How Long Do Network Engineers Have Left?

Key Takeaways

All right, John, how do I say your last name? That's the only thing I want to ask. I have no idea. We've been saying it all week. We call you John Capo. >> That's perfect. Cappo Biano. >> Cappo Biano. Okay, awesome. Now, I've seen you pop up out of nowhere, which for me it was nowhere, but you just kind of popped up my feed and I see you everywhere because you're talking about AI, MCP, A2A. I'm running out of acronyms. I don't know what else there is. Network engineering, what's happening with AI? Tell me like what are you excited about? So, like kind of take me, man, where did we start? How did you get into this? >> Sure. So, I um I got caught up in the inflection point of Chad GPT 3.5 in November of 2022. >> Yeah. >> So, when it gets released, I was one of the first million users >> and then when they released the API, the API came out and I thought I might be able to connect this with network automation, >> right? It's just another API call. So, what if I could get some data from the network and ask AI what it thinks about it? And it worked. And when that worked, it it felt like visiting my first web page. It felt like my first phone call. It felt like seeing black and white TV turn into color. >> Okay. >> It it it really clicked with me that wa I can just ask what is my default route? >> And it'll say here's your default route and here's the next hop in natural language like you and I are talking. >> Yeah. Yeah. So, I remember when I first saw Chad GBT, I kind of my that wasn't my first reaction. My first reaction was, "Oh, no. Existential crisis. What What's my role now? It can do all of this." But that wasn't yours. >> But why do you think it was that you were more excited versus like we're doomed? >> Well, I see it as um like the computer >> is for human convenience. Now, I'm old. I'm very old. My first computer didn't come with a mouse. >> Oh. The mouse was a peripheral that we screwed into the back on a serial port and now I could move the mouse around. Okay, >> I know it sounds hilariously trivial and I'm really old, but it's about convenience, is it not? >> Mhm. >> So now instead of menus and gooies and clicking and memorizing commands and I can use my own natural intuition and just sus out questions in a conversational way. Now, should we be frightened? Uh, I think we should be a little frightened, right? There's with the great power comes the great responsibility, but it's an opportunity, Chuck. I see this like >> like the protocols we just talked about MCP and ADA. Why they've got me so excited is because they're protocols like HTTP or SMTP or spanning tree protocols. Your code can work with my code because we're developing them with standards. >> It's not the wild west anymore, right? And it's very much like the late '9s, mid90s where protocols suddenly changed the world. >> Right. >> Right. Think about HTTP, hypertext transfer protocol and HTML. What did that do for the world? >> Right. It brought us online. >> Yeah, it did. Yeah. >> So now these new MCP tools are going to bring us even further with artificial intelligence. >> Now, real quick, so I know many people, they heard the MCP buzzword, but they're like, "Okay, I don't have time for that." That was my first reaction. I'm like, AI, AI, AI, MCP, I I can't keep up, right? >> So, real quick, what the heck is MCP? >> Right. So, protocol most important, but it's a protocol for the model like a large language model. >> It's context Windows and it's a protocol that is client server. >> So, what's neat is we can build a server. Let's say this is a good example is a subnet calculation. >> Yes. >> I've seen you cover subnet calculation a >> little bit. Yeah. Right. Well, there's a sevenst step manual process if you want to do it by hand. >> Mhm. >> Or you could go to a subnet calculator. Well, now with MCP, which is 30 lines of code as a server, >> I can send it a Cedar notation IP and it will tell me everything about the subnet. First usable, last usable, number of addresses, broadcast, network, just by giving it the Cedar notation. >> Okay. >> So now I'm not even using an online calculator. I'm using AI to say here's an IP and a subnet. What can you tell me about it >> all through this protocol? Now, what's kind of cool is this protocol because it's client server, we can take VS Code, Visual Studio Code, and make it an MCP client. So now, right in Visual Studio Code, when I open up Copilot, it has access to that subnet calculator tool. >> Okay. So whenever I'm developing code and I need to do a subnet calculation, it just calls the MCP. >> Okay. So that's So you're giving you're basically giving AI a tool that otherwise it didn't have before or we're have to like figure it out with some like hallucination magic and we you don't want to risk that. So with MCP, we're kind of giving it guardrails, giving it a tool it can use. >> Right. >> Now, what's I I keep hearing that MCP is the new API. Is that kind of what we're seeing now? >> It is. And it actually is an abstraction