IT Is Becoming an HR Department for AI Agents

Hey everyone, welcome back here to Techstrong TV. I am really happy to introduce our next guest. It's his first time here on the show.

His name is John Capobianco. He is the head of AI and developer relations at Itential. And let's welcome John.

How are you, man? I am really fired up today. This is really a nice personal milestone to be on Techstrong TV.

Awesome. I've been in the industry a long time, and it's really exciting for me to be here today, and I'm really proud of it. I appreciate it, man.

We're thrilled to have you. So you kind of opened the door. You've been in the industry a long time.

Give us a little bit of that history, man. Let's hear it. All right.

Well, if you're non-technical, I started as an aluminum factory worker. So I had a little bit of trouble at Queens University, and I ended up having to take a job at the local 343 union. Not doing IT, I'm not talking about help desk or networks.

I'm talking about forklifts and heavy gauge aluminum, and steel-toe boots and 12-hour shifts. The union let me go back to school, so I worked day times, and I went to school during the day in a computer programmer course. But I found myself working in networks and infrastructure more than coding for the majority of my career.

I found myself, I ran Empire Life, a Canadian insurance company's network for about five years, and then I was actually given the opportunity to be the senior network architect for the Canadian Parliament, and I was there for about a decade. And what's interesting, Alan, is that halfway through, we had built the new infrastructure for this awesome new national importance network, but we found operating it was starting to become a challenge because of the scale and the complexity. So around 2016, we started with network automation, and that was a really big change for me.

But what I found was that programming education and now my experience with networks were sort of coalescing. Where I could program networks and talk to them with computer programming languages like Python, as opposed to logging into the command line and doing it by hand. This actually led me to join Cisco.

I was at Cisco for about three years. I was an AI technical leader for two of those years. Some wonderful formative years of my career working at Cisco.

And then I ventured out into the world of AI. I was with Selector AI, a startup in root cause analysis with machine learning, and now I'm head of AI for Itential. It's an evolution, and I sort of live at the intersection of network automation and artificial intelligence.

Working at the factory by night, going to school by day, becoming a computer programmer, wind up running the network for the Canadian Parliament. The IT guy for the Canadian Parliament. What a great story that is.

And now, like everyone else in today's world, it seems, AI. Head of AI. That's right.

Thank you. I appreciate it. I still am in contact with some of the folks I work with at the aluminum factory, and they used to call me Manchild because I had a baby face but this giant body that I live in.

And they're all proud that Manchild made something of himself. So it's been a rollercoaster. But I take this very seriously and I really try to ...

I'm very serious about the idea of technology and the democratization of technology, now that AI is really accessible to all of us, of breaking generations of poverty, of breaking the bonds of having to do physical labor with my life, and use my brain and my wits and my intelligence instead. And I really want people to maybe-- I have laid my journey bare on websites, on blogs, on videos. I try to chronicle my learning in public to maybe help that kid at the gas station who listens to my YouTube, move into IT.

Or that woman who wants to get back in the workforce and needs somewhere to start. I take it very seriously, Alan, and because of the change in my life. I can't tell you how hard physically and mentally it is to do those 12-hour shifts, three of them in a row on a weekend, pumping aluminum out versus writing AI agents.

In my lifetime, that has changed. It's crazy. When we think about, look, I'm older than you, I'm assuming.

Think about what I've seen in my lifetime. Yeah. And I do.

I think about this all the time, this whole AI thing and what it's wrought. What a tremendous time to be-- I'm grateful to be able to be living through this after everything else I've been through, the dotcoms and everything else. That in itself is inspiring and exciting.

Someone of your age and has been through it all and seen it all, and now embracing. Does it make you feel youthful? Have you used these tools?

Oh, absolutely. Has it not? I wrote a book.

Right. For 20 years, I threatened to write a book. AI was what got me off my butt to go do it.

And I did, and I'm grateful for that. John, I want to turn over to Itential, though. This is a new company for a lot of people in our audience.

