AI is everywhere. While there are many questions about whether AI is a bubble, the ethics of AI, and the cultural impact of AI, in this episode of the Hedge Mike Bushong joins Eyvonne, Tom, and Russ to talk about how network engineers can use AI. Are some use cases better than others? Are there pitfalls network engineers should avoid? Listen in and find out. https://media.blubrry.com/hedge/media.blubrry.com/hedge/content.blubrry.com/hedge/hedge-319.mp3 download
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
- Join us as we gather around the hedge, - where we dig into technology, - business, and culture with the finest minds in - computer networking. - Well, hello, Tom. I am very happy to - see that you have stuff on your whiteboard - today. - Yeah. Doing some projects. Working in the real - world. This isn't, you know, software stuff going - on here.
- And Audrey, Audrey's doing well. I know that - Yvonne is sitting there thinking, - I need to go and fix this plant. - No. No. No. It's a really good plant. - I do have a - new peace lily behind me. - We we had a freeze earlier this year. - My different Bachia didn't survive, and so I - now have a start of your peace lily - back there that is yet to be named, - but, it's thriving. It's been there a few - months and thriving quite well. So
- I'm a decent plant parent. - And today, we are joined by Mike Bouchong, - who I believe is in are you at - home? Mike, are you in California? I am. - I am. I'm in San Diego, six blocks - from the water, - a few hours from the weekend. Do not - be jelly. - We will do our best, but no promises. - Yeah. Exactly. - Yeah. - Yeah. So alright. So we started this conversation - talking about
- AI as always because that's, like, the big - topic everybody's really crazy about. - And, - Yvonne, you were saying you were at KYNOG, - and what was your impressions - of KYNOG - and AI? - Well, I think - so, you know, I'm I'm in a world - as a leader, and I'm a leader of - infrastructure people at - most forward thinking AI folks in the world - are doing. And, you know, we have we
- have, - managers that are building agents to to do - forecast and that are you know, so it's - it's - we've got a lot of agentic development going - on and lots of new training going on. - And so - and I and I guess what I was - surprised to see - is the huge chasm that exists between that - world - and what a lot of folks - in - enterprise network engineering understand - AI to be, which
- is basically - a chatbot and a meme generator. And I - think there's this - the way I've said it is the chasm - is bigger than I've ever seen it in - my career - between what is emerging and what the the, - you know, frontier capabilities are - and what, like, the rank and file technologists - understand them to be. And so I'm I - I I've told a lot of people this. - I'm still trying to make sense of it. - Like, I don't have a a coherent - philosophy or or even thought about it yet, - but it's,
- the gap is significant. - Okay. - And - I don't know. My my own impression is - that's true. - And some of that is - because - people honestly just don't know what to do - with it yet. - They're just honestly they're like - and when they try it, by the way, - the times that I've tried it, - they don't spend enough time learning how to - use it, - and they get frustrated because it doesn't do - what they want it to do. - Like, I I've I've seen this in the
- open source world - where people will use AI - to build something, - and then it doesn't build quite the way - they like or it's total garbage in all - honesty. - And then they just say, well, this is - just like, whatever. Go away. I don't even - care about that. Okay. - You know? I'll I'll come back to it - when it's actually - real without realizing that there's a much more - nuanced approach - to these technologies - that might be useful. - So, Tom, any thoughts? I've I've noticed, it - they they're definitely the minority, but I've noticed
- a couple of infrastructure engineers that have really, - figured out that they want to - embrace it and start using, like, it's not - just, it's not just a chatbot, but but - that's definitely the the minority - of network engineers that I know. But but - there are some of them. - It feels in some ways like - like, - you know, just another tool - to a lot of engineers, I think. - And they see it as sort of a - smart tool, but my old tool works. So - I'm gonna keep doing it till someone forces - me. And all we hear in the news,
- it seems like, is mandates for AI. And - if your company is not doing the mandates, - they probably will soon. So I'll just wait. - You know? Seems to be a seems to - be a one of the sentiments. Interesting. Alright. - So, Mike, - finally, - Mike's been sitting there patiently. - He's he's itching. He's ready to go. - So many thoughts. - So if I look at, like, AI generally, - it's it's pitched depending on who pitches it. - Right? It's pitched as a panacea,
- you know, a transformation thing. It's, - you know, the future. It's the way the - the next wave of whatever. It can be - tool to productivity. It could be a tool - to, you know, massive job loss. - It's sort of, you know, depending on who - has the the microphone at the moment. - It's, you know, whatever you want it to - be. - When you look at, like, network engineering, - it's meant to be, like, maybe it's the - great equalizer. - Right? Maybe it's, - it's gonna, you know, remove the toil from - the job. - Right? And so there's expectations that you're, you
- know, wildly more efficient or maybe you get, - you know, kind of expert level insight that's - at your fingertips. And so I think when - people talk about, you know, what's gonna happen, - it's - you know, that your expectations are that it - has to be transformative - for you to have any experience with it. - And when your ex when your expectation is - that it's gonna be transformative and then your - first contact is that it's, you know, - a bit mid, - my 14 year old twin boys will be - happy that I'm working their vernacular into our - the, - then then what happens is you you draw,
- like, an initial conclusion. That conclusion might not - even be right, but I think what it - does is it sets you down a path. - I'll give you, like, really easy examples. - You know, a year and a half ago, - they or maybe two years ago, you know, - all of the dialogue was around, you know, - AI and hallucinations. Well, there was an entire - wave of people - who adopted that. You know, they've ignored - follow on, you know, iterations in the space. - And so, you know, what was a problem - then? It's still a problem in some cases, - by the way. But but that problem is - less pronounced today than it was, you know, - maybe two years ago. But because your expectations
