Intro

Andy Lapteff 0:16

This is the Art of Network Engineering, where technology meets the human side of IT. Whether you're scaling networks, solving problems, or shaping your career, we've got the insights, stories, and tips to keep you ahead in the ever-evolving world of networking. Welcome to the Art of Network Engineering podcast. My name is Andy Lefta, and I am joined by the founders of so many things. But today, the Network Automation Forum, we have Chris. Is it Grundemann?

Chris Grundemann 0:43

It is Grundemann, yeah. Hi, Chris. At least that's how I say it. I don't know. My ancestors may hate that, but uh it works for everybody in Europe. That's right.

Scott Robohn 0:50

And the other voices? Scott Robon, same thing. I don't think Robon is really the way they pronounced it centuries ago, but we're here in Munich, Germany. We are. We're closer to whatever German roots we have, however we mispronounce them. Exactly.

Andy Lapteff 1:03

Yeah. So quick fanboy moment. Scott and I have done a bunch of podcasts together. I finally have the Chris Grundemann. Like, listen to his radio voice. Like it's so there's only been like one other person on the show, I think, that had such a built for radio voice. Like I could just listen to you read like Catcher in the Rye. You know what I mean?

Chris Grundemann 1:20

Like, or the phone book.

Scott Robohn 1:22

Remember when we had phone books?

Chris Grundemann 1:23

The phone book, yeah. Well, that's about when radio was still cool. That's right. That's right.

Andy Lapteff 1:27

For radio. So we're here in Munich, uh, Autocon 5. And I thought it'd be cool to get the founders of NAF and this amazing conference that's happening. I mean, I've been recording podcasts here all day, and pretty much all we're talking about is network automation and the gap and why it's a 70-30 split, your data has shown. And so these conversations are awesome to have because without them, we can't surface the root cause, right? How the hell do we fix this and push the industry? So, your organization and these conferences, in my opinion, and are doing more to try to push this conversation and the industry more toward it than a lot of other things I've seen, you know, of anything I've seen. So, why did you start NAF? Like, why did you start it? How did you start it? Did you have

Founding the Automation Forum

Andy Lapteff 2:07

any idea that like upwards of a thousand people would show up at these conferences?

Scott Robohn 2:11

Let's start with your last question first. No, um, no idea. Yeah, but you know, I'll I'll take the first cut, right? We get introduced through a common friend. I think I knew about Chris this three, three and a half, four years ago. We'd never run across each other really, but we both had interest in what's happening or not happening with network automation. Right. We started talking and we just said, hey, you know, we're two people with our opinions. We should probably try to get other people talking about it, right? And that turned into Audacon Zero, November 23 in Denver. And uh we set a bogey of like 200 people. If we got 200 people, we thought that would be successful. Chris to you.

Chris Grundemann 2:51

We had 340 people show up, 345. Wow show up with the first one. And and and that's a wild experience for sure. Doing something, uh, you know, I've done some community events and things before, but something on this scale where we really uh threw the idea out there and set and set a target date, which I think was

AutoCon Zero Success

Chris Grundemann 3:07

less than six months from when we decided to do it. We it was about six months from when we had met for the first time. Guys are cowboys, and said, you know, here's the date, you know, we're gonna do it. And then and then sponsors showed up. Yeah. And and we said, oh shit, now people have to come, but we're really on the hook, right? Uh and luckily people showed up, we had great content. I mean, it was amazing. And and it really has just exploded, we had no idea it was gonna explode like that. Where was Zero? Was it Austin or not? In Denver.

Scott Robohn 3:31

And we we tapped into something that already existed, right? We it became clear to us pretty early on that there was like there was unmet demand for people who wanted help with network automation and wanted to help

Unmet Demand for Automation

Scott Robohn 3:44

each other with network automation. If you think about the name, you know, network automation forum, it's a place for people to get together to talk, right? Using the traditional concept of forum. We also have used Oasis in the past, right? Where like the watering hole. Yeah, where the where are the guys keeping the area clean, you know, dusting off the seats so people have a place to sit? They're primed to already talk about this and support each other in this.

Andy Lapteff 4:08

Yeah. Why did you care about network automation? And you know where it's coming from from me, but to to have enough passion in it to create an organization around it, throw these things and really do a ton of work. I can't imagine how much like why did you want to move this forward in the industry? Because

Why Care About Automation

Andy Lapteff 4:24

like there's a gap, it's a thing, and everyone else has failed.

Scott Robohn 4:27

I will let me key this up for Chris Chris to give the give the mic drop answer.

Chris Grundemann 4:31

NAF is the only organization that continues to ask the question. Uh, why haven't we seen full adoption of network automation?

Andy Lapteff 4:39

Yeah, but did you guys practice this right before? You know, we're but we're we swim, we swim in this, right? So like we're kind of you know on the same page here, you might guess. But why do you care? Do you know what's behind the question, right? Sure. I mean, sure. It's not that nobody cares, but you guys are investing like a pound of flesh to try to do that. Yeah, literally, right?

Chris Grundemann 4:55

We're really doing this ourselves. More than um, yeah, yeah.

Andy Lapteff 4:58

But there's vendors with certifications as an example who are trying to like make this, you know, that now the the Cisco track as an example has a lot of automation in it. So the industry is moving, but the adoption like it's been very, very, very, very slow, right?

