108: Don't Let AI Replace You—Learn This First

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Everyone keeps saying the same thing.
AI is coming for your job. And I'm here to tell you that's
not necessarily true. But what is true is that the
people who understand how AI is actually implemented are about
to outpace everyone else here inIT.
And the wild part is that most engineers have never even heard

(00:21):
of the technology that's quietlymaking it all possible.
Today I'm joined by someone who's not only talking about the
future of automation, he's actually building the systems
that power it. William Collins has spent over
20 years in enterprise IT startups automation, and is now
helping shape how AI is actuallyintegrated into real networks.

(00:44):
This isn't theory. This is someone who's in the
code, in the conversations, and in the rooms where this future
is being created. By the end of this episode,
you're going to understand exactly what is coming, what
skills are needed next, and why MCP might be the single most
important acronym you learned this year.

(01:05):
Welcome to the show, Williams. Thank you so much for taking the
time to join us today. Yeah, glad to be here.
Love the channel of all the workyou're doing.
I really appreciate it. And, you know, kind of before we
kick things off, do you want to give us a little bit of
background about who you are, what you're doing, and why the
folks at home should should listen to this?

(01:25):
Awesome. Yeah.
So just to get started. So what I do today, so I'm, I'm
the director of tech evangelism at a company called itential.
And we're really a network automation and orchestration
company or just general infrastructure really.
We can automate anything and my job is essentially to help
bridge that gap between, I guesswhat our technology can do and

(01:47):
the real world problems that network engineers and IT leaders
are trying to solve. And I also host the Cloud Gambit
podcast, which recently got adopted into the packet pushers
family. But as far as like my
background, I started working intech really in high school back
before I knew, well, actually really back before I knew you

(02:08):
could make money and a career out of it.
I kind of had a something that happened early on in life that
had me sitting in a house with nothing else to do.
So I got into like building computers, and that's kind of
where I started. Yeah, I, I, I think I definitely
had a similar background. You know, I started building
computers from a young age and got a really kind of passion

(02:31):
from technology for technology there.
Didn't end up transposing until a career until later in life.
But that passion kind of really stuck with me as well though.
Yeah, same here. And it's, it's funny because
it's almost kind of like a builder, a tinker mindset, You
know, you just building things is fun, making things work is

(02:52):
fun. And in contrast, hey, networks,
we want things to work. Like there's nothing more
satisfying than, you know, building a Greenfield network or
fixing a problem in brownfield. When you have different overlays
and abstractions, you get into the nuts and bolts.
You have to go through troubleshooting, maybe throw
your red Cape on for a second. It feels real good.

(03:14):
Networking is super fun to work in.
It is, and I think that's what drew me to networking so much.
But you know, to kick things offhere, I want to hit on something
that you said during our, our pre podcast chat that kind of
really stuck with me. You know, you, you said that the
hype around AIA right now is so,so thick that you could cut it
with a butter knife and that that is definitely the truth I

(03:38):
feel like right now. But what do you see in the
market right now? What's, what is real?
What is all, you know, what is fluff?
I feel like AI, like you and I were at Cisco Live and AI was
just vomited over everything whether we wanted it there or
not. What do you think is going on in
the space right now? I'm just gonna go completely

(03:59):
transparent and speak from the the heart on this one.
So historically, if you look at so you have suppliers, you have
vendors, they need to sell software, they have to sell
things, you know, to, you know, in order to continue to exist
and be a healthy company. But then you have enterprises,

(04:19):
you know, down the mid market and smaller companies that are
buying technology and they have to be able to adopt it.
So one thing I'm seeing out there right now is it's so hard
for enterprises and just the thecompanies that are buying this
stuff to understand what actually something is.

(04:40):
What are we buying? What is the value proposition?
What does ROI look like? How do you put that in a budget?
Is billing going to be kind of like cloud computing is like
what does all this stuff look like?
And then you have companies up and coming that are just trying
to sell things. And a lot of times what you have
is, you know, historically goingway, way back, marketing teams a

(05:03):
lot of times will just go out there and, you know, say
anything to try to sell something.
Yeah, our product kind of does everything.
It just does all the things. And then that causes confusion.
Well, AI has kind of taken that problem and put it on steroids
because now so many teams and somany different sellers are using

(05:25):
AI to basically say, hey, take what I do and put it in a
marketing strategy to sell to like all these different
verticals. And then you just get all this
stuff. So I, I think there's just a lot
of confusion out there, which iscommon.
Of course, it's common this early in the hive cycle.
We're still so early with AI. And what does AI even mean?