- were set and then your first experience was - was thus, - and therefore, you kind of rewrite it off. - And so you see waves of people who - disengage. And and so then when those waves - eventually, you know, crash on the shore, yeah, - you've left some people behind. - Now I'll wrap this up in a in - a kind of a an interesting p or - I think it's interesting. I guess I shouldn't - speak for the audience's - behalf. But, - so, Lizanne Bainbridge in in the early eighties, - she actually was working in automation, not in - network automation, but in sort of the more
- generalized automation. - And she trained she called, a thing that - the transition paradox. And the idea behind automation - was that, - if automation is gonna make you faster, right, - the the the, - the the the paradox there is that you - actually go slower when you first use it - because it's new methods, it's new tools, you - have to build things up. And so you - have an expectation of speed, but you have - a a realized - experience - of of of it being slow. - It was supposed to, you know, kind of - save a bunch of money. People were supposed - to unlock, you know, lots of TCO or
- OPEX or, you know, whatever. - And then you find out that, you know, - in the first in the opening, - you know, days, weeks, months, quarters, - it actually cost you more. Because you gotta - do what you were doing before. There's no - savings, but you're doing the new thing next - to it. So you hire people in. Your - costs go up. And and then finally, you - have this expectation that if it's automated, it's - gonna be more correct. I'm gonna make fewer - mistakes. I'm gonna drive fault out of the - system. But what you see is because you're - introducing new tools and methods that you actually - take things that used to work with their - mops and sops or whatever, and then all - of a sudden, you know, through automation, it
- doesn't work as well. And so this transition - paradox is that all the things that you - expect, all the value you expect it to - deliver, it doesn't it doesn't actually deliver. - Now if you take a workforce and you - say that that's true across individuals, not just - companies, and they have these expectations, - and then when their first experience isn't great, - you know, then what you see is people - disengage because none of these things are true. - For me, the question is not so much - of how do you get people into AI. - It's how you correctly set expectations in both - the short and the long term, - and then allow people to to tolerate or
- even to lean into - the transition that they have to go through, - you know, learning it and applying it in - different ways and then, you know, sort of - mastering it over time. - And so then and then here's my thrilling - conclusion is now in the AI space, you - have this wave of people who've gone through - that mastery. And so they're, like, so far - ahead of everyone else who kind of had - the first contact. - And so to Yvonne's point, like, I there - there is a chasm, and I would argue - that it's it's widening. - And the the interesting thing is that the - longer this goes, at some point, you're gonna
- have an entire generation of people who are - gonna declare that that gap is too wide - for me to cross, - and so you're gonna strand them. And so - to me, that this is gonna be the - challenge of our of this generation. It's gonna - be how do you like, do you leave - people behind? - But we can't hire enough people to to - to make it work otherwise. And so, like, - like, that to me, that is the problem. - Interesting. - So how does this - relate to automation? Or because it seems to - me that networks still aren't automated,
- much less AI. - Like, people - a lot of networks only, what, 50 or - what what was the latest number? Does anybody - remember the automation? - If you take the high the the the - the outliers and the hyperscales, - most people in data center would put automation, - you know, somewhere in the mid twenties. - Yeah. And that's that's Gartner is the most - bullish, and Gartner would say that, you know, - two thirds of companies will be or 70% - of companies will be less than 50% automated - by, you know, 2027.
- That's see, that's insane. That's just absolutely that's - insane. - We should be much more automated. - And then the other thing that I often - wonder about is our approach to AI - and automation - has been much more you need to go - build this. And by the way, that's partly - industry's fault. - That that is partly the fault - of - vendors and - standards organizations - who all want to be the standard
- or differentiating - themselves - so that they can say, - you know, - buy me or use me or whatever the - case might be. Pick me kind of a - thing. - That's to some degree, that is true. And - so there is some - some sense in which - when you go into this stuff, - it feels very sloppy - to me. Even automation, - it's very - I've got to sit down and figure it - out. - There's no there's no easy way
- for me to do this - in in any way or shape, which is - why the public cloud service versions, - you know, Grok and - Copilot and all the others that are run - out there are so popular in this space, - and that's why people have that impression. It - takes - I I I have a Quincoder as everyone - who listens to this podcast knows. It's sitting - right over there. I have a Linux box - running Quincoder - on a llama, - and it took me two days just to - get
- Visual Studio code connected to the Quinn server - so that I could do use it as - a tool for code reviews. - Like, - that's people aren't gonna do that. That's just - not that's not gonna happen. - So I don't The analogy that I've that - I've used, and I think the the industry - is a little bit too focused - on - this moment instead of taking a longer term - approach. And one of the things that - that I see in my world anyway is
- that it is a great time to be - a cowboy. - Like, it is a great time to be - the person - who wants to just - ride in and play with all the new - things and try them out and throw stuff - against the wall and see what sticks. - And I think - the the industry in this moment recognizes that. - But what we're doing is we're also, in - some ways, devaluing the people who work as - city planners, - who go in and take a look at