Chris Grundemann 5:11

And I think to me, automation, like well, so I didn't know what it was at first, right? I mean, I so I I started building networks. That's right. And and shell scripts. Like, that's how we built networks on the service provider side, right? Um, now it was very primitive, it wasn't, it never really kind of evolved beyond that,

Three Eras of Networking

Chris Grundemann 5:27

but that was kind of how we did networks. So then to me, my first deluge with like automation, you know, in quotes, was SDN. Right. And and really was OpenFlow coming out. Sure. And that kind of thing. But but for me personally, and I know Scott, you have your own reasons as well, but for me personally, automation is really about finding ways to help people who actually operate networks, right? So I kind of split it into these three eras, and I've probably talked about this before in other places, but I think there's, you know, this first era of networking, which was maybe the first 30 years or so, was this gonna work? Can we actually do this? Is there a standards-based way to create global communication? Like, I mean, there's a number of things that go into this, right? And at some point, probably around the 80s, we realized yes, there could be a global internet based on TCP IP, right? And competing protocols kind of fell away. And then we said, okay, well, how do we make that work for the users? Right. So how do how you know, how do we make it work, you know, better? And and that lasted all the way up until for me, I pinned that about when we got working Wi-Fi on international flights. Uh at that, at that point, the internet has kind of disappeared for most people. It's kind of like power. I go to the wall, I flip the switch, the light comes on. I don't actually think about whether we're burning dinosaurs or not to make that light come on. And and same thing with with the internet, right? I I don't even plug into anything anymore. I just pull my phone out of my pocket and use it. And I expect it to be connected to everything else on the planet. Well, that's great for the users. Now we need to, I think, shift our attention towards the operators, the people who are actually running networks and make their lives easier. And I think the way we do that is with automation. And if we can't get companies to go their head wrapped around how to actually use automation, then people are gonna just continually just be struggling with the same things I was struggling with in my early career 20 years ago.

Scott Robohn 7:06

Yeah. And just to layer some things on top of that, right? To use a wireless example, right? Think about how easily you flip back and forth between Wi-Fi and public 5G, right? It's it's invisible now. We've automated that. And that's like that's protocol level behavior, but we really have made that automatic. Remember when you had to worry about it and you weren't quite sure?

Chris Grundemann 7:28

And we were even trying to zoom call would drop if you turned the white. That's right.

Scott Robohn 7:31

Yeah, that's right. And that that almost never happens, even you know, any country you go to in the West now, right? It works just as well here in Munich as it does back in the world.

Chris Grundemann 7:40

You can be on Virginia that's right through cells.

Scott Robohn 7:43

Unless unless it's Amtrak. Amtrak really needs to do a better job. There is something unique in network engineering where there is a lot of resistance. Um, and there's still this, you know, the tread networker

Adoption as a KPI

Scott Robohn 7:56

mentality where I I really like cutting and pasting and I like the swivel chair. It's still there to a very large extent. And um, that's why you know we haven't changed the theme of of the program, right? You know, we keep asking the question why haven't we seen the full adoption network automation? We'll adjust it when we think we've we've hit a limit. We'll make it a lot, right?

Chris Grundemann 8:18

Yeah, the questions come up. They said, well, you know, after after the first event, they're like, hey, what's what's the tagline for the next event? And I said, Well, until we see full adoption of network automation, the tagline remains.

Scott Robohn 8:27

Yeah, yeah. I'll I want to throw a couple other specifics in there, you know, align with Chris's journey. My first network automation project was in 1996 at Bell Atlantic, using Expect scripts to log into Bay Network's routers and then Cisco routers. And so, like, so for some of us who've been doing it for a very long time, pieces of automation have always been part of what we're we've done, right? But how do I how do I do it at scale? How do I get to that, you know, Wi-Fi to 5G automatic behavior, right? And we're miles to go before we sleep on stuff like that.

Andy Lapteff 9:00

I don't know what's wrong with me, but I have three questions. No, it's okay. I can never just ask one. Well, so circling back, or let's start at the end. Um you're you're you have you're keeping score. You have questions. Well, yeah, you guys you guys say awesome stuff, and then it makes me think of things. So um back to like KPIs for lack of a better example. Um, you know, we won't stop until we see the full uh adoption of network automation. Like, how are you measuring that? Is it the survey?

Chris Grundemann 9:25

So we do do a yeah, we have a survey um that actually predates Autocon a little bit and NAF a little bit, but but it actually has come into its own with the SNAP community, right? I mean, that that survey, I think the first year it was like 36 people responded, right? And like maybe it was like 76 the second year. And then last year we had like 400 people respond. So we're gonna pumping it with the community. Exactly. Yeah, because the Snap community is there and there's a place where you can actually go ask these questions and people are responsive, it works really well. Now, hopefully we'll get even more people to respond this year because there are like 10,000 followers of our LinkedIn page and several thousand people on the Slack channel and a couple thousand followers on YouTube. So, you know, 400 actually sounds pretty low to me as far as people responding to the survey. But that is one way to take to take tally. The other way is just, you know, here at the events in the hallways, right? Talking to people, and there's so many people who come here and they're like, I'm here because I need to take some of this back so that we can get started. So many people, right? I mean, not everyone, but uh like close to half of the people who are here, yeah, and this is very anecdotal, but close to half the people I talk to at Autocon are telling me, I'm I'm trying to go back and start an automation practice where I work. I think that's better than than validation, right?