(05:48):
What does it mean if you're working in infrastructure, if
you're a network engineer? Like, what does any of this
stuff actually mean? What is the value to you?
You know, how deep do you have to go down the rabbit hole for
learning? You know, what are you supposed
to learn? Where do you start?
There's just so much confusion. But it's starting to get
wrangled in because there's onlyso long that these wheels can

(06:10):
spin that the real use cases endup, you know, falling through
the cracks. And that's where, you know, I
think it's important to focus. No, I'm 100%.
You know, I kind of feel like this is a repeat of what
happened when cloud computing came out and you kind of touched
on that a little bit. I felt like when cloud computing
was just debuting, everything was moving to the crowd.

(06:32):
They were throwing everything that they could at the cloud and
seeing what was stuck and and then that kind of died off a
little bit and the good things stayed.
You know, I'm not saying cloud was just, you know, a hype and,
you know, a buzzword for a it was.
But you know, what I mean is that the things that were
actually useful stayed in the cloud and then some things kind

(06:53):
of started to fall back down to where they belong, in all
honesty. And I feel like that's the same
thing. What's going on with AI is
everyone's throwing everything at AI, seeing what sticks, and
the good things that are actually making our lives, that
are actually doing good are the things that's going to actually
be around and help move the industry forward.

(07:15):
I totally agree, you know, and it, it just, it always, if you
think about zooming out, it's always like so important to zoom
out because I was talking to someone, just a few, I'd say
like two or three months ago and, and they worked for a
pretty big company. It was like 25% through rolling

(07:35):
out SD Wan like they had just kind of gotten started.
And I was just thinking in my mind like, wow, we did that.
Like we were done, like at one company I worked at, we were
done like 10 years ago. And it.
It it's just, it's been around like so long, but you have this
diffusion of innovations like how it proliferates through the

(07:56):
market. You have like your your game
changers that are basically building these things like the
Googles of the world. Like they aren't even looking at
where the puck is going. They're, you know, constructing
the building, the power, they'redoing all the things.
They're engineering hardware, they're just building silicon.
They're doing all the things. But then you have the companies

(08:17):
that are looking at, OK, where is that puck going?
That's where we want to be all the way down to maybe some of
the laggards that maybe they don't have the expertise or
maybe they don't have the budgetto get things done, you know,
which is a challenge in and of itself.
So yeah, it takes a while for these technologies to
proliferate through the market, and AI is going to be no
different. No one 100% you know when funny

(08:40):
is, you know, when you first start talking about, you know,
that company implementing SD Wan, I'm like, man, that's a
that's a buzzword I haven't heard in a while.
It's just like baked into everything.
Now it's just, it is like, it's just like another part of our
day, But I remember when it alsocame out too, It's just like,
Oh, we got to do, we got to implement SD Wan.
I'm like, and you know, at the organization, I was like, I'm
like, why? It actually doesn't make sense

(09:02):
for us, but you know, we we had to implement it just because it
was the new thing. So IA 100% agree.
And I'm just out of curiosity, really quick for those
listening, you know, what advicedo you have for like people that
are kind of getting overwhelmed at this point by just how
prevalent AI is in everything right now?

(09:22):
You know, it's funny, funny you say that because my answer to
that question has changed so much in like the past month
even. So first of all, no matter what,
like saying AI is like saying like the medical field.
Like what does that even mean? Like you have, you have folks

(09:43):
that build robots that do surgery, you have folks that use
the robots, you have doctors actually doing surgery, they're
doing different performing operations on different parts of
the body, etcetera. And you have RNS, you know,
LPNS, you have so many people that are expertise in different
things that contribute to like ahigher goal, you know, to

(10:06):
execute on something. So one of the things I, I used
to recommend is always like, learn some, you know, at least
one or two things rather deeply.So if you're in network
engineering, that might be like TCPIP or BGP, like learn
something that's fundamental that isn't going away.
And as you try to navigate the confusion of everything else

(10:28):
that's stacked on top of that, one thing I found, particularly
when I was working on the enterprise side, that was really
useful is, OK, what's a problem that I'm solving or that I, I
have to work on day-to-day with my job today?
Is there some way that I can figure out, OK, this AIMCP
stuff, can it fit into what I'm doing?