- everything and go, okay. - Here's how we build a thing - that's durable, - that people can use, - that where they can get from point a - to point b. Now granted, when we look - at AI as an industry, like, we are - we are Lewis and Clark standing East Of - The Mississippi looking West. Right? We don't really - know exactly what's out there. We know it's - something. We're pretty sure it's got a ton - of value with it, but, like, there's a - lot of dangerous stuff between here and there. - And we need people with the personality type
- that are gonna go explore that. - But what I'm also seeing is, like, we're - taking people that are more, - maybe structured, - less risk averse, and and - they're not sure where their place is going - to be in this new world. I think - what I would argue is that we're absolutely - going to need them and need that skill - set and need that mindset. - We're just not there yet. And and I - think, - like, we're being too shortsighted as an industry - in how we're thinking about that. Because at
- some point, you gotta structure some of this - stuff - because, first of all, everybody's not a cowboy. - And if you put cowboys in shot, in - charge, lots of people get shot in the - street. And it's it's it's no way to - live. Right? - So anyway, I probably overdid that analogy, but - that's kinda how I've been thinking about it. - No. No. Well, it's interesting because the governance - piece of this is something that really should - worry us. Like, I'm sitting here looking at - this article on Shire on Security, and he's - talking about a German court ruling
- that Google is liable for its AI searching - search, - summaries. - And we don't think about okay. I use - an AI tool - to build, - to to filter my telemetry. - And I don't notice that something goes wrong - because the AI tool doesn't find it. - And all of a sudden, the network goes - down and customers - lose you know, my customers - lose whatever they lose. - And everybody wants to point back to the - AI and say, oh, the tool broke. The
- tool broke. Yeah. But see, that's not gonna - be acceptable in the business world. That kind - of governance is not going to be acceptable - in the business - world. There's going to have to be - accountability. - And this is where you were talking about, - Mike, about hallucination - hallucinizations and stuff. That's, I think, where people - worry - about that kind of thing. - I I don't know how to solve it - though, unfortunately. - So Do you think, I'm I'm curious, Mike. - Do you think we're how close are we
- to the point where someone says, okay. We - can't really be spending all this money, - and we're gonna have to scale way back. - Have we have we started to approach that - yet? Or, where where is that? - We're - we're there at at almost every part of - the of the industry. So if you look - at news - even over the last week, right, the the - major hyperscales, - so just to give you some some numbers, - CapEx - across, - Google, Microsoft, Meta, - Amazon, and Oracle
- was, like, 250,000,000,000 - in 2024, - and it'll be over 750,000,000,000 - this year. - Like and that's approaching, you know, a trillion - probably over the next two years. And so - the question is, is there enough money to - sustain that kind of investment? And that's in, - like, the ones that probably have the most - clear path to monetization, by the way. - When you start lumping in, like, an anthropic - or an OpenAI, - those numbers obviously go up. If you start - moving down even a little bit from there - into the NeoCloud space, you know, it goes - up even further. - And people are trying to figure out right
- now, how do you monetize? - If you look at different numbers, - depending on sort of which week and who - did the poll, you know, people will tell - you that that north of 80% of enterprise - AI projects are failing. Some will say that - put the failure rate north 90%. - And so people are this is actually Yvonne's - comment about the cowboys. People are are and - this is not just, obviously, network automation. - This is, you know, sort of the whole - AI arena. People are are cowboying - their way through, and not everything is succeeding. - And then you're seeing companies claw back
- the the the money. And and so I - think we're we're there now. There will be - expectation of ROI. - You know, I would say, - you know, some of it's happened already. Some - of it will happen, you know, over the - next, you know, six to twelve months. And - then I think - at some point, we're gonna, - it's uncomfortable to say this. Right? But at - some point, we're gonna have our pets.com moment - where we realize that not every AI thing - was, - had merit. And it was all, you know, - exploration. - It was our Lewis and Clark moment. I
- think it's a good that's a good analogy. - And And then what you're gonna find is - that, you know, you didn't - not every place, you know, West Of The - Mississippi ended up being, you know, equally inhabited. - Like, we picked spots that worked and spots - that didn't, and and that's what's gonna happen. - And there's gonna be winners and there's gonna - be losers. The challenge on the cowboy side, - like, everyone thinks about the cowboys who who - spot the winners. Right? Those people like, when - you're there first, right, you make bank. Like, - that that's amazing. Sort of the the underside - of that, by the way, is there's gonna - be people who become the faces of failure.
- And when you're the face of failure like - that, there's there will there will be a - cost to that. - And so and then everybody else, I mean, - we know that, you know, the unwashed masses, - what they do is they sit back and - they watch and they see, you know, who's - gonna, you know, essentially who's gonna win. They're - gonna pick sides later. - And then, you know, the question is what - what moves those people? But, like, I you - know, to your to your question, I mean, - I think you're gonna see expectations of ROI, - you know, - nominally this year, next year, - and that and things like memory pricing and
- whatever. Like, that stuff's just gonna make it - even more important because it's I mean, this - is it's not even like this is, like, - a a relatively cheap investment. Like, it's free - to go and part and and, explore. - You know, like, at least for the Lewis - and Clark angle, like, you can like, if - you wanted to go and forge a a - path, it was kind of that's up to - you. Like, you could you raise the - you get your stuff and you take the - risk. - You know, in this world, it's like, I - don't think that people are going alone anymore, - and and it's not just their risk to - bear. That means the consequences are gonna be - higher, which means the scrutiny is gonna be