Scott Robohn 10:33

Like if you if you have a sense that half the people are trying to take away and and come up with ideas that I can try, you know, that that's we don't need to question that, right?

Andy Lapteff 10:42

I think I think that's a pretty clear, clear drive for folks. I have a comment on, I don't know if I'm using the term correctly, but like self-selecting, right? So the people that are here aren't the Trad NetOps folks that we're trying to move. And I'm assuming based on the numbers you just shared, the people answering the survey probably are the 100% involved here.

Chris Grundemann 11:04

Everything that we see, it's worse. Right. It's definitely worse.

Andy Lapteff 11:07

Right. Yeah, yeah. So uh it's stating the obvious, this isn't even, I guess, a question, but where this started is you said make net NetOps lives easier. And my question is how I have all these preconceived notions and biases about how I think, you know, what the benefits are. I think for me, it would be saving time. Why spend three hours on something I could do in five minutes? Right.

Chris Grundemann 11:28

Well, I mean, yes and no, though, right? Because like you're still gonna work eight hours a day, whether or not you save those three hours into five minutes. So it's what do you want to be working on? I think is the real question, right? Like this doesn't actually save you any time, it reallocates time. Well, maintenance windows is an example.

Andy Lapteff 11:42

Fair. I think it could save a ton of time in maintenance windows. That's very true. I would say I would spend three hours working through a 24-page notepad plus plus script I put together. Well, what I mean is like if you like I could push it in a Python script with like one button and do some validation and like so like, oh my god, right?

Chris Grundemann 11:55

Yeah, but then there's a time up front. And maybe you don't even need a maintenance window,

Reallocating Time Not Saving

Chris Grundemann 11:58

right? Because now you're you've you've actually validated this, you've tested it, you know it's gonna run safe, and maybe you don't even need, you know, some things are still gonna need a maintenance window for always, forever, but some things maybe don't need a maintenance window. But but my point was uh that's a time because like I don't know about you guys. I mean, I'm sure some places are worse to work than others, but when I do a night maintenance, you know, I got a comp day the next day. I don't have to come into work. Yeah, right. And so again, I'm not actually saving any time. Yeah, I'm reallocating time. Yeah, that's a good point. And so what I'm reallocating time to is architecture and and like actually understanding what the network is for. Now, here's the thing, not everyone wants to do that, and I get that. Yep. And that's where some of this conflict comes in, but you get to kind of choose your path. If you if you start automating the stuff that you don't want to be working on, you get to spend time doing other stuff that you do want to work on.

Andy Lapteff 12:42

So, is that how we make their life easier? I know you answered it, and then maybe I'm tired. So it's not saving time. I'm reallocating time.

Scott Robohn 12:48

I want to I want to put a really fine point on that. You are saving time on specific tasks. I I like the reallocate time, you know, uh verbiage. Like I definitely agree with that. But it's not like you're going home at 2 p.m.

Chris Grundemann 13:00

every day now, right? Like that's well, maybe, maybe either way you were. It depends how good you are, right?

Scott Robohn 13:05

If your automation is really good, let's just go straight to universal basic income. No, just kidding. Right. This idea of elevating how I think and how I operate, and in shorthand, you know, Damian Garrus is out talking along these lines, thinking more like an architect, right? Now I I'm the puppet master and I can pull different strings, and I don't have to babysit the upgrade. Like if I do, if I automate an upgrade right for three different vendors, right? I can abstract everything that needs to be done with pre-checks, with getting the file, with making sure the file checksum works, making sure there's space to storage on the vice, checking the pre-checks, then actually doing the upgrade, then switching to the new version, et cetera, et cetera. Like if if I don't have to babysit that anymore, I can think more about capacity planning. I can look at more new things coming at me, right? From a from a bigger picture perspective. So there's there's a a very clear ability here to think more about bigger picture and what else can I do in the network and how can I mature my network infrastructure more?

Chris Grundemann 14:05

But sometimes it's not even that, right? And I don't want to be too contrarian here, but so I I think a good analogy for me in some of this, right? And there's a there's a lot of areas here, there's a lot of shades of gray, but in some of this, it's you know, so let's say I I go by a Tesla and I've got self-driving, right? Now here's the thing I don't actually save any time.

Scott Robohn 14:23

Right.

Chris Grundemann 14:24

I'm in the car, I'm doing the whole trip. But you know what I don't have to do is be stressed out, paying attention to every little thing. I just get to lay back a little bit. I got my hands on the wheel still, I got my feet near the pedals. Yep. The car's driving, but I'm there to take over if I need to. So I'm it's just a much more comfortable, much more relaxed experience. Now, I haven't actually shortened my day. I didn't shorten the trip, but what I did do is improve my quality of life. Yeah, you know what I mean? I I love that. You've reduced your cognitive load. Exactly. Yeah. And now I can actually pay attention to the things that matter instead of everything all the time, right? It's not fire hose anymore. It's it's sprinkler.

Andy Lapteff 14:57

This is also obvious to do. Have you guys over the time been able to identify like, is it fear? What's holding back? I know my story that I've talked about ad nauseum, so I won't. But I failed out of computer science and thought I wasn't smart enough to do this. Now,

Fear and Identity Barriers

Andy Lapteff 15:13

I was just interviewing the Bart. I don't know if you know the Bart. They're awesome. Bart one and Bart two. But we were talking and which, though. Same. Well, uh well, I'll tell you later. Bart Smeadling makes stinky eyes at me. He has this very grumpy face he gets, so I know Bart Smeadling.