Give us the scoop. So I have a long history with Itential, even before joining them. They have been a leader in automation and orchestration for infrastructure for about seven or eight years.

Very large enterprises, very large service providers, a wide variety of companies in the financial sector and beyond. And like I said, it's not really revolutionary for Itential, but it's evolutionary. We've introduced a new platform to use reasoning and action agents with things like skills and MCP tools to help tame this beast of massive networks and a lot of tickets and a lot of churn and a lot of errors and a lot of human toil.

So we've tried to make it very easy. We have builder skills. So like you mentioned Cloud Code, right from Cloud Code, I can build an agent using natural language to describe things in my vernacular, in network engineering terms.

I need a network agent to go test interfaces on these devices and send a report to this email, and if you find a serious problem, send me a Slack message. And that agent will go do the thing. But what differentiates us, I suppose, is this idea of governance, role-based access control, audit, authentication, all of the things that, it doesn't make press or the news, but how do we go from labs and science experiments to actually production and actually seeing the value and capitalizing on these tools in a production environment?

We just came back from the Lumen Innovation Summit, and it is incredible the things that we can do now with agents. We had a 45-minute workshop together with them. A lot of them, the first time seeing and using our product.

And then it turns out they built approximately 50 agents in that 45 minutes, about 35 engineers. And real good agents, firewall agents, routing agents, switching agents, testing agents, auditing agents, source of record agents. So it's a really exciting time, and that drove me to join Itential at the beginning of this year.

We had longstanding relationship through automation for about four or five years. And I have a really close connection with the people there and the culture there. If anyone has been watching me, I'm really thriving.

I'm enjoying my work. I'm doing some incredible things. I'm able to test the latest models.

We're going to be talking about decision-making models in a little bit. I was able to bring that new decision-making model capability into our platform in about 30 minutes with very little struggle, very little friction, and now I'm able to help augment those network automation workflows with a decision-making capable agent. So I'm very excited and very happy with my work at Itential.

We're going to have a workshop in December ... with actual live equipment where you can sign up, you'll be given a login, and you can actually build agents with real infrastructure. So you may not have heard of us.

I am trying to do my best and use whatever credibility I've established in my career, I've thrown it all behind Itential. I saw what they were doing with Model Context Protocol. They were the first innovation with MCP in the industry, the very first, almost two years ago now, where some of the larger players are just starting to release MCPs.

We are now incorporating skills. We have a whole governed platform and gateways. I am here to help you learn more about us, so please connect with me if you have any questions at all, reach out to me.

That's the DevRel part of my title. I love doing the AI part, but I also have a DevRel component where I'm here to connect with you. I think we are moving to the point where a lot of us are recognizing we don't need these big general-purpose LLMs, these models, for every task that we want our AIs to do.

As a matter of fact, for a lot of tasks, number one, we could use smaller models that aren't quite as token and intensive, cheaper, right? But number two, we could, and we should be, developing models that are highly specialized for specific decision-making circumstances and environments that will allow them to make just better-- The agents will work better. They'll make better decisions, more reliable decisions, right?

They'll evaluate what's out there, validating actions, determining what results before they rush into it. And I think that, we talk about governance, we talk about guardrails, we talk about all these things. That's really what we want with our agents.

We want our agents to get more right. We want our agents to understand where the parameters are and what is the real task at hand. And from what I've read, this is something Itential's working on.

Absolutely. And you've really captured the essence of what we're trying to do. So we find ourselves now building multi-agent orchestrations, and I completely agree with the right model and the right tools for the right job.

Jensen Huang, two years ago, at a keynote for NVIDIA, the CEO of NVIDIA, and at the time, they sort of laughed, and people made memes, but I took it really seriously. He said the future of the IT department is going to be an HR department for agents. Isn't that brilliant?

And that's how I find myself two years now working where I am not managing infrastructure directly. There's a layer between me and the infrastructure. And we've seen this before.

Let's take wireless, for example. I had 3,800 access points at Parliament. Do you think we logged into each one of those individually and controlled them individually?