- higher. So Yeah. You know, keep your eyes - open that way. Particularly in the infrastructure space - where you do something - and you actually can impact - a lot of people very quickly - regardless of whether or not that was your - intent. It's much harder to firewall off. - Like, the Lewis and Clark thing, that's great. - You know, I have this little team. My - team is at risk. They go out and - they do their thing. It's kind of like - they're firewalled off from culture. If they would - have failed, - well, you know, they failed. Right? Somebody tries
- again later, - and they learn from that failure. - No. When you - accidentally - take a network down - because you are Lewis and Clark, - like, - no. Things get real, real fast. - And that's that's - there's there's a lot of I can understand - the hesitance - in that space. - So The the the bad news stories, though, - just I mean, they're gonna travel fast and - far. Yeah. - Exactly. Happens, what that's gonna do is create - a chilling effect. So if I'm on the
- sidelines, maybe I'm skeptical. Maybe I'm just maybe - I just don't know enough. - And then I see a data point that - supports me staying on the sidelines. Like, what - it does is it just in it entrenches - my position. I'm gonna wait until there's, like, - success. And I don't it's not enough for - someone to have some success. It has to - be, you know, breakthrough success. Meaning, - it's, like, repeatable in some way. - Yeah. Yeah. And so how do you like, - how do these things become repeatable? Like, that - that part I mean, that's an interesting question. - Like, what what what constitutes or what are - the properties
- that would make something repeatable at an at - an industry or even, like, a a vertical - level within just peers? Like, to me, that - part, I actually don't know the answer to - that. I mean, I - but that's what people are gonna be looking - for. And - I'm telling you, if if things are failing - in a 80 or 90% clip, what you're - gonna see is more stories. - And and then the dangerous and and maybe - the the the, you know, the - uncomfortable truth is that we can't let even - if if like, 90% failure is probably normal. - Yeah. Yeah. And that can't stop you from - breaking forward. Because if you do, then then
- the whole industry stopped. And then and then - if if somebody else makes it and you - don't - Mhmm. Then you're you're at a disadvantage. Your - company's at a disadvantage. You know, your your - network is at a disadvantage. Your career is - at a disadvantage. And so, - like, you you we have to pay attention. - I I do think it's high stakes. - But but I don't think you can I - I think not participating - is is challenging? - So yeah. So I was thinking of an - example about automation - being the same sort of a thing. - When Facebook had that big outage
- that caused them to go down for several - days or hours or whatever it was. - Everyone knows about that outage, - and they know that it was automation or, - you know, related, that there was an automation - problem. - But instead of taking away the lesson of, - well, this is why you build the system - to have failure domains, - and you build the system so that one - failure can't leak into other domains - with firewalls in the system, stuff like that. - The the takeaway was, well, automation is dangerous.
- And I think that that kind of thing - is actually slowed down, - and the same thing is gonna be true - for AI. Right? Every time there's a failure. - And the flip side of that is you're - like, well, how does it become repeatable? - Part of the problem to me is that - if you're a company who's made it succeed, - you have zero interest in telling anyone about - it. - Right. - Yeah. I was just thinking about that. The - the breakthrough successes are pretty easy for everybody - to talk about, but, - I I think it's probably the small
- small efficiency gains that add up over time - that really is the value of the whole - thing anyway and that's just not a newsworthy - thing. Like - I, you know, in my work yesterday I - was I was having this moment where I - was I I was waiting for the for - the l m to do some stuff. I - I I could tell it was gonna take - it like, I don't know, a minute, two - minutes and I was like, - I can go up and get a drink - or something, but I can't really like go - fishing or whatever. - And so and I started thinking about what - what about context switches? Can I use this - time somehow? Just a little very microcosm, you
- know, of what we're talking about. But but - I did not have to be engaged the - same way I was before. - And I think I don't know. I I - I think that'll be there'll be small wins - that nobody really is incentivized to talk about - and, - the people who are waiting on the sidelines - just aren't ever gonna see that. - Even even the breakthroughs though, Tom. Right? If - you're in a company - that suddenly gains - competitive advantage in your entire market - because you have a breakthrough, - you are not going to talk about it
- because you don't want your competitors to realize - what you just did. - Yeah. High frequency trading space is a good - example. Yeah. They do stuff. They're printing money, - and they're like, we're happy to privately and - quietly, - you know, do our little thing in the - corner, and we won't we're not gonna show - up to your conference, and we're not gonna - talk on stage about whatever because - every time we do, we we chip away - at our advantage. - That's exactly right. - But to go back to Tom's, like, - incremental - example - of you know? So so maybe, you know,
- this week, you you - you gain a, like, 10% - incremental improvement. - Like, those incremental - improvements - compound over time. - And so if if you think of it - as something like a compounding interest, - then - if you continue to make those incremental improvements, - whereas there are people sitting back and waiting, - not making those incremental improvements, whether it's as - an individual or a team or an a - whole organization. - At some point,
- those lines are gonna diverge. - The problem is if you look at, like, - a compounding interest chart, - the line for the longest time looks pretty - flat, and then all of a sudden, you - get this hockey stick effect. - And I think right now, we are at - the stage - where the lines still are - look almost parallel, - but at some point, there's gonna be a - divergence. - I can't tell you exactly when that's gonna - be, - but I'm pretty confident that it's coming.
- And - a a a point I wanted to make - while, - while Mike was talking earlier is that sometimes - we talk about - is it a bubble, is it not a - bubble, all those things as if - because we think it's a bubble, we mean - it's not real. - And those two two concepts are mutual they're - independent of one another. Right? The Internet was - very real. The capabilities of the Internet were - very real. The change that the Internet was - bringing was very real, and we are all
- here today because of it, - there was still an economic bubble. - Both of those things can be true at - the same time. - And so just, - you know, regardless of what you believe about - the economics of it whether or not it's - a bubble, it doesn't mean the technology - isn't meaningful - and worth some investigation. So I I think - it's important - to kind of separate those two concepts a - bit. - Yeah. The the technology is gonna be durable. - There's no doubt. I think,
- you know, I - think people will they need to assess whether - they need to participate or not. For the - people watching the two parallel lines, though, I - mean, how do you encourage someone who's watching - that? How do you - I mean, like, in in general, how do - you how do you get them off the - the sidelines? How do you get them off - the fence? So they because it the the - when, - if you think about, like, the the the - effort you have to put in - is gonna be sort of related to how - how far those lines have diverged. And when - those lines are, like, on top of each - other, then it's, like, it's relatively lower efforts
- to participate. You do it over a longer - period of time. So it's, like, low effort - but over a long period of time. If - you would wait until it totally diverged, then - you have to learn a new discipline. You - know, you don't you saved all the time - along the way. So maybe, like, during that - time, know, there was some opportunity cost. You - did something else that was useful, valuable, whatever. - But now you're sitting and watching. Like, how - do you get people to - to engage? - So so before we talk about that, because - I actually have ideas, I wanna go back - to automation for just a second and ask,
- are we at the point where people should - leapfrog automation and just go in - and start looking at AI, - LLM based tools - as a part of or superseding? - Does it supersede automation? What's the relationship here? - Do y'all think? Like, just just in general, - are we, like - we we haven't even succeeded in automation - at this point. Are we - are we, like, at a leapfrog point? What - what are we gonna do with this? So
- I'm - sorry. Go ahead, Mike. Well, so I'm looking - in cloud. I'll just I'll keep it super - short. In cloud, there were people who YOLOed. - They moved to cloud. They abdicated responsibility. They - said, we're in the cloud. It's up to - AWS now. You know? Jesus, take the wheel. - And then that didn't always work out for - them. - In fact, it almost never did, but okay. - So there's a lot of repatriation - going on. Go ahead, Tom. - The thought I've had, and I know it's - I know it's not this simple, but, sort
- of, I guess, - theoretically, - now would be a great time, I think, - to stand up an MCP server in front - of your infrastructure, inform about how to get - into your infrastructure. And I know it sounds - a little crazy, but we're cowboys. Right? You - know what, Yvonne? So Mhmm. You know, stand - up this MCP MCP server in front of - your infrastructure, teach it how to talk to - your infrastructure, - and - use, you know, use whatever you're gonna use, - cloud, whatever, - to to talk to this MCP server - and and do things for a while - and and figure out how to make those
- patterns that you already use, - to how to turn them into an automation - system, - or or something along the like, if - we're trying to a thing that learns from - language can learn from patterns of usage. - And so, you know, if I was starting - fresh and had nothing, - I would probably do something like that. Like, - let's let's figure out what the work is, - have something right you know, plan some software - for me, and, - why not let it watch what I'm doing? - And I don't know. It it it seems - like there are capabilities for standing up these
- systems that, - don't involve - sit down for a year and plan out - your work and build a very elegant architecture. - Do you think that the average bystander can - get started on that? - I don't know. Know. I mean, you do - have to be able to write a little - bit of code. I think you have to - but I I don't know. I'm I'm optimistic - that a network engineer could could do something - like that. Because all of us at one - point had to sit down and say, you - know what? I need to stand up a - web server, and I need to not wait - for some server to do it. Like, we - had to we had to all do that. - And so, you know, why not?
- None of us came out of the womb, - not even Russ White, knowing BGP. - Right? Everything we know, we learned. - And and, you know, and I think it - may present a unique opportunity. Like, - one of the things that that I've also - seen as an example - is that I have actually seen people who - are non developers - be more successful - implementing - systems and code with AI than developers. And - that's because they go to the AI, - and they say, I don't know what I'm
- doing. Tell me how to build this. Right? - And and so instead of, - like, you say, okay. I want you to - build it, and I want you to build - it this way, and I want you like, - so go to your AI tool of choice, - Gemini, Claude, whatever, and say, hey. - If I wanted to send up an MCB - server and talk to my infrastructure, how would - I do that? - And then it will tell you. And then - it you you can set it up so - that you have to approve everything that it - does. - But in a lot of ways, it lowers
- the barrier of entry. Like, you don't have - to write the Python from scratch, for example. - I I still think it's really important to - know the vocabulary - of all those things. Like, you you ought - to know what git is and how it - works, and you gotta have some general programming - constructs and concepts in your mind. And you've - got, like, those kinds of things. - But, actually, - you could use the tool - to - and and - and leapfrog a bit, but, like, the those
- standard concepts of automation - don't go away, - but you now have a helper. - And so - if you use it properly, - it's a unique opportunity, - to kind of either pivot or learn or - grow or implement in ways that you couldn't - in the past. And - and I I see a lot of that. - I I'm still excited about Mike's question, but - I don't wanna jump too far ahead - because I have thoughts too. - Yeah. Well, I mean, we can go there
- now. I mean, I just wanted to went - throughout because we talked about automation versus AI - in the very beginning. I didn't wanna lose - that thread until we Yeah. Kinda came back - around to it. - So I think from my perspective, - it's much more my my perspective on it - is is to find ways - to intentionally use, as you said, Yvonne, - the AI as a helper. - Find very - specific - task - that I may not know precisely how to - do - or I may just not want to spend
- the time doing, what whatever that case is. - Right? - And I use the AI - to do that thing for me. - I never just I think a lot of - the problem with AI right now is that - people walk up to it and say, you - know, log on to Claude or Gemini or - whatever it is, and they say, - write me a piece of software that does - this. - And Al spits a bunch of code, - and they don't know how to read it. - They don't know how to evaluate it. They - don't know how to do anything with this. - Right?
- It's much better - to start and say, I need to write - a piece of code that does this. So - you sit down and you start writing - yourself. - And then when you get to an impasse, - you say, - I don't know how to do that. - I'm gonna ask. And you ask the little - AI assistant, and it spits back some code. - But it's a very specific - thing - that you, - in the context of your own development, - know how to evaluate what it's done.