Scott Robohn 15:26

Actually, it's a very it's a great example of any cast, right? You just you send a packet to Bart, and the topologically closest Bart Bart will answer and provide the service.

Andy Lapteff 15:35

So what I wanted to get at is something that I talked to the Bart about. You know, I spent eight months in Cisco Neticad spending, I don't know, roughly 10 to 15 hours a week studying the Cisco CLI. Sure. And I had a guy I worked with, uh Chris, who said, dude, if you can do like a two dozen page notepad plus plus in CLI, like that's programming, dude. Like Python isn't that different. But to me, it seemed completely not related. And and I thought as I was speaking to them, I have not put in the time learning Python that I put in learning CLI.

Scott Robohn 16:08

We remember, we remember you living that very publicly. Well, right. Chris and I had some conversations about this laptop guy. We need to talk to him. And I'm not I'm not kidding.

Andy Lapteff 16:18

No, but I never put the time in. Sure. And and then what I was also saying to them was I was also working four or five maintenance windows a week and I was burnout, and I was also studying for the MP at night. And I was also trying to get my AWS practitioner because cloud, like, you know how it is now. So there's all the things, and you might have a family with kids, and it's like, when am I supposed to sleep? So Python just wasn't priority for me because I thought I could hide from it until I couldn't. I don't know where I'm going with all that, other than to say I'm trying to figure out what that constraint is for people. Is it fear? Because you know, our buddy Mike, like he talks about the fragility of networks. Sure. We break everything without automation, and then we use change freezes and process to create the illusion of you know reliability. So I am partly afraid of automation when I was in production because I'm like, oh my God, I'm gonna move faster. I break things slow because of technical debt and all the reasons and all that. So I guess you guys haven't identified the thing. If you have, you would have solved the problem by now.

Scott Robohn 17:14

But what's holding them back, right? So I'll throw like one really good example. Um, is this guy from Lumen, Greg Freeman, who spoke at Autocon 4. They're nothing fast about the automation that they've driven in their network. That's a multi-year effort where they've

Tools Downstream of Software

Scott Robohn 17:30

defined over a thousand workflows that they're doing, taking very carefully over time. You talk about Python and being afraid of picking up new skills, right? I don't think anyone here is saying you need to turn into a programmer to operate a network. However, I will never be a carpenter, but I know how to use a hammer and a saw and nails and screws. I know how to use the tools for the jobs that I want to do, but I will never professionally be known as a carpenter. There are tools that are downstream from software engineering that are really, really, really, really, really, really useful. I don't know how many really I use there. But like the tooling makes a lot of sense, right? And like learning Git and learning Python and understanding what a pipeline is and and reducing your ego to the point where somebody does need to check your configs before they're pushed into production, just like code is checked before it's pushed into main, right?

Andy Lapteff 19:23

I love that tool analogy. I'm putting an attic floor in right now in the dead of summer. Not the smartest thing in the world. But I worked a couple years in construction and I miss working with my hands and I'm really enjoying it. The fresh cut of wood and you know, hurt my hands like I I can smell it now that you've said that. Yeah, I know. Working at a desk is just not as satisfying as like building stuff, right? And for some reason, for most of us, those tools either seem inaccessible or out of reach or too complicated, or maybe there's too many of them because first it's Ansible, then it's Python, then it's Rust, then it's Git, then it's I love this organization and the conferences because I don't think there's a clear answer. It's not a one size fits all.

Chris Grundemann 20:03

Well, there's a couple of but how do you solve it? Like there's a couple pretty major challenges, though. I mean, I think we're further down the road or half, so at least on the identification side, right? I think one of the big ones on the individual level, especially, is identity.

Speaker 3 20:14

Right.

Chris Grundemann 20:14

And and this is something that I think is just a fascinating topic in general. I I can't get enough of understanding like psychology and sociology and people's minds about this kind of stuff. More than once. Yeah. And and I think in this case, right, what's happened, and you know, Anna Claiborne in her keynote talked about this a little bit, the certification track and how this kind of led to that, right? So we don't, you know, I don't think we can blame Cisco for the lack of automation. We can.

Scott Robohn 20:36

That was Autocon 2.

Chris Grundemann 20:38

You can go look at it on like on YouTube. And and this is a part of this, which is that so what what does the CCIE really say? Well, what that means is I can hammer CLI commands really fast. Yeah. That's what the CLI tests. Now, there is networking knowledge in there as well, yes. But the test itself, and most of the lead up to it, is how fast can you bang out CLI commands? And for the CIA itself, right, half the day you spend building, and then they you leave for lunch, they break it, you come back, and it's how fast can you troubleshoot on the CLI? So the the test itself, the pinnacle, the PhD of network engineering is how fast can you type CLI commands? And so of course you get your identity wrapped up in how fast you can type CLI commands. And when somebody comes along and says, you don't have to type CLI commands anymore, you say, Well, then who am I? Right. Yeah. Then who am I? Yeah. What do I do? What do I do here?