No. All of them answer to a single controller, and we configured the controller that pushed the config down. Now, just imagine another abstraction layer that is an agentic system with reasoning capabilities and the right tools to manage that controller.

So it's just another step of abstraction. We've been doing abstraction since the beginning of time. And now I am managing the agents.

That's my new role at Itential. Not so much the things downstream, but like any good manager, I've transitioned from being a good engineer in the field up into middle management because it's later in my career, and I can apply that expertise with a different perspective, a different focus. So I'm focusing that 25 years of experience to manage an agent, one agent, my orchestrator agent.

And you know what? It itself is managing five, six, seven other agents below it. So in a workflow, I might have my planning agent that runs maybe a foundation model, Sonnet 5.

It's cheap. It's good enough to do planning. But my orchestrator might run a frontier model, Astra 6, because I want the best, most expensive model.

Like a manager, you wonder why the CEOs make all the big bucks. My orchestrator's going to make the big bucks and have the most expensive tokens and most capable model. But it's going to hand off to a less capable model that's capable enough to write the plans.

And then maybe I have a decision-maker, and that decision-making agent looks at the plans and runs it through its calculations and gives me a probability score. I'm giving this a 99 out of 100, John. Everything, but maybe the rollback plan could use a little work.

But it's good enough to pass, but you may want to look at the rollback plan and augment it. I'm happy with that. The human in the loop says, approve the plan, and let's push the config with a different agent, and that agent has the tools to actually connect to the network.

Very limited scope. That's all that agent can do is configure things from the plans given it from the orchestrator. And then maybe a testing agent, a communication agent.

The communication agent can use Gemma 4, an open-weight model, completely free. It's generating emails and Slack message and communications. It's perfectly suited to do that job.

Right? So again, that's me trying to do more of an HR, how much do I pay these agents? What agents are capable?

What tools do they get? How do they work together? I hope people are following along.

No, no, they absolutely are. Well, I get a lot of people looking at me like a dog who's been shown a card trick when I talk about agents. They're kind of tilting their head, and I've kind of lost them a little bit along the way, but really think of it as not- You're talking to the wrong people, John.

Well, don't think of it so much as a technology problem, as an orchestration problem, as a human resources problem, is what I'm trying to get at. So for what it's worth, right? I think people need to think of agents as their coworkers.

Yes. And they may not be equal. Maybe they work for us.

I don't think we work for them. But nevertheless, we got to wrap our heads around that these are persistent helpers that are doing work at a scale we can't do. And we got to stop treating them as novelties or as some sort of Skynet, Terminator evil things.

And they are what they are. They're perhaps the greatest tools we're ever going to invent to help us get work done. I love the messaging.

I really do because I don't know of any other technology. Wireless networking didn't invoke a personal emotional response. This technology, though, seems to mix personal politics and personal emotions, and I don't know.

Some people are actually having sort of identity crisis around this idea. Well, let's call it what it is, John. Some people are threatened.

They're threatened. They feel their livelihoods are at stake. They feel they're in competition.

And it doesn't always bring out the best in some people. But anyway, John, we're over time here, man. I got to wrap this up.

Oh, I'm so sorry. I could talk to you all day. I'm so sorry.

No, you don't have to be sorry. I was asking the questions and engaging you. But you know what?

This is the kind of thing, John, where, look, we could talk all day about it. But maybe we'll have you back on. You know what we didn't mention, though, as my AirPods are falling all over the place here.

Itential, where can people go get more information? Absolutely. com.

We will have a strong presence at AutoCon Six in Tucson, Arizona, in mid-November. We will be there. I'll be there in person.

I can't wait to meet everyone. And Alan, again, this is really exciting for me. I really am excited.

I'm going to make note of this on my own website and let everybody know that I made it on Tech TV, so this is great. We appreciate you, man. John, a pleasure meeting you.

Good luck. We're going to take a break here. John Capobianco, Head of AI and Developer Relations at Itential.

You're watching Techstrong TV.