(10:50):
And also there's tons of ways tolab this stuff like free tiers
that you can run things on in the cloud, cheap subscriptions,
just buying some tokens, like it's not super expensive to just
start labbing and figuring things out.
So I'm always I'm a big believerin like practice by doing

(11:13):
experiment, you know, that's oneof the best ways to learn.
And then lastly, I actually recently wrote a prompt I just
started using like clod code here in the past two months,
like just trying to figure it out and, and start using it to
optimize some of my workflow. But just going in and, and, and

(11:37):
just figuring it out and I, I built this prompt that that
basically said, Hey, I am AI, forget how I worded it.
It was like, I am a college graduate from computer science,
yadda, yadda, yadda. I built like a little profile.
I, my focus is network engineering.
This is the year. These are the big innovations.

(11:58):
What is actually real and applicable to my career that I
should be focusing on and that what it spit out?
Was. Actually really powerful, really
useful. So use those prompts like don't
just go in like Leroy Jenkins and say, hey, what do I need to
learn today 'cause that is not gonna get you anything valuable.

(12:18):
It's gonna just start assuming and hallucinating and guessing
on all sorts of stuff. You know, the, the power of good
prompting is huge. You know, I, I, I'm in a
crossroads here. My employer is not very AI
friendly. They're they're scared.
You know, the, the big boxes arescared of it.

(12:40):
Where me on the other hand, I'm so intrigued.
I'm in some, I'm in some large language model.
Every day I'm, I'm doing different things with it.
I just find it fascinating. So I have to do it at home.
You know, I have my home lab networks and things that I
tinker with, but it, I, I learned early on, you can't just
give it some, some Joe Schmo like basic prompt, like, Hey, do

(13:04):
this for me, because it's going to just completely go off the
rails and you're going to be fighting it to, to do what you
actually need to do. So the power of, you know, good
prompting is huge and it can be used in so many different ways.
And I love that example you justgave because, you know, a lot of
people are using it to help learn what they need to for this

(13:26):
career field to help try to stayahead of the curve.
And with AI, it can be difficultwhen those hallucinations
happen. And you might not know whether
or not it's hallucinating. It can really lead you down the
wrong path. And then you're sitting, you
know, in a job interview you're in, you're spouting the stuff
that you think you know, and they're like, are you from Mars?

(13:47):
Like what? What are you talking about?
That's not how that works. And, you know, I bring that
example up because I seen it, You know, I'm a hiring manager
and I had someone tell me something.
I'm like, no, like, where did you learn that?
And he's like, oh, well, I, I used AI and it told me that's
true. I'm like, go back.

(14:07):
And yeah, exactly, exactly. And I just, I had to, I had to
hold myself back from laughing because like, I felt bad for
this person because like, they had been, I don't know how
they've been using it. But yeah, that's, that's a
that's a huge thing to also consider.
You know with AI it's I still feel it's so much in its embassy
stages and who's late hallucinations can really run

(14:31):
wild sometimes. Yeah, and even using it when you
should. And I'll tell you even the other
day, so I was working on AI had this a spine leaf topology that
I'd built out in container lab and I forget what one of my
spines or all of my spines, I can't remember.
There was some problem. And I was like, so tempted to

(14:52):
just dump the configs and chat or in an LLM to, to figure out
what the problem was. And I'm like, I can't do this.
I can't do this for everything. So I sat there and I did network
troubleshooting until I found the issue, which took a little
longer than I was like. But if that keeps it kind of
keeps your edge in place a little bit because I don't want
to like lose. I don't want to make my, my job

(15:14):
in some of my things so easy to where I'm just using it for
everything. And I'm doing that to be
intentional. Not that I don't want to.
Start time back. Yeah, exactly.
No, absolutely. Because that's what's going to
keep you from being replaced by just someone who can prompt, you
know, your job all through, you know, away.
So, you know, having those skills in front of mind.