of an API. So when the REST API came out, it it was monumental. It moved us into CI/CD pipelines, network automation, infrastructure as code, all of these things, network automation. But there's still some friction there. >> You still have to know how to say post the body of a JSON against an API. >> Right? >> Now, what I just said there, if that makes no sense to you, that's the friction I'm talking about. You still have to know a little bit of coding to use a REST API. >> What if I could take the functionality of the API and abstract it behind simple natural language that the LLM can interpret the question and say, "Oh, I have a tool for that and it invokes the API call." Um, I've written an MCP for Cisco PITS. >> So, real quick, what is PIATS? Sure. Yeah. So, I know we're like we're trying to people coming up and learning, you know, Cisco and CCNA, they may not even be like at the automation point where they're learning that stuff. >> Right. Right. So, that's a good point. Cisco PITS is a network automation framework that's free and it's written in Python. >> Mhm. >> And it lets me connect to devices and much like a REST API gives you JavaScript object notation, structured text, >> right? >> PITS will parse a show command and give me JSON back in structured text. Now why that's important for AI is because the AI doesn't necessarily speak Cisco CLI >> the output that we get but it certainly speaks JSON >> right >> so I can use PTS is almost like a proxy to get the data in the format I want to then send it to an AI to interrogate it. >> Okay. Okay. So your MCP uh which interfaces with PIATS. Yeah. >> Okay. And what are you able to do with that? >> You can do anything you can think of. >> Okay. So I could say, could you help me understand my routing table? >> That would be what I send to the MCP. The MCP knows that it should run a show command and it's going to run show ip route on the device and get the routing table back as JSON and it will explain it to you. You have a static route to this address. You've got an OPF network to this address. Um, you could add a loop back interface. could you please configure loop back 100 with this description? I'm on network chuck and an IP address and enable the port. So an actual paragraph or a few sentences of words >> to achieve what >> you know you would normally log in and run the command by hand, >> right? So let's say I'm like okay MCP how much how specific did you have to be when you're programming that server for that tool to be useful? Did you have to like tell it like this is the command you'll do when they ask you about like this or that? How does that I've never built an MCP server? >> Right. So you actually so um if you've developed REST APIs, there's something called fast API >> that lets you develop APIs fast. There's a fast MCP development kit. >> Oh, I didn't know that. >> It's called fast MCP. So then Chuck, let's say I wanted to do a get against Cisco Catalyst Center uh inventory >> and that's an API call against an endpoint in Catalyst Center. I decorate that as a tool. I say at tool >> and now that tool has been abstracted and advertised to the MCP client. The client doesn't have to know the endpoint, the body, the structure of the payload. They can just start asking how many devices do I have in my catalyst center inventory. >> The decorated tool will then call the API. >> Mhm. Okay. So, I mean, so what we're saying is that you can abstract everything with the CLI. Is the CLI dead? It is going to be less and less important I feel now that natural language and and network language models and large language models can basically do that heavy lifting for us. >> Yeah. >> Right now I don't think employers are paying or are hiring you because you've memorized every single CLI command. >> Right. It's not necessarily how you interface with it. It's the outcome of building a solid robust network. Right. uh that adheres to good standards. Now, this gets really crazy with MCPs. >> They call them the USBC of software. >> Okay? >> And they're that simple. Meaning, if I wanted to plug in the Cisco Mari MCP, >> I just plug it into my client and now I can start talking to Mari tools. >> So, you don't have to do anything special besides just like saying VS Code, you're an MCP client. Here's the server and that's it. Just go. >> That's it. Off you go. >> Okay. >> Off you go. Big question for people getting into networking. First of all, should they even try? Is there a future for that? And if they should try, what do they learn? Because if if CLI is being abstracted, which we see it happening, is it more beneficial to just go full AI and just play with the integrations? What do you think right now? >> So, I think that there's two sides to this coin. I think there's AI for networks, which is kind of what we're talking about, and there's networks for AI, >> right? the infrastructure, all of those thousands of GPUs and racks and racks of equipment, someone still has to do the networking there. Yeah. And there's new protocols like Rocky V2 when