Andy Lapteff 21:29

There's so much sunken cost into those. I was I didn't make it to the IE, but I spent, it seems like an enormous amount of time learning the craft, which was CLI driven. Yeah. And yeah, to have somebody tell me that's not like my entire C CNA and CCMP years of my life, it wasn't all CLI, but to your point, that's the thing. That's what you're tested on, that's what's in the quiz list, that's what's in the troubleshooting thing. And because of what my buddy Chris said, uh uh Chris Margette, Python shouldn't they seem tangential. Like, why does Python if if I spent 15 years being a

CLI Sunken Cost

Andy Lapteff 22:08

CLI expert typing them out fast, why am I so threatened by Python? I don't know the answer.

Scott Robohn 22:13

So so there are new It's more open-ended. Yeah, there and there are new threats entering the room. Like, and we should talk about that, right? Two letters. Yeah, that's that's the drop the AI bomb, right? And get that out of the way. So on Python in particular, why Python for network automation? It it's really convenient, it it's very adaptable. It's easy to learn, right? It's easy, it's easy to read and learn, but it's not the only thing like we we Perl scripts expect. Like we've all done other things to script things.

Andy Lapteff 22:41

Well, you guys came up scripting, which is interesting to me. I didn't. Yeah.

Scott Robohn 22:44

Service provider, I guess telco, If you go way back, like there's a convergence point way back, I think, that predates both of us where the original network engineers were typically computer science people, right? And they kind of forked out from there and did the specialized thing. But Python isn't the only way to go upon completely intended. Right. Go is also an alternative that you know Google is for for very

Why Python for Automation

Scott Robohn 23:08

important reasons really interested in, right? And now I've got two of them.

Chris Grundemann 23:12

You could be an actual programmer and use like C or Yeah, sure.

Scott Robohn 23:15

You could and I'm There's a there's a guy I follow on X, Uncle Bob. He's a developer and I don't understand most of what he's saying. But um, you know, what languages he uses for different things, right? I made sure that I got C as part of an independent study in my undergrad because I kind of knew that this was going to be important in the 1980s. But uh now I have tools that can write the code for me. And is it as important for me to know the Python, the go, whatever the syntax of whatever the underlying language is? And like six months ago, I was I was I was saying things like, Yeah, you shouldn't drop your Python studies yet. Right.

Andy Lapteff 23:55

I know.

Scott Robohn 23:56

Now I'm not so sure. Yeah. Right. Because you know, it's certainly useful, right, to know, okay, I can actually check the output of whatever this model is saying. Um, but there are ways to fact check that more.

Andy Lapteff 24:09

The reason I'm signed up for two days of packet coders, Python stuff coming up is I really want to be Rick Danato. Hey, I'm just thrilled to be, you know, trying this for the hundredth time in my life trying to learn this stuff. But I'm I keep getting closer. I'm you know, I'm chipping away. Right. Um, I think I'm over the identity part, which honestly was a big part of it. Um and no small and my identity was crushed going through what I went through. I'm like, oh, okay, I need to start over. And that's what happens, right? We get torn down, they get I get it gets built back up again.

Scott Robohn 24:38

I think we can look at like previous generations, maybe maybe your parents, right, who like saw their careers disrupted. Yeah. You know, my mom was an electronic parts drafter. You know, she started out in an architecture school that that didn't work out for her, but her she had great skills as a drafter, and CAD just kind of passed her by. Like she never really got into you know, using CAD for producing those drawings. Um, and she had a career change as a result of it. Right, right. But stuff comes at us, and there's multiple ways to adapt. A career change, or you know, I'll invoke REO Speedwagon again, you know, roll with the changes, right?

Andy Lapteff 25:17

Um how would I know that the code that I'm generating with like say claw code as an example? You know, so some of those folks who have the

Careers Disrupted by Change

Andy Lapteff 25:26

IE numbers, sure, some of them I've seen are a bit bothered by by AI and LLMs because again, it's kind of doing their job in ways that are better than them or faster.

Speaker 3 25:39

Yeah.

Andy Lapteff 25:39

So I six months ago was talking trash on vibe coders because I'm like, of course you have to understand the fundamentals. This is insane.

Scott Robohn 25:46

But vibe coding isn't the only way to do this.

Andy Lapteff 25:48

Well, I have been able to build so much amazing stuff with clawed code. I guess I'm vibing because I don't understand it. So really where I'm getting at is the tooling is amazing. I love it, and I can build things I could never build before. And I have no idea if it's slop code or not, because I don't understand code enough.

Chris Grundemann 26:02

And and is that a problem? Well, this is a thing that like I see, you know, one of the most visible aspects of AI is content stuff, right? People, people writing blog posts or or whatever they're doing with it, right? And and and I've seen this, you know, we tend to jump to like, oh, that's AI slop, right? And I'm like, okay, well, well, hold on. If it's slop, it's slop. I don't know that I actually care whether the like if the k if the person who wrote this actually wrote it themselves, is that better or worse than if they used AI? Because it's terrible either way.

Scott Robohn 26:29

Yeah, but whether it's human generated or actually matter, yeah, right.

Chris Grundemann 26:33

And and and and so, but but so what's nice though is with code, is it's not just about reading pretty sentences, it's about actually doing something. Right. And so if

Vibe Coding with Claude

Chris Grundemann 26:42

the code does the thing, does that mean it's good pretty good? Well, I I mean I know it doesn't as opposed to bad, yes, right now. Is it secure? I don't know, right? But but but if one of the jobs to be done is to be secure and it is secure, then do I care how it was written?

Scott Robohn 27:00

I mean, I don't know that I do.