(15:34):
Exactly. And yeah, Speaking of context,
you mentioned MCP, which was theother part of my, I forget where
I was going with that answer, but like MCP is definitely a
model context protocol. It's been the game changer for
AI as far as being adoptable I think in enterprise.
So. Absolutely.
And that's was literally the next thing I was going to ask

(15:56):
you is, you know, MCP is kind ofthe hot new thing for
automation. And, and I I feel like a lot of
IT pros still don't understand it or even quite understand,
yeah, what it is. You know, they may have never
heard of an MCP server. What it does do you can you kind
of break it down what MCP? And like like I said in the

(16:17):
beginning, why it's probably oneof the most important things
when it comes to AI and networking that people should
really Start learning. Yeah, absolutely.
So MCP has become so important for the, the success of making,
you know, AI actually work at anenterprise scale to enterprise
level. And one thing I heard a lot of
when I first started looking at MCP was, hey, it's just a

(16:40):
wrapper around an API. It's not anything else.
It's just nothing beyond, you know, the existing, you know,
function calling mechanisms and things we build.
And I, I dug in, you know, because I kind of wanted to
believe that because believe it or not, Dakota, like whenever
anything new comes out that I have to learn, my first reaction
usually is like not another thing, another thing.

(17:03):
I. Just don't have time for all
these things. It's a little crazy.
So as I started digging into MCPand I guess we should start out
with a oh, like an explanation of like why what is MCP or why
does it exist? So it kind of just hit its one
year anniversary. So it's only a year old.

(17:24):
So it's got its one year onesie threw a birthday party for it.
It was great. Everybody loved it.
So MCP is basically it's an openprotocol.
It's open source. You can actually go to the
GitHub repo and look at it. And if you look at the GitHub
repo that I think the official definition is it's a open source

(17:47):
protocol that enables integration between LLM
applications and an external data sources and tools.
So basically a way to securely use an LLM against
infrastructure that that you haven't managed.

(18:07):
So if you, you can think about it like, so MMCP is kind of to
AI like what BGP is to routing. So it's a, it's a standardized
way for different systems to exchange information and
capabilities so they can work together.

(18:28):
And I guess going beyond that, like what is the problem it
solves? So imagine if every router
vendor invented their own routing protocol that only
worked with their equipment, yadda yadda yadda.
We all know that never ended well.
But that's essentially kind of where AI integrations were like

(18:50):
pre MCP. So every AI model had its own
way of connecting tools, databases, external systems,
etcetera. You know, you got all these
developers, you know, writing all this custom glue code
everywhere for integrations. Cause developers going to
develop, that's what they do. They build a lot of stuff.
And then what happened was MCP kind of got shimmed in there and

(19:15):
it creates a common control plane between AI models and the
tools and data they need to access.
So instead of like a, it's kind of like going from hub and spoke
to full mesh if that kind of makes sense and making
integrations a lot easier. No, absolutely.
It's funny, the first time I actually heard the turn MCP was
at the attentional booth at Cisco Live.

(19:37):
That's kind of where I got introduced to it.
I'm like, what do you mean? It it's the bridgeway almost
from AI to your network and it it didn't like, I still, to be
honest with you, don't fully understand it because I'm not in
the weeds enough with it. You know, it's not something
that I can necessarily implementyet into my networks, but it's I

(19:58):
can definitely see how it's really going to help shape the
future and bridge the gap there,you know, and I guess you know.
Just to kind of talk about what you guys at itential are doing
with MCP, it kind of intrigued me.
Do you mind kind of diving into that a little bit?
Yeah. So first of all, the way that

(20:19):
itential was approaching AI was one of the reasons I just
started working at itential thisyear, earlier this year.
And one of the things I loved somuch is like when LLMS sort of
came to market and everybody jumped on just every little
thing every, you know, all tons of vendors had these chat bots
that would allow you to chat with like not very much

(20:41):
documentation. You know, there was just a chat
bot for everything and all thesedifferent things.
And I loved how I, Tenchel didn't just jump on that wagon
immediately, but they were thinking about and I remember in
my interviews, it was one of thethings that was important to me
was, hey, we're not just going to go and try and roll anything
out to market. We want to wait until like a

(21:01):
real use case comes up and really build around that real
use case to provide real value to our customers.
So what that kind of looked likeis, and again, we've been on a
aggressive innovation trajectorythis year.
I've had a front row seat. It's been crazy exciting and

(21:24):
just awesome. But it, it kind of started out
in May 2025. Yeah, we're still in, we're not
2026 yet, but it's close. Yeah, starting to get the wires
crossed. But we, we launched the our MCP
server in Prague, Czech Republicat an event called Autocon 3.

(21:44):
So the core idea was deceptivelysimple but powerful.
And it was that, you know, enterprises are adopting,
they're trying to adopt this agent stuff, LOMS, they're
trying to use copilots and it's going so fast.
But in our minds, nobody had really solved the let's use

(22:05):
network analogies all day. So the last mile problem, let's
just say last mile problem, how do you let AI actually do things
in production infrastructure? That just makes me nervous.
Yeah, exactly. But you've got to do it without
losing governance, compliance, auditability.
You have to have those things. So how do you do that?