- So one of the I remember when I - first started writing code, one of the very - common piece of advice floating out there was - you can't learn to write code academically. - It it it can't be a a stupid - thing. It has to be some real problem - no matter how small it is. And I - think the opportunities right now are still that, - but, you know, have it write your software - and then have it explain your software to - you in your your hobby project or whatever. - And that's something you couldn't do before with - software. You couldn't it was like, okay, go - get your Python book and and learn how - to do this. You can you can tell
- the - machine, do this this pet project for me - and then now explain. And then and as - you read through it, why does this what - does this do? How does this work? And - you can ask a billion questions without annoying - it. And you can you can, you know, - take thirty minutes a day and just write - software and have the rest of the week - explain to you the software that it wrote. - I think we can actually accelerate our education - in all sorts of disciplines - beyond anything we've had before because you could - have He said that I would always treat - the AI as a code partner - rather than as the creator
- myself. - Right. For the for - the real thing. Right? You do what I'm - suggesting is you if we're trying to help - people kinda see the value, like, have it - write the software for you for this small - project, explain it to you, and then you - throw that all away and do it yourself. - Or now you now you know how to - interact with the machine. - But it's funny how we always fall into - coding. Sorry. Go ahead, Yvonne. No. It's fine. - No. I will say, I don't think it's - always gonna be the case where - you and AI are writing code together. I - do think we'll get to a point where - AI is doing most of the writing. But
- regarding, - like, Mike's question of how do you how - do you bring people along as a leader, - you always have a distribution of people that - are somewhere, like, you got your you got - your cowboys and your city planners, and then - you have your, you know, your your day - workers, you know, on on your team. Right? - You you've got your people that that are - are out front. - You've got people that are, you know, they - they're not comfortable being out front, but they're - really good at structuring, organizing. And then you - got the folks that are like, you have
- to tell them what to do, and they - do what you tell them, and that's all - they do. Right? - But as a leader, like, one of the - things I've seen be successful - is - you you give your your most forward thinking - people room to run, - and then - you work with them to develop something - practical - that everybody else can actually use - and apply it to their daily work. - And then once they can use it and - apply it to their daily work, - then all of a sudden, you've sparked some
- creativity. Like, oh, I kinda know what this - is now. It makes more sense to me. - And then they'll start getting creative with that - thing and figuring out how to incorporate it - into their work. But you've gotta figure out - how to find that bridge of there's this - new scary thing out there that I don't - understand, and I don't I don't know what - I don't know what to do with it. - And then build some breadcrumbs for the folks - that maybe don't know how to make that - connection. - And I found that incredibly successful. And that's - not just an AI thing. That's any new - technology.
- Yeah. - But, but I think, like, giving folks a - practical thing to do that makes their daily - work easier, better, faster that they can see - the personal value in is crucial. - The the the initial returns, though, are gonna - tend to be - It's gonna feel like you're wasting a lot - of time. Yeah. That's - So what what you gotta do like, if - if I wanna learn guitar, guitar, then I - know that I'm gonna be terrible at playing - guitar for some number of, you know, days,
- weeks, months, whatever. - But because it's a thing that I want - to do, - I'm sort of comfortable with that learning curve. - If I don't wanna learn AI, but I'm - told I have to, I don't have the - same forgiveness. - And so I'm like, this sucks straight away. - I don't wanna do it. What you gotta - do is figure out how do you lower - like, you gotta lower their upfront expectations. You - gotta raise their long longer term expectations. - But if you're not willing if people are - not willing to, like, faithfully engage and actually - try to do something and endure a little - bit of the early part that's
- a bit cumbersome, - then they're not gonna learn enough to actually, - like, turn into a valuable two tool. Now - the thing where I think AI is a - little bit different than guitar - is that AI is willing to help you - learn it. - So to, you know, to to Tom's point - about, you know, you can ask for help. - Like, it's willing to to to help you. - I think maybe that that helps a bit, - but I think I think you have to - find people who are - maybe in the right mindset. - And then, Russ, you asked a question earlier
- that I wanna kind of tie all this - back together to that. You you you asked, - can you are people gonna learn automation, - or are they just gonna learn AI and - leapfrog it? I think if you don't know - workflow, if you're not fluent in the language - of automation, - then, - you know, you can fail with any tool. - Yeah. And so I I I I think - what people they're what they're gonna have to - do is they're gonna have to like, - I think if you just if you think - that you can ignore operations generally, ignore workflow - generally, and then a tool is gonna come - along and then and then solve it for - you. I I I think that's probably,
- optimistic. - And and and where that will be true, - it won't be the tool that solves it - for you. It'll be that you have some - coworkers - that are doing hard work around you, and - then they're gonna spoon feed it to you. - Like, that that may be some people's experiences. - But, generally, if it's a company that doesn't - if it's a network that where no one - who's doing it knows - anything about operations and, like, how to architect - stuff and, like, how to think through, like, - the hard work of network engineering. If you - don't know that, - AI is not gonna suddenly give you that - discipline. I think it's gonna make it it's
- gonna give you a really, really fast path - to break things at scale. - Well and I think - to to just double down on that, Systems - thinking becomes more important, not less. - Yeah. Because as you start to build with - these tools, - you have to build them - in in a systems approach because, ultimately, what - you're doing is building a a system that's - going to - do all the things. Right? - They they've shown this in hiring and stuff. - So there was a belief that what it - was gonna do was eradicate the expensive, you
- know, your engineer with twenty five, thirty years - experience. - Turns out, those are the people who are - best equipped to use the tools. That's right. - Yeah. - Yeah. But now but there but there are - two - things there that I think are important. The - first is, like, when you talk about, like, - the the illustration of guitars is interesting to - me because - I know that when I was learning jazz - piano, the one thing I could never do, - you could not you cannot give somebody Goldberg's - variations - to teach them piano. - There's no success.