Chris Grundemann 27:01

I don't, I don't know that I do.

Scott Robohn 27:03

So and and we don't have enough time to to dive deep into this. I know that we're all playing with different aspects of this. You know what we've in a very short period of time, we've gone from you know vibe coding as an enabler and prototyping really cool stuff for people who couldn't write, you know, anything to save their lives, to now using models and different models to generate and then check. Um, and like it's not it's not a simple linear process. The whole idea of iteration with code development or even generating configs for your network, right? You try something, you learn, you fix it on the next iteration. You know, we're seeing a lot of that be automated, you know, through agentic use of AI um and just countering models countering each other for you know, checking outputs.

Chris Grundemann 27:50

So and and and this is where like some of the like actual like programming skill and some of the thought process and stuff comes into it is that LLMs are by by nature non-deterministic. If you want to have consistent results, you need to build some some some framework around that. There's some companies that are doing some work there. If you want to play with it yourself, you should probably be doing some work there, right?

Scott Robohn 28:10

Because it's getting better with each with each pass and with each, you know.

Andy Lapteff 28:14

What does that mean? I hear people say LLMs aren't non-deterministic. I don't know what that means. You can give it a same a prompt. You get different stuff each time.

Scott Robohn 28:21

Copy and paste a prompt, and like I'll get one output on one.

Chris Grundemann 28:23

You're not guaranteed to get different output every time, right? You're not guaranteed to get anything consistent. And is that just the way it's built? Like, I would 100% is the way it's built.

Andy Lapteff 28:30

It's math you would get the same stuff on the case.

Chris Grundemann 28:32

No, but it's but it's but it's calculus, it's stochastic. Oh Jesus, another thing I failed. Right? Yeah, it's stochastic, and so it's never too late. It's always there's always a range. Uh-there's always a range, and there's always a margin of error around that range. And so you

LLMs Are Non-Deterministic

Chris Grundemann 28:45

cannot guarantee ever that you will get the same answer to the same question.

Andy Lapteff 28:49

So if you're generating complicated BGP configs, you might want deterministic yourself, right? Like, yeah, exactly.

Chris Grundemann 28:54

Well, and that's where you got to build the like you got to build a system around it, right? The LM cannot be the only thing. Yeah.

Scott Robohn 28:59

And if we're comparing human language to, you know, let's let's use Genos because we're two J N C I E's, and that's what's really matters here. Okay. No, it's all good. There's much less room for semantic wiggle room, you know, in a router config than there is in human language. So it's not the same set of broad, um, broad issues. Now, it's gonna it it takes one form in iOS, it takes another form in SR Linux, it takes another form in Junos, it takes another form in EOS, right? Equal opportunity, you know, we we can't cost.

Andy Lapteff 29:36

You just reminded me, John Capo turned me on to there's some kind of like uh RFC connected, whatever. I don't know if it's an MCP or whatever, but you can invoke an RFC, let's say the OSPF RFC, yeah, as you're generating. Yeah, which like, oh, I didn't even know that was a thing, and that's really smart. Like, okay, that might be one of those constraints you're talking about. Exactly. You know, work within here, yeah.

Scott Robohn 30:28

So so it's interesting from um using

RFCs as Constraints

Scott Robohn 30:31

RFCs as constraints, but they don't guarantee what a config should look like.

Chris Grundemann 30:36

Not at all, right? They have nothing to do with configs. Yeah, right, because because the RFCs are actually written for Cisco, Juniper, Nokia, Arista, whoever. They're not written for users of these things. Now, now that's not to mean that as a good network engineer, you shouldn't understand the RFCs. I I found it very, very helpful to understand the RFCs. Yeah, but the RFCs are written for the person writing the OS, not for the person writing the configs. The configs have another layer of opscape of abstraction there where you're actually using iOS or S. My brain just exploded in a great way.

Andy Lapteff 31:03

Yeah, no, it's like wait, the RFCs are for configs.

Scott Robohn 31:06

Whoa. Well, let me let me give you a really, really specific example. If you take MPLS interprovider VPNs, there's option A, option B, and option C. All three are part of the RFC and RFCs, but they're mutually exclusive. And this this is what drives people to say things like the wonderful thing about standards is there's so many to choose from, right? Right. So I obviously want, like if I have a Juniper router and a Cisco router that need to do one of these options, I want them configuring in the same option. And that's all about how you write your constraints and how you prompt a model.

Andy Lapteff 31:40

So I want to get people's attention, and I'm always speaking to those people that aren't automating, aren't here, aren't filling out those surveys, and I fear will be disrupted at some point in the not too distant future because of the two-letter word that is getting really, really good at things. Is there any magical way to compel these people to engage?

Scott Robohn 32:01

There's no way to compel anybody to engage, but you can create the watering hole, keep the seats clean, make sure there's enough seats, make sure there's enough of a support environment to pay for the upkeep of the watering hole if you catch my drift.

Chris Grundemann 32:15

Right. I mean, I think you know, the CLI is like any other addiction. No one can quit until they decide it's time. Yeah.

Scott Robohn 32:22

Have you ever used that analogy before?

Chris Grundemann 32:24

I think that might have been the first time.

Scott Robohn 32:25

I feel like you just dropped a major that we should end the podcast right here. The 12 steps of CLI exactly. Admit you have a CLI problem. I'm Scott and I'm a J N C I E.