you're and other protocols involved in in these GPU farms. So, I don't think that >> networking as a career is going to diminish. I think it's going to become more and more specialized, >> more and more specialized to support, you know, high performance computing or GPUs or large language models in the data center. Um, I think we still have 3 to 5 years of a safety net. I don't think that it's going to be that revolutionary and that fast. >> Beyond 3 to 5 years, it's really hard to project or predict. But I think that your CCNA and your Cisco certifications, particularly their new automation certification and the CCNA together are are are still a path forward to get into it. I don't think the network engineer um we've seen the change, right? Now you have to know voiceover IP and the right now you have to know Wi-Fi. Now you have to know wide area networks, softwaredefined networks, automation. >> It's it's a it is a difficult career. Like I remember when I think it was 2018 or 2019 like the big thing was network engineers are developers now. Everyone's a developer. And I'm like what? Okay. So is that the end of network engineering? Are we all are the programmers going to come for our jobs, right? >> And like that didn't happen. We just learned a bit of code. We learned how to automate. >> But AI feels different. AI is happening too fast and it feels like it's becoming smarter than we were ready for. >> Do you feel like we're ready for that? Do you think network engineers are ready for that? Do you think the industry is going to adapt fast enough to I don't know get teach us to be in the right positions? >> I think just asking those questions is a good start. >> Mhm. >> Right. Having that discussion with your peers and your colleagues and the communities around us. Um it it is it's it's unprecedented that this technology is here and and it's here and working and and is dramatically revolutionizing our industry. So I think are we ready as a society? I think that the the I think that this is important having the conversations throwing out these new protocols and acronyms to wake people up because there's a lot of people I see asleep on this. They're they're asleep on this, Chuck. They're not paying attention. And and part of it is >> there's so much, right? We're so overloaded every day. There's some new thing in AI that it is tough to pay attention. >> But I would argue that we need to incorporate it in our day-to-day life. >> Like an hour of AI a day >> or as you solve problems, use AI to help you solve those problems. I think the time for skepticism is behind us. I think that everyone who is using AI has an advantage over people who are not using AI. >> Okay? >> Right? And as we have this conversation, if you're not using AI, there are people are there are other people using it. >> Right? And there's going to be a parody gap here. >> And that is what I'm worried about are the people here that aren't using it, that aren't paying attention, that are going to wake up one day to a whole new reality that they're not prepared for. >> I completely agree. Um, so for people who maybe by this interview are starting to get woken up a bit. What's the first step? So you said AI a day, what's the today? What can they do? >> So either get yourself an AI account and and I don't care who with what vendor. There are multiple AIs to choose from. >> I have them all. I feel like you have >> I have them all too. I have them all too. But there are open-source free alternatives. Mhm. >> And I think some of your audience because it's such a worldwide reach, maybe $20 a month isn't feasible or they're earlier in their career, they're working on their CCNA, they're in school and they don't have that kind of money. There's an opensource package called O Lama, >> not Llama, O Lama that you can download for free. Download a llama model and start chatting with chat GP or chatting with AI for free locally and private. So don't let the cost barrier for a cloud service hold you back, right? Look into the open source ways of doing things and and and you know when I say an hour a day, I don't I sort of stop tracking like everything that I do involves artificial intelligence. >> So I guess if if they were to like okay jump into AI doing it every day, let's say they get O Lama installed, what do they do with it? Do they just start using it for organizing their thoughts? Do they try to implement uh it into some kind of networking situation? What should they do? Or just try everything? >> I I would try everything. Here's a lowhanging fruit of SIS logs. >> Who wants to be in the data center at 3:00 in the morning pouring over pages and pages of logs? >> Mhm. >> Take the logs, paste them in a context window, and say, "Help me understand these SIS logs." >> Right? Or if you're looking for an interface flap, were there any interface flaps at between 2 and 3:00 a.m. in these SIS logs? >> Mhm. SIS logs is a great place to start. Readonly activities are safe, right? If you want to ease into this, reading network data and interpreting it with an AI is a totally safe exercise. >> Now, do you trust MCP and any certain AI with configuration, making actual changes? Are you there yet? >> Oh, yeah. >> Okay. All right. >> I'm okay with that. >> I'm okay with that. >> I'm not. >> Now, but here's an example of how MCP can help. So there is a RFC MCP >> the request for comment there's an MCP for that there is okay >> so you know what I can do I can say can you please reference RFC 4954 >> and generate a differ quas policy based on the standard >> and then right the MCP reads the standard and now the AI has been augmented with the standards knowledge and generates a valid according to Hoy oil config >> and it's better than even me doing it on my own because it's referencing the actual standard via MCP. >> Oh, the only thing it's missing though would be the context of the network, >> right? >> So, how do we fix that? >> Well, we can well, we can use other MCPs to help get more context from the network. It's sort of rabbits all the way down. Um I think configuration management uh it's it's a similar challenge with art with network automation. People kind of conflate network automation with change management, >> but there's so much you can do with automation that's safe and that's not related to config documentation, visualization, testing. >> Yes. >> Right. So tests are a great place to start that are non-intrusive and won't take down your network. Um, right. You could write a little pits test and and then have AI help you understand the failure. >> Right. So >> you could do show interfaces. >> Mhm. >> And take the interface counters and say, is this interface healthy? >> Mhm. >> No, it's not. I see drop packets right now. Would you rather look at every field and show interfaces? >> A little bit. Yeah. >> By hand. >> Yeah. Sometimes >> you would. Or would you rather just >> It's because I'm nostalgic. I don't want to give up show IP interface brief. Come on, you could just send that data to the AI and it will do it. So, >> I am as lazy as a human as you're going to find. >> And I say that facitiously, but it's about convenience. >> Yeah. >> Right. The more convenient we can make this, the better the answers, the more accurate the AI. Um, >> I get a sense that there's still a bit of a trust issue with you and generally speaking in AI. Is that fair to say? >> So, for me, I'm like, take it all. But I know my audience's take on AI. They're like, "Oh, more on board with it. I want them to be okay with it." So, that's where I'm at. >> So, I'm glad you're bringing this here because you're you're in it. You You see what's happening every day. You're What are you excited about right now? What's What's the biggest thing you're looking forward to or working with? >> Oh, I I see. So, we've talked about MCP a lot. >> Yes. >> Now, for your viewers here, relate this to the OSI stack. Think of MCP as layer 2. >> Okay. And think of layer one might be GPUs, the physical layer. >> Above that, it's going to be an agent layer. And there's varying protocols, different ways to do this, but agent to agent protocol. >> So what we've been talking about is human to agent. >> Hey, what's the subnet calculation on this IP? Right? >> What about agent to agent? when I have a a communications agent that says sends my emails and deals with my calendar and does uh bookings and interview time >> and then I've got my PITS agent that can talk to my network, a security agent that knows security. So these MCPs are tools >> that we bind to the agent >> and that agent has an agent card on the web that says here are my skills. So, not in the near future, there will be a network chuck agent on the internet that your fans can just chat with >> a pseudo you. Now, what would the MCPs be? They would be all your transcripts, all your video library, all of your appearances. Those would be the MCPs. >> And then you would have an agent that you could talk to the world through. >> I don't like that. >> No, take it away. We're done. No, just burn it onto the ground. Where's the server farm? That's Do you do you want that? >> I think that would be wonderful. >> Why? >> Well, I think I think agents are going to sit like there's going to be a an internet of agents. >> Yeah. >> Doing agentic things together. >> I think it leads to new drugs. I think it leads to improving the human condition. >> Let's take every university in the United States or around the world. What if they all had agents and those agents could share data amongst each other and let's say the Harvard agent has MCPs that abstract Harvard databases, white papers, knowledge bases, whatever, right? Well, now we've just hyperconnected all of human knowledge through agents that humans can talk to and agents can talk to, right? >> That's that's scary. I mean, so I I agree. Like I think when you think about the human aspect and what it's going to bring to humanity, that's incredible. But then I think where do humans come in, >> right? >> Where are we going to do? Um like I don't know about you, but the more I use AI, I feel my