(22:26):
And the MCP server was our answer there.
And yeah, I mean that that's thecensus when you just started.
Like how do we let AI into the production?
I'm just like, my stomach started churning.
I was just like, you know, cuz I've seen, you know, AI
hallucinate when I've used it tohelp me troubleshoot the
problems. But yeah, MCP kind of puts those

(22:49):
guard rails in a sense into on AI.
And do you mind kind of diving alittle bit deeper on like how
how that happens 'cause I, you know, just for, I know I'm
curious. So I'm sure that the audience
listening is. Yeah.
So again, the MCP server was ouranswer to try and solve for
that. And so as we discussed, you know

(23:10):
Anthropic came to market with model context protocol as a
standard. So it acts as a control layer
between the AI systems and the potential platform.
So every, there's so many words for things now it's like out of
control. But for every, I would say like
AI generated action, whether it's like a configuration
change, a compliance check or something for remediation, like

(23:33):
a remediation workflow, it gets routed through a like a policy
enforced workflow. So the existing workflows,
validations and approval systemsthat you already had in place,
the governance you already had in place with potential, like
all those guard rails are actually still there and still

(23:54):
working. It's just everything.
On top of that, how you interactwith them, how you're able to
integrate them with other thingsis much more simple with MCP and
the exposing and filtering and dynamic calling of different
tools in tandem with other tools.
So this you can think of MCP is kind of like the, the catalyst

(24:18):
to get things really going to where using agents is a Gentek
AI thing can become like real life.
So we recently at Autocon 4, which was a few weeks ago
actually, we, we launched our Flow AI, which is a full
orchestration platform. And this takes everything we

(24:40):
learned from the MCP server and kind of extends it into what
we're calling AI to Action Continuum.
So going into, I guess, staying kind of more focused on MCP, but
like you kind of need that. You need safe integration, you
need things like Oauth under thehood, you need guard rails.

(25:01):
You need all of those things to really understand and define
what is contextual and like where you know where the
basically where the D mark is from reasoning to determinism.
So what I was talking about earlier with those, those
workflows that go through and, you know, test something or they

(25:25):
validate something or they do a specific thing, they include
guardrails. Those things are, are
deterministic workflows, but allthe things that require so much
reasoning that a human would have to do above that layer,
that often takes a ton of time. And now a good comparison to
that would be like finding an available IP address in a range

(25:48):
in some system that you use. Well, there's many different
checks to where you can determine if an IP address is
being used. You can look in that system and
see, oh, it's there, it's reserved.
It's not reserved. You can do checks to make sure
it's not reachable from different networks within your
organization. There's lots of stuff you can
do. So you can think about all that

(26:10):
advanced reasoning being the agentic stuff on top and, and
really the Sky's the limit with the the agent stuff.
There's so many different very cool use cases.
But yeah, we we don't have to gothere for the conversation.
It's, it's, it's fascinating that those things is what excite
me about the future of automation.
And to I think the untrained ear, that sounds like it's going

(26:33):
to replace people. But to me, it feels like it's
just more tools to make your workflow better and to actually
allow you to do the things we'vebeen dreaming about for years.
You know, the, the, the things that we've been begging for.
You can actually now start implementing some of those
things, but I want to backtrack a little bit before that.

(26:55):
You mentioned something that really fascinating.
You know, we have these start-ups right now that are
building the tech that these enterprise companies want, you
know, but enterprises often can't really adapt it without
the right foundation. And I kind of feel like MCP is
again bridging that gap for them.
You know, it's helping put thosesecurities in place so it can

(27:18):
actually be adopted by more organizations and, you know,
become more mainstream. Am I right there?
Am I misunderstanding it? No, that's true, definitely.
But it's like the whole analogy of, you know, how do you eat an
elephant? Like one bite at a time.
There's so many things that you have to do.