Chris Grundemann 32:35

I'm joking, but I mean this is a personal decision, right? I like I agree with Scott. There's there's no compelling people to come here, but I do think if they get exposure to these tools, this is not a threat to your job. This is the future of your job. And in one way, shape, or form, right? And we don't know exactly what that's gonna look like. Honestly, I don't know how disruptive AI is gonna be. I really don't. It's it's potentially scary in the in the same way that automation was finished. I I remember when people were scared of Python because they thought Python was gonna take their job. Well, that was 15 years ago. And guess what? I don't know anybody who lost their job because of Python, Python in the networking world. Now, now that's not to say there hasn't been anybody. You know, I don't want to, you know, cast too wide a brush there, but but right, like like automation didn't wipe out a bunch of network operations jobs. And I'm not sure that AI will. I do think that you can make your life a lot easier, whether or not the rest of your company's on board.

Andy Lapteff 33:26

You can make it a lot easier once you get over the hump of learning the new thing, right?

Chris Grundemann 33:30

Yeah, yeah.

Andy Lapteff 33:31

Which which used to be my why did you get into technology though? You don't want to learn things? I wanted to learn networking.

Chris Grundemann 33:36

Yeah, okay, fair. But it's still this is now how you do networking.

Andy Lapteff 33:39

Well, that's the problem. I wanted to, and I'm again, I'm the Lorex, I'm speaking for this, but like I wanted to stop climbing poles for $22 an hour and getting stung by bees and bit by dogs. Like I wanted a better life because I fell in love with a girl who's not my wife. I didn't know. Nobody told me.

Chris Grundemann 33:53

You

Compelling Non-Automators

Chris Grundemann 33:53

have to keep learning new things.

Andy Lapteff 33:54

And you would even you would said this, but Scott said this a bunch of times, which I love. Like, you know, if you don't want to do the thing, there's the door, like go. But but this is tech that like you have to stay skilled up, you have to do the thing.

Chris Grundemann 34:05

Because the vendors change the OS, right? So you even if even if you're a Cisco fanboy, which which OIOS, but like which version of Cisco OS did you do you know? How many have you learned, Andy? Honest, honest question. How many versions of Cisco OS have you learned up to the point before you learned Python?

Andy Lapteff 34:18

Three. I don't know.

Chris Grundemann 34:19

Right.

Scott Robohn 34:19

The other the other thing to throw in here too is like we can we can pick on network engineering, but this is much bigger than that. Like whether it's like if you zoom out for IT, right? You signed up for a career in IT, any any discipline. You you said you wanted a life of disruption. You may have you may not have thought of that explicitly.

Andy Lapteff 34:40

I didn't, yeah.

Scott Robohn 34:41

So that's which is what I push back.

Chris Grundemann 34:43

That's wild to me. That's wild to me.

Scott Robohn 34:44

Yeah, we rode the wave at certain points in our career, and other we waves keep coming.

Andy Lapteff 34:49

I was working on last mile like customer premise equipment for 50 grand a year, working all the overtime in the world. And I heard there was a job higher up the stack where those things connected that made twice that to start. And you get to sit down and stop climbing poles. I didn't have, and I hear other people say this in places like I didn't have representation in my life to know. I never I didn't know a network engineer, I never knew anybody in tech. Like I come from carpenters and cops, like so. Then I get in, I'm like, oh my god, look at all this. Like cloud, cloud came through, and I'm like, are you serious? Which one? All of them, come on, you know, and like it's just one thing after another. But I think that you're you're right with the identity, and I think you're right with like if you don't know what you're in for, and then you get in here, like I don't think the 70% of those folks are all me, but I just wanted to be a network engineer at a desk

Never Stop Learning

Andy Lapteff 35:41

not climbing poles, and then they threw the kitchen sink at me. Sure. And I dodged and weaved as much as I could because again, four maintenance windows, I got little kids, I'm studying for all this different stuff. Like, guys, I'm killing myself. Tough shit, that's the job. Okay, yeah, and then this stuff, I'm like, oh my god, right. And I have my story based on my identity. I think the identity thing I'm kind of locked into, but it's big on the individual level.

Chris Grundemann 36:05

Now here's the thing, right?

Andy Lapteff 36:06

You can't change anybody's mind if their identity is this isn't me, this isn't what I do. Exactly. So there is no getting them here, right? Like exactly.

Chris Grundemann 36:12

Well, it's it's very much like it's very much like politics, right? Like if you choose this identity of whatever it is, if you say you're a Republican or say you're a Democrat in the United States, there's a same thing.

Scott Robohn 36:20

It becomes very, very hard.

Chris Grundemann 36:22

You know, as soon as somebody talking to you says they're the other one, you stop listening because your identity is you know, you know what I mean? Because you're protecting yourself at this point. And so if you are a network engineer that defines your identity through the CLI and through that knowledge of that OS, you would literally cannot hear anyone else say anything else. Can you change identity? Yes.

Andy Lapteff 36:39

Is that even if you can't well I mean, could you influence other people's identity is really good?

Chris Grundemann 36:43

Yeah, and I and I think that's where it comes down to. I mean, it's it's was it was it Gandhi, right? Be the change you want to see in the world. I mean, that's that's how you do it. Right. You show people, you you expose people to this. Yeah. Hey, guess what? Like, I get to use all my networking knowledge, and I never have to learn an OS ever again. That's a really good way to put it. You should work in marketing. That's really good. All I have to know is Python. Yeah, right. Instead of knowing every OS of every device ever made. Which one's actually easier? That's compelling.

Scott Robohn 37:09

But you'll you'll still need to learn the nuances of the APIs on each of the vendor platforms, right? The LLM will figure that out for you. I want to I wanna I wanna I want to throw this back at you and like your oh, I I want to stop climbing poles because it's warmer inside, right? And it's not as hard on the body, right? Um, you were absolutely right in understanding you could move up the stack and you could make more money, but there was never an implied promise that you'd need to stop learning.

Andy Lapteff 37:34

Right, right. That was an assumption I made. And and I thought there was a finish point of the learning. Yeah, I'm here to tell you. I'll get my MP and I'll be a network engineer for 30 years and I'll be done. Right. Which isn't true, obviously. And I don't know where I got that idea, but that's that's true. Like we actually want automation.

Scott Robohn 37:50

Every other career we don't talk about it.

Chris Grundemann 37:52

Don't you arrive? Like, I don't know if that's true. No, especially not, especially not in 2026. Well, I don't know if that's true.

Andy Lapteff 37:58

But this was this was 2012.

Chris Grundemann 37:59

Well, it's like because changes.

Andy Lapteff 38:02

So there's CE credits in those higher, but I guess in tech it's the same, right? Like it's always changing, you gotta stay skilled up. But in my mind, I'm like, I will arrive. I will be done.

Chris Grundemann 38:13

Yeah. Right. I was just I don't know how many jobs where you can just like coast on your knowledge for a whole career. Anymore, anymore, anymore. Like no, there's probably some. There's I'm sure there's some.

Scott Robohn 38:25

And I wasn't limiting it to the scope of just IT. I was only using that as well.

Chris Grundemann 38:29

Well, no, but I think it's something we're talking about because we do talk about this, like, oh, well, if you signed up for IT, you signed up for learning. No, the problem is you were born after 1970. You're gonna have to learn some shit.

Scott Robohn 38:38

Yeah, but I don't know that that's that new either, right? I mean, I think I think the rate of change might be increasing, but uh I could talk to you guys forever.

Andy Lapteff 38:46

Yeah. Is there any parting thoughts? I mean, I I love what you've done. I was the anti-network automation guy, and I love the organization. I love the work you're doing. I just spent a week building a website that's a hundred times better than our old website, but the uh the repo on GitHub. Now it's open and people can uh submit issues and pull requests and help me iterate. Like me, the guy who six months ago was like, Hell no, I hate this. I'll never do it. Well, not six months ago, but I stood on your stage 18 months ago. Like I remember it wasn't that long ago. So you guys had a part in that, so thank you. Standing on your stage in front of 900 people, thank you. And is there any like I'm always talking to those people not automating? Is there anything we can say as a parting thought to to hook them?

Chris Grundemann 39:31

Well, I think it's just you know, show them these answers, right? Like, like you, like the the NAF YouTube is full of all the videos of all the talks that have ever happened on the stage, and there's probably one in there that answers a question that somebody's asking you, right? Point them that direction, right? With whether it's showing them that video or not, but like you know, give them the tools, show them the light, so to speak. Um, exposure, I think, is how this happens over time.

Scott Robohn 39:51

Yeah. No, that's definitely part of the equation. Do your homework, right? See what's there, see what's interesting. If you have questions, ask us, right? You can hit us on LinkedIn very easily. You know, we don't really scale, but we'll do our best to get back to anybody with questions. Come check out an Autocon, right? We're we're here in Munich, June 2026. We'll be back in the US November 2026. Details to be announced in about uh five days. I was gonna try to pull it from here.

Andy Lapteff 40:16

Yeah, no, this won't be released until after you could tell me a month.

Scott Robohn 40:20

It'll be watch, watch the NAF Slack, watch the watch the uh the LinkedIn page well announced, but come come check it out. Um you don't have to take our word for it. You know, there are they're gonna be 600, 700, 800 people that you can talk to in the halls that'll be presenting. So you'll connect with something. I mean, it might not be what you expect.

Andy Lapteff 40:40

If the public anti-network automation guy can change his identity and embrace this stuff and have repos in GitHub where people can interact with wooers, we're making an example. But if I can do it, anybody can do it because I couldn't do it until I did, and now I can. So you know, we got to bring them along. Um, thank you guys for everything. Um for all things Art of NetEng. You can check out our Linktree, Linktree forward slash Art of NetEnge. Check out the fancy new website at the same domain, artofnetworkengineering.com. There's an awesome resources section with a with a bunch of people, communities. I think I have NAF in there. If I don't, somebody should issue a pull request and tell me to put the NAFS thing in. Wink wink, help me make this better. But uh awesome website. There's a newsletter coming up. So as always, thank you so much for watching and listening.

Exposure Through NAF Content

Andy Lapteff 41:23

And we'll catch you next time on the Art of Network Engineering podcast. Hey folks, if you like what you heard today, please subscribe to our podcast and your favorite podcatcher. You can find us on socials at Art ofNetEnge, and you can visit Linktree forward slash Art of NetEng for links to all of our content, including the A1 merch store and our virtual community on Discord called It's All About the Journey. You can see our pretty faces on our YouTube channel named the Art of Network Engineering. That's YouTube.com forward slash Art of NetEnge. Thanks for listening.