my my brain starting to atrophy. Some of my reasoning capabilities atrophy, some of my idea ability atrophies. Do you feel that? >> I It's funny. I feel I have a sense sometimes that I'm relying on it to do some of my thinking >> that has crept into my mind. >> Right. >> Is this an original thought? How much of this thought is even mine? >> Yeah. >> Or is it a hybrid mix of the AI reasoning and my logic? >> Right. Right. >> Um atrophy is a good word. I think that's a strong word. I don't know about atrophying my skills. If anything, I feel like my skills are getting better. M >> um you know I I'm not I did go to school for programming 25 years ago, >> right? But I didn't learn Python in school. I learned C++ and Java and things like that. >> So yes, I can do some really good Python code through automation. >> But more and more that less and less of that code is my own code, >> right? So Vibe coding. >> Vibe coding. >> Are we are we at Vibe networking now? >> Vibe ops. >> Vibe ops. >> Vibe operations is what I'm saying. Okay. >> Hey AI, is my Wi-Fi healthy? Hey AI, is my spine okay? >> Are there any drops on these interfaces? Could you whip up a ACL for this router? >> Right. So, I love that. And I love the idea of like waking up at 3:00 a.m. Hey, something's down. >> Hey, what's down? Can you help me troubleshoot this? Right. What What's actually going on? Oh, the the PSU failed. Go replace that. >> Right. >> Okay. I like that. But at some point, they're we're not going to have anything to do. But I think we all need to think about that. >> Yeah, we have to. >> You know, does that lead to universal basic income? >> Does that lead to >> I keep seeing that thrown around like that's that's the future like 3 to 5 years people are saying. >> Yeah. >> So the other thing is I don't know I find this is going through my head a lot as I move let's say to get on my flight and come here and get something to eat and try to get a space to record the video. I would rather be dealing with robots than humans at this point. >> Right. My hotel room's not ready. >> Checkmate. You got me. >> Right. >> That was frustrating. >> That was frustrating. But if that was a robot that could scan us and go, "Yeah, come on in." >> Uh-huh. >> So, I I I'm excited about the next like right now, it's still intangible AI. >> It's still very ephemeral. It's it's it's really through coding. What about when the when it physically manifests itself in automobiles and in trains and in planes and shipping containers and like it's going to be everpresent. >> So I see I see this meme going through X is do we say Twitter or X? I I hate saying X. >> I know >> Twitter. We're going Twitter. >> Twitter. >> Um >> I see this meme going around like the guy on the bus. One one guy is like AI is going to take all the jobs and the other guy's like AI is going to take all the jobs. Which side are you on? I'm on the AI is going to take all our jobs, right? >> Okay. >> And I'm and I am an outlier, right? Like I I have my own unique uh take on this, but it is it to me it being AI and large language models and natural language and chatting with the computer. It it is like the promise of the last 30 40 years. I feel like we've been on this trajectory and we're sort of reaching the operand of Mo's law and some of these right we're really starting to maximize our compute power and I I again I believe in a utopian outcome >> now whether or not we it's going to cost a lot of jobs >> I think that as yes some jobs will be displaced >> but you know two years ago there wasn't such a thing as a machine learning engineer >> right so new emerging categories of jobs will will come out. >> So, you're overall optimistic? >> I'm optimistic. I'm optimistic. Now, in IT in particular, I love um um Jensen from Nvidia. He believes the IT department of the future is going to be an HR department for agents, right? >> And I sort of can see some of that happening here. >> If I onboard a new agent, does it know the corporate policy? Does it play with your agent? Like it's more of a onboarding exercise than a technology exercise. >> It is. It is like and the more I play with agents and I it is it's not easy right now. It's almost frustrating because there's so many things like okay it didn't know that idiot. Stop it. I have to give it this context. We didn't know. So it's frustrating right now but I know eventually it'll be abstracted. But anyways I know we're probably being pulled for time right now. John, thank you so much for this amazing interview. Um where can people find you? Yeah, find me on my YouTube is probably the best place and I have a ton of stuff on MCP and AAA uh LinkedIn X, Twitter and um feel free to reach out. I I I am open to communication. I would love to help you on your journey and I really want to thank this guy for having me on today. I've been following him for a long time and it it was heartfelt when he reached out to me to do some content together. So, thank you. >> Oh, it was my pleasure. So great to meet you, John. >> Thank you. Yeah. Yes, sir.