(27:39):
And one of the things that we'rebig on identical is it's it's a,
you know, these things aren't absolute must haves, but it's a
good idea to have some foundational things in place
with automation first. Like get some of these wheels
turning and think of like, OK, like if you're in that spot

(28:00):
where you're maybe a team of three people and you're the only
ones doing network automation for the entire company.
And if one of you left, you haveproblems and it's back to doing
a lot of manual things and a lotof things break, that's that's a
problem. And you're going to run into
that same problem. If you start to just throw AI on

(28:23):
top of things, it's actually maybe going to make some things
worse. So there are some fundamental
mechanics that are, I don't wantto say they're absolute must
haves, but they make a lot of sense, you know, before you
start embarking on this journey.And one of those things I would
say is, you know, taking a platform centric approach to how

(28:43):
you build and maintain infrastructure.
So that way you can really orchestrate across like
Federated systems, like typical big, you know, like enterprise
companies have. So it's not just Tay me,
William, I need to automate like30 switches or something and

(29:04):
that's it. I'm in my own little world.
It goes beyond that because you have ticketing systems, you have
to update, you have, you know, monitoring systems and things
you have to onboard, you have other teams that you have to
update. And you know, that's one of the
beauties of like where this MCP paradigm fits in because, you
know, one of our partners Selector AI, they do telemetry

(29:27):
and they take all this contextual data from network
devices and do just a lot of triage and, and let you know
sort of the the problem behind the issue.
So it's like you plug in their MCP and they're the visibility
and kind of the first face of the troubleshooting phase.
And then you lock our MCP in next to that and we do auto

(29:50):
remediation for said problem based on all this contextual
data that normally a human wouldhave to get.
So if you think about looking atthe, the routing table for BGP,
like let's say you had a BGP flapping issue, there's a lot of
different things that can make BGP flap, whether it's upstream
or something, an optic or some problem on the site.

(30:11):
Well, going through and figuringthose things out, like when I
worked in OPS before, we could just adjust BGP or like shut
down a neighbor to, you know, for stability sake, we would
have to go through and check a bunch of stuff and we'd have to
document it in a ticket before we could do it.
Well, imagine if that busy part of the human aspect of that work

(30:34):
was already taken care of. Like the ticket was already
documented, you knew what the issue was, and then you could
just run one of a few remediations either
automatically or with the human in the loop.
So that becomes a really big time saving is what I'm getting
at is utilizing these tools, you're saving yourself time.
And that's the biggest complaintI had as a network engineer for

(30:56):
as long as I was a network engineer is I don't have time
because I'm firefighting. But what happens if you could
just limit that firefighting by 50%?
That's a lot. That's a lot of time.
Yeah. I mean, especially when you have
organizations that are working with really small teams, you
know, for years I was the only guy in, you know, working for

(31:19):
this ISP, managing our network operations center.
And I found, found myself constantly putting out fires.
And I can never do things to advanced systems to I was, I was
always fixing problems. I was never able to prevent them
because as soon as I got one thing fixed, oh gosh, this
thing's gone fire now. And so I, I could definitely see

(31:44):
how having these systems because, you know, things aren't
going to necessarily change there where, you know, it's not
just like I'm going to snap my finger and then have 20
employees under me. That's just not how businesses
work anymore. But having the proper tools that
help me do my job better is, is huge.
And you know, that's just, I feel like 1 use case like for

(32:06):
MCP and you know, you're in the weeds more you, you see how MCP
is being used, you know, other than, you know, trying to help
make our lives easier, what are some other advantages that you
see businesses like how are theyimplementing it to help further
along their organization? Well, one thing that I've seen a

(32:27):
lot. So usually it's like when when
you adopt something like networkautomation, like I remember when
network automation started to gain steam.
And the first time, yeah, the first time I was actually
allowed officially to use network automation in in a big
network, it was all read only. Like we weren't making changes.

(32:49):
We were pulling specific types of data into a machine.
We'd built some Python based on like SOAP, XML craziness back in
the day before Russ stuff was popular.
And then we would transform thatdata that we pulled from these
different places in the network and then we would populate it in
a dashboard. And all of this took a lot of

(33:11):
lines of code and a lot of different things to get just
that one thing working and that dashboard up for our network
operations center that showed this, these top things that we,
we wanted to see at all times. And, and there were things at
the time that we would like, we were actually waiting on a
custom integration with one of our monitoring vendors that we

(33:33):
had. And you know, we just waited for
a while and we just decided to build it.
Now MCP is really useful in thatinstead of, if you think about
how a Restful API works, if I'm building something like that's
based on REST, you know, MCP being a protocol.
REST is not a protocol. REST is actually an architecture

