The AI Shift That’s About to Hit Networking | The Bearded IT Dad

Dakota (Host) • 00:00

Right now, there are network engineers who are about to become wildly more valuable. At the same time, there are ones who are just quietly becoming irrelevant. Not because they’re bad at their jobs, not because they don’t work hard, but because the way networks are built, operated, and fixed is being rewritten right now by AI and automation. My guest today didn’t just see that shift coming. He leaned into it and built his entire career on top of it.

John Capobianco • 00:28

I firmly believe that automation is much safer, has much less failure, much less risk than human beings doing their work.

Dakota (Host) • 00:37

Today, I’m joined by John, one of the most respected voices in networking, automation, and AI-driven infrastructure, and someone who’s just joined Itential to help shape what the next generation of networking looks like. Today, we’re talking about how he made that leap, how AI is already running parts of real networks, and what that all means for anyone trying to future-proof an IT career.

John Capobianco • 01:07

Thank you, Dakota. That was an incredible introduction. I’m really excited. I love your message. I love your positivity. I love the casual nature of just having a conversation with people in tech. And I think we can help some people today — maybe inspire some people to try some new things.

Dakota (Host) • 01:38

The general conversation around AI is typically doom and gloom when it comes to careers. But I feel like it shouldn’t be that way. AI is a really useful tool. People who are willing to accept it and learn to adapt can really use it to accelerate their careers beyond their wildest dreams.

John Capobianco • 02:02

And it rhymes with network automation. For 10 years, we heard: network automation is going to automate you out of a job. Practically speaking — because I’ve lived this, I did this in production as part of my real career — every time I automated something, I was rewarded for that. They said, wow, you automated all the documentation. Could you automate the testing? Could we automate configuration management? It’s going to lead to more opportunities. And then you become the author of those automations. Similar to now — you’re going to be the author of the agents.

John Capobianco • 02:45

So our careers might shift, and now there’s an entity in between us and the network. Instead of us directly configuring or monitoring networks, there’s going to be an intermediate — an agent that we build. We build the agent, we talk to the agent, and the agent autonomously does all of those things humans used to do directly against an interface. It’s just the natural evolution we’re seeing.

Dakota (Host) • 03:27

It’s just the next phase of the tech industry. Things aren’t going to stay the same forever. The only people who truly have to worry about AI replacing their jobs are the people who aren’t willing to evolve and continue to upskill.

John Capobianco • 04:05

For me, artificial intelligence was an evolutionary step on top of what I had already established in network automation. When ChatGPT 3.5 came out in November 2022, there was already 10 years of network automation history. Augmenting what we had been doing was a force multiplier. I was an AI skeptic — it was kind of neat. But when I revisited it with an API key and sent interface state data to the AI and asked “are these interfaces healthy?” — and it said no, named three interfaces that weren’t healthy, and told me exactly which counters it detected — that was an inflection point. I immediately saw the value.

Dakota (Host) • 05:44

Before you were the AI automation guy — what did your day-to-day actually look like as a network engineer?

John Capobianco • 05:54

At Parliament, the early part was greenfield: architecture, design, IP address layouts, VLAN assignments, zoning. Then it switched to operational mode — keeping up with growth and demand. We had a change involving 54 routers. It was going to be start Friday at six, hopefully done Sunday — outages, problems, a lot of human copy and paste. We asked leadership to delay the change three weeks so we could do it in the lab with an Ansible playbook.

John Capobianco • 08:01

The operator runs it, and maybe three minutes went by. It reached its end. I started to apologize — I didn’t know what went wrong. The operator said, “No, hang on. It worked.” We expected 20 or 30 minutes. Three minutes to push, test, and automate the entire change. No rollback needed. We were cheering like we’d won something. It would have taken six or seven of us three days, with service disruptions. That was a real triumph.

John Capobianco • 09:41

I also have the opposite story. Our first time allowing Ansible to touch the core of the network. I personally guaranteed leadership there would be no problems. The pagers went off. The core went down. Someone had made an out-of-band change between when I wrote the playbook and when we executed it — and it didn’t get updated in the source of truth. When our idempotent playbook ran, it wiped out their fix and reintroduced the problem. Twelve hours of figuring that out. That event held our automation journey back six to eight weeks. But even so: automation is still much safer than the notepad-per-device approach.

John Capobianco • 16:21

After my first API call with PyTS, I made conscious decisions to move out of the Cisco training team and into the Cisco AI Technical Leader role. Then to an AI startup, Selector. And now to Itential to do pure AI agents and agentic operations. Where I used to spend hours studying for certs, I now put those hours into RAG, knowledge graphs, and MCP.

John Capobianco • 17:25

MCP is only 16 months old. If you’ve never run an MCP, you’re not that far behind. Andy and I just did a Hello World on the Art of Network Engineering podcast — installed the Netbox MCP, connected it to Copilot. First Git clone, first Python virtual environment. Then he asked: “How many Netbox circuits do I have in New York City?” It came back with all of them. That’s a whole new way to interface with technology.

John Capobianco • 19:32

I built NetClaw — a ReAct agent based on OpenClaw — and it already has over 300 stars on GitHub in just a month. People around the world are using it to talk to their networks through WhatsApp, Telegram, Discord. Enterprises are using Itential’s technology to completely close the loop on network automation. Agents are first-class actors. FlowAI provides the framework for that.

John Capobianco • 20:21

Should engineers stop learning traditional networking? Fundamentals — Git, Python, subnetting, routing protocols — are the foundation that makes AI prompts better. My sister-in-law can’t design a network just with an AI. You need grounding. The Google Developer Expert credential I just earned required no exam — just a 45-minute interview and my GitHub, YouTube, and body of work.

John Capobianco • 36:52

Something to look for on the horizon: spec-driven development. GitHub has a spec kit you can install into Claude Code. The first thing it builds is a constitution. From the constitution, specifications — user stories. From user stories, tasks. From tasks, code. All natural language, all constitution-based. It moves us beyond vibe coding. William Collins at Itential just did a full write-up on it.

John Capobianco • 40:50

Gartner believes that by 2030, 70% of IT tasks will be automated or have AI performing them. Someone needs to build those agents. Our jobs will shift: instead of managing devices, we manage agents that manage devices. Onboarding new agents. Performance reviews. Making sure they play nice. It becomes more like an HR exercise than a CLI session.

Dakota (Host) • 43:05

You’re orchestrating the agents. You’re going from orchestrating the network to orchestrating the agents that keep the network running.

John Capobianco • 46:25

If you’re the person who brings an agent demo to your company and says, “Look what I’ve done” — you become extremely valuable. Companies right now are looking for AI leaders, AI prototypes, proof of concepts. Imagine the security you’d offer yourself if you were the person that knew AI in your org.

John Capobianco • 49:00

Every couple of days I search “artificial intelligence cancer research.” In Canada, they’re using AI with colonoscopies and false positives have dropped 70% in six months. Seven out of ten people who used to be told they had cancer by mistake. That’s one example out of thousands of positive stories. Try to find the positive, cut through the noise. When Dakota lights up about the apps he’s building, and when I light up about NetClaw — that’s genuine enthusiasm for achieving things we never thought possible. I built NetClaw in a weekend. Find something that excites you, and build it.