, can be easily guided with that system. Yes, it's part of a benefit that you can get out of AI. Great.

Should we develop a personal relationship with these AI agents? Am I going to start naming them, or are they really just tools and I shouldn't get too attached to them because they'll come and go and they're not permanent fixtures? That's interesting.

Technology has to be consumed as technology. We have a lot of technologies coming in and going out, not consumed, and until now are being consumed. So my perspective is we have to see AI as a technology consumed wherever we need and however it should be taken for the business.

As you mentioned earlier, too much of AI, is it going to give the business outcome what you expect, or you're going to spend more on the tokens giving it to the AI provider, right? Now that's the trade-off. I would say if your business is in a very contained way, AI is not going to add value.

It's going to be a fancy thing. I would say having your existing system is sufficient rather than spending more on AI, no outcome for the business because of the AI. All right, folks, you heard it here.

Hey, don't get too attached to your AI agent, but by all means, start using them because they will make your life easier, assuming you can properly govern them. Hey, thanks for being on the show. Thanks.

All right. And we'll be back in a minute.