(33:55):
style framework that uses a protocol.
It uses HTTP under the hood. But Restful APIs, they're
stateless by design. So every request is a
self-contained request. There's no server side session
state or anything. So when I'm building something
with MCP to do something useful like that, I'm saying, OK, I

(34:17):
have a Restful call to like maybe pull a list of devices
from inventory. And then I'm going to have
another call that's going to filter this other device type,
OS type maybe, or something along those lines.
And then from there I'm going topull the config from that
device. I have a separate call for that.
You know, there's, so it's like this Lego building of different

(34:38):
Restful APIs to get to your, youknow, outcome where MCP is more
like it's not built for humans to write software against like
REST. It's more for AI models to
interact with these external systems.
And it's stateful by design. So it maintains connection
throughout the duration of the session.
So you have things now to where it's easier to integrate and

(35:02):
pull data from System A and System B and System C and then
maybe to like send a report to your boss daily with your
company's brand guidelines stamped to the report so it
looks official. So with MCP, you know, this is
hypothetical. I could be in my chat host and I
could say, hey, I want to pull the software version or the

(35:27):
status of these certs or this orthat.
I want to pull that data from this subset of devices.
Once you do that, I want you to put these in a table,
categorized or stack ranked thisway.
And then I want you to generate this in APDF and I'm going to
provide you brand guidelines. And then from there I have the

(35:48):
other MCP for my mailbox or something else.
I want you to mail that to this person, which is my boss.
I want you to do this like everyother day or something so I can
type that in with human, you know, just text and get that
outcome at the end or the end ofthe tunnel.
So and doing that manually or building something via REST with

(36:10):
some of the custom things we need to see would be very
challenging for most network engineers so doing.
Huge time consuming too. Yeah, exactly.
Time. The thing that you can't get
back. Yeah.
No, and IIA 100% see the value in that because you know, a lot
of network engineers are alreadyspread thin, you know, they're

(36:31):
trying to do they're trying to do everything and just being
able to buy back that time to build those systems.
And then honestly, there's a chance that maybe even build it
better than you could have in the 1st place just because you
know you, you only can know so many things.
You know, we, there's, there's only so much knowledge a human
can take on. And so by being able to do

(36:56):
multiple things that maybe you don't quite have all the
knowledge on, you're making yourself a more qualified job
candidate as well. Because now all of a sudden
you've unlocked this other skillset without having to invest
years to learn like a whole nother coding language and all
this other thing just by being able to communicate in a human
way with these systems now. Yeah, yeah.

(37:20):
And it compounds too. So when you think about back to
your comments earlier about the job loss, that's one thing I
hear all the time and is someonewho recently started to really
take a stab at vibe coding here in the past two months or so,
three months, you cannot. I mean, you can build a very

(37:42):
small prototype. I mean, maybe you can get maybe
20% of the way there, but once your code base gets too big,
it's really hard. And once you have to start
rationalizing some very complicated design decisions,
the AI and the way that AI workswith tokens and quadratic

(38:03):
complexity of tokens and like losing context over long,
there's just a lot of reasons why it's really hard for someone
that doesn't have coding experience to build a
full-fledged application 'cause it's really hard.
Apps are hard to maintain. Things change and there's so
many moving pieces and it is a tool, full stop.

(38:24):
It's not. So I'll tell you what it it may
take your job if you're not looking or trying to leverage
any of this stuff to make yourself better.
But if you're using this stuff and you're getting in the weeds
and you're, you know, leveragingit to make yourself better, make
yourself more productive, producing more, more efficiency,
that's going to be winning goingforward.

(38:47):
So those are the folks that are really going to thrive, I think.
You know, I, I was actually doing a live stream last night
on my my channel and that question got brought up is like,
what does the future look like in AI?
And the people who are leveraging AI as the tool that
it is are going to thrive. This is this is just the natural
evolution of tech. This is just an, you know, this

(39:09):
is how the industry's evolved over years.
You know, new systems got implemented and people lost
their jobs because of that. But then that new system that
got implemented opened up like 5other style of jobs.
Like it's just the shift, the natural progression of the
industry. Something this industry is
always in innovating, it's always evolving.