- Right? - It's it's scales. - Like, okay, I can do scales all day - long. I know the technique. Whatever, whatever. - Unless I see incremental - success - over time - at doing something, my interest level is going - to wane - very quickly. - So people need to be very intentional about - finding those things. Right? Like, to even get - away from coding - to get towards more towards networking, - don't just throw logs at the AI and
- say, tell me what's wrong with my network. - Throw logs at the AI and say, alright. - I've noticed this application might be running slow. - You know? I've noticed - a problem in this area. I'm gonna ask - very specific questions - based on my understanding of networks - that's going to tease information - out of the AI, which then gives me - a success. - It's a narrow success, but it's a success. - Matt Oswald? - Yes. Mhmm. So Matt Oswald, when he was
- doing the NRE Labs thing, which was, like, - such an awesome project, like, that was that - was he used to say that the one - of the hard things about automation, I'll just - you know, I would expand it. I would - say it's true in AI as well, is - Is that to some extent, you have to - be an expert before you're allowed to be - a beginner. - Yep. - And, like, I whenever he he he said - that, it was such a simple thing to - say. And and but, like, it stuck with - me. Like, there's so many disciplines where it's - like we and then you're going back to - the beginning of this whole conversation. Like, when - when Yvonne talks about that gap at at
- at k y nug, - do you call it kynug? Like, is that - a is that a - like, I don't know how to - right. - At at that place, - the - that gap, like like, the further ahead the - leading people get, the more those two lines - diverge, the harder it is to get started. - Like, I I almost feel like if if - people don't don't figure that part out now, - then, like, you you're gonna have to be - an Olympic athlete at some point. - Yeah. I don't I don't Well, I I
- mean, I think eventually, - like, an ecosystem - will develop that allows people to - catch up to a degree. - But the people that are gonna have the - outside success - are the are the the the people doing - it Well, it'll be now. So a vendor - will take this and they will pack it. - And then the cost of that packaging - will quite literally be the cost of that - packaging. Like, you will literally have to pay - more for it. Yeah. You will lose a - bunch of flexibility. Your operations will be
- whatever they essentially enable them to be. And - I think when people do that, you're abdicating - control and you're saying that my fate is - now like, I will accept whatever you hand - me as my fate. - And I think that that actually works. People - can as as long as as long as - people move in the direction you want and - at a pace that you're comfortable with, I - think that actually works for some people. And - we've seen that with, you know, insert vendor - name here and there. Yeah. I think when - it doesn't serve you and, you know, like - you know, so so Tom's day job, he's - got things that he does because it didn't - serve.
- Right? Like, I think it - I I think it's not right for everybody, - and people are gonna have to opt in. - Like, am I gonna am I gonna float - along the lazy river with, you know, with - with my vendor, you know, - floaty? - Or am I gonna, you know, swim - and go do it and and and go - where I wanna go? - And the the part that's cool is that - not everyone has to have the same answer - to that question. Like, there's no judgment. Like, - sitting in a floaty is awesome. - And and some of that's just a business - decision - of - how different - how differentiating - is this capability for me, and does it
- enable me to generate outsized revenue - or not? - And the industry is just not mature enough - yet to answer that question. It's just we're - just not there yet. I think we'll get - there. But Yeah. - Yeah. - So what scares me, the other thing I - was gonna say there, Yvonne, was when people - say things when you say, like, okay. - We're gonna get to the point where the - AI does the coding and we just use - the product, - is that then we say we need systems - level engineering, we need systems level thinking.
- But the only way I know to build - systems level thinking - is to build systems. - I don't know any other way. - I I I I literally I've been in - the training world - way before I came and when I was - in the air force, I was teaching people - radar systems and stuff like this. - And literally, the only way I know - how to teach people systems level thinking - is to get them to go through the - exercise - of building a system.
- And when we throw off and when we - throw off the system building to the AI, - I don't know where the next generation of - people - who know how to build systems comes from. - That's that's I think that's a yet unsolved - problem because we know, like, we need lots - of, like in the coding example, like, - senior developers are the most, - successful - with these tools right now. We are actually - we don't need as many junior developers, but - without junior developers, how do we get senior
- developers? Like, that, I think, is a yet - unsolved problem. I Well, I I I don't - know the answer. Nobody, - like, nobody builds a system by themselves anyway. - Doesn't matter if you're the smartest guy at - at Hyperscaler x. Somebody had to even if - you came up with the whole company on - a whiteboard, - somebody had to go actually do it and - it wasn't you. Like Yeah. And and it - can be this I know there's a lot - of there are people that think that I'm - the smartest person in the company. I'll draw - the white board. Everybody else will execute. It - doesn't actually work like that. And so Never. - Like if you if the idea of building
- a system is that you're the only one - doing it, that's false. And so someone else - is gonna be the built doing doing the - building - and Right. Like so so I guess what - I'm saying is - why couldn't you learn by analysis because you're - not building it all yourself anyway. Even you - might think you are but you're actually not. - You're learning from other people's experience. You're seeing - other people implement your things like, - white people learn by analysis. But there's a - difference between - like, I'll tell you that when you sit - in a room with 20 people
- and one person sitting in the back - and not actually participating - and doing any of the design work, - taking on any of the components - and figuring out how those components interact, - they're not learning. - And they're not gonna be successful in the - long term. And the danger with AI that - I see, one of the dangers I see - is that people do that. - Build me an app that does this. Sure. - Okay. - And then they're like, okay. Yeah. Yeah. Oh,
- I'm gonna go analyze the code later. - Trust me. You won't. - Yep. It's gonna be Trust me. You won't. - Was with that. Like, once you have to - scale it, like, the - even if AI is fast, like, reinventing the - app every time you need to scale it - a little bit - Yeah. Probably not the the most optimal way - to get there. So I I think you - have to learn from it. I thought I - thought Tom's comments earlier about, you know, start - with AI and then let it, you know, - learn from AI and then - and then sort of iterate and move on. - To me, that's consistent with how you learn - anyway. Yep.
- Yeah. Yeah. But I I just don't I - I get really, - like, we can't be passive - in this. - Right. Right? That's not our goal as humans. - I think what we're learning, actually, - is that folks that are being successful with - AI, their cognitive load isn't decreasing. - It's increasing. - And this is a conversation for another show, - but it's actually we're seeing increased burnout. And - everybody who's effectively using AI
- and is deep into it says they're not - working less, they're working more. - And so there's a whole set of, like, - socio technical problems - that we're gonna have to figure out how - to solve because - our our brains need margin - and you can't be doing high cognitive load - tasks - the entire day. You know, when I get - when I get kind of in this - brain fog space where I've overdone it, I - go and, like, approve expense reports - or do some kind of
- mundane routine task. You need new hobbies. I - know. I mean, this is in my this - is this is in my day job. I - go for a walk. I water my plants. - You know, something like that. - And we've still got to have that kind - of margin - as humans to be able to function. And - I we that's something else we've not figured - out yet. So That's actually another show, and - we should talk about that because that's so - I I'll we'll do the other show, but - I just wanna throw at something that that - made me think of Yvonne. The the first
- time I ever took off an - airplane, with myself at the controls. And I - had my friend with me who is not - an instructor, and and he was kind enough - to let me learn some experience. - I just remember it was and I don't - know if y'all have ever fun, but, like, - there is so many things going on at - once. And the first time you do it, - it is it is crazy. It is to - experience pilots say, yeah, whatever. But when you're - doing it for the first time and he - made the comment, after we got up in - the air and everything was fine. He's like, - oh, you're and he's like he's like shouting - at me and stuff. Like, do this. Don't - do that. I'm like, what the and and
- he made the comment. He's like, yeah. Your - CPU is like a 110, 120%. - Right? And and I was like, oh, yeah. - But but it was still - that 110, 120% - phase did end pretty quickly into the exercise. - And then it dropped down to a more - manageable 90 or 95. And so, like, that - I I think the - organizing yourself and and doing your work in - a way that that acknowledges that is gonna - be pretty critical. But if you if you're - able to do it, you'll be able to - go for the long haul and see the
- curves start to go back up. You know? - Yeah. I'm not entirely certain AI is built - that way. But, again, that's like we should - we should actually do a whole school. That's - an interesting discussion. We should get Mike back - on - to talk about that. Managing cognitive load when - it comes to using AI systems. Love it. - Yeah. Yeah. - Thanks. Mike. - My CPU is at a 110%. - That one feels hard. - Well, that's because we're at the end of - this one. They just need to come back
- fresh after you've taken a walk or proof - proof some expense for something. - Scheduler thrashing in process scheduler. - There you go. That's that's what it is. - Alright. Cool. Well, I think we beat this - horse till it's time to dismount it. - We'll get Mike back on to talk about - the other topic - and find other cool things to talk about. - So okay. Tom, say your word. - The word is LinkedIn. That's where you can - find me. That's where
- everybody can find Tom is LinkedIn. - Yvonne. - Yeah. - On LinkedIn, - I do more trolling of Twitter x these - days than I do posting, but I am - around. So, feel free to reach out. - And, you where is your other podcast? What's - going on? Oh, oh, the Cloud Gambit. Yeah. - Over on Packet Pushers. I - silly. I didn't mention it. Yeah. There we - talk about cloudy stuff and AI stuff and, - you know, more of this these kind of - things, William Collins and I. We always have - on an interesting guest,
- so check us out there. - Awesome. And, Mike, I don't think you have - any blogs or anything. Right? You're still kind - of - I backed off the socials. Like, - link LinkedIn is where I'm at. Just - if you're gonna reach out to me, don't - use AI. And for god's sake, please don't - it cannot be a sentence, a carriage return, - a sentence, a carriage return. - There's, like, a bunch of hashtags at the - end. Like, just be a be a person, - and I'm happy to. - And and I will just add the the
- high - by itself with nothing else is also not - recommended. - I used to have this friend who it - people would walk up to him and say, - yo. And he'd say, yo. Yo. - Like, what? What am I supposed to do? - You have intelligence. There's some intelligence right there. - Like, what? What do I do with that? - Alright. Well, I'm Russ White. You can find - me here at the hedge on rule11.tech. - Occasionally on x, I log in to LinkedIn
- every now and again - and - packet pushers and here and there. And so - feel free to reach out to me if - you have questions, thoughts, or whatever. - And we know that we live in an - attention driven economy. - And so, therefore, we thank you for listening - all the way to this the bitter end - of this episode of The Hedge, and we - will catch you next
One of the more interesting ways to use AI tools is to use local hardware and open weight models, or even cloud-based closed-weight models, to build skills you can then use a tools to help you accomplish specific tasks....
Brian Trammell joins Alvaro Retana and Russ White to discuss his IETF draft Optional Security Is Not An Option, and why optional security is very difficult to deploy in practice. Brian blogs at http://trammell.ch and...
Social media services have largely become a marketplace designed to capture and hold your attention, rather than places where people share information and talk to one another. A large part of this is because they are...
Who should be held liable when AI fails? Who is liable when an AI system "breaks into" a network or system, causing a major breach? Listen in as Tom and Russ discuss two articles covering these topics on this roundtable...
Network engineers do not need to become full-time coders to succeed--but some coding skills are really useful. In this episode of the Hedge, David Barrosso (you can find David's github repositories here), Phill Simonds,...