(39:31):
And if you're in tech, you have to constantly be evolved with
it. You can't just stop learning.
You have to be lifelong learners.
You have to constantly be educating yourself on the latest
and greatest things. And not all these technologies
work out. You know, we've seen plenty of
these buzzwords that come out and just die and tank, but, and
you still have to learn them because what if it is the next

(39:54):
big thing and you have to be willing to evolve with it.
And if you evolve with it and you, you, you get excited about
these things and these new opportunities, you're going to
excel. You're going to be writing your
own check by the end of the day because you're going to be light
years ahead of the crowd. You know, another thing that
came up during that live stream is that, you know, job market

(40:16):
saturation, that there's so manypeople going after these jobs.
You can't get a job. And I have someone complaining
that, you know, they, they just find it bad that, you know,
someone with a college educationcan't even get an unpaid
internship. And I'm like, that's not the
problem. You're looking at this wrong.
Like, what are you doing to makeyourself stand out from the
crowd? How are you differentiating
yourself? You can't no longer say I got a

(40:38):
college degree. I should be good enough.
You know why these people shouldjust be tripping over hand to
foot to offer me these jobs. That's not how it works.
And maybe, you know, 20 years ago, then there's a chance that
might work that way. But I even no, I don't feel like
it was that way. I feel like you still had to
market yourself. You had to work on your personal
branding. You had to be that diamond in

(41:01):
the rough and convince the employer why you are the one.
And by leveraging these tools, you know, there's still plenty
of people, like I mentioned in the beginning, who've never even
heard of the term MCP. This is going to be what's going
to make you stand out to those big organizations and get those
large paychecks. I mean, I, I, I hate seeing

(41:23):
talking about the money part because if you're in tech for
money, you're, you're in the wrong industry.
But that's what people resonate with the most.
You know what, I, you know what I'm saying?
So. Yeah, you mean you're right
about the job saturation. Right now it's on believable and
every that's another thing is AAI generated resumes.
I ran into that at one of my last roles when that first

(41:44):
started and I was a hiring manager.
Every resume was AI generated. It seems like they were all
experts at all the things and you would get into like 5
minutes of the interview or 10 minutes and you'd realize that
like half of the stuff was like they didn't even know half the
stuff. Like it was just AI generated.

(42:05):
And I talked to someone actuallyand they said that what's
happening is there. There's actually tools out there
that will do this for you, I guess.
But you put in the job description and say this is my
resume template build and it will.
Yep, just build. It out for you and then you.
I've had, I've had those companies reach out to sponsor
the channel and I'm like, no, that's, that's, that's, that's

(42:28):
counter, you know, intuitive, that's counterproductive.
You know, I, I, I had this issuemy, my day job.
We hired someone, you know, they, they resume looked great.
So I brought him in for an interview.
He talked all the right stuff during the interview.
So I'm like, you're hired, get him into the job.
I'm like, OK, can you SSH into the switch?

(42:49):
And he's like, I don't, I don't know how to do that.
I'm like what, like how do you not know how to SSH into a, a
switch? Your resume says you're well
experienced in that. Like, yeah, we, we, we brought
him in on a Monday and handed him his final check on a Friday.
Like, you know, and that, that'son me because I didn't push it
enough, You know, I didn't try to probe into him enough.

(43:11):
I just was like, oh man, this guy's got to be the guy for the
job. But yeah, we like you mentioned
there, there's those tools out there.
But you know, if you're able to actually lab the skills and
demonstrate, you know, these skills, that's, that's the
biggest advice I give is, you know, building a home lab or,

(43:31):
you know, or, you know, using itat the cloud platforms, you
know, there's, there's free tiers for all of this and you
can absolutely 100% practice allthis stuff and then document it,
show how you are doing it. Again, it's it's, it's major.
Blog about it. Create videos to be the next
Dakota, the AI Dakota, right? You know, I am totally for that.

(43:55):
If anyone wants to go out and create their own YouTube channel
talking about their journey, what they're doing, giving
advice like you that that is you're going to stay like, you
know, you, you submit a resume and you go for an interview and
like, Oh yeah, you want to see what I'm doing here, go to this
YouTube channel and it just has your everything you're doing.
It's just going to be, I'm goingto hire that person in a

(44:15):
heartbeat. I'm like, that's amazing.
I can actually see you do the job.
You're hired like, yeah, you know, that's great.
And, you know, YouTube's not foreveryone.
You know, I, I understand that, but there's still different ways
you can, you can go about it. You know, it is really amazing.
You know what you can do? It's.
A living resume. Absolutely, yeah, you know, and

(44:37):
I'm glad we brought that up because, you know, there's,
there's multiple sides to this coin.
You know what you can do now with AI and it's actually living
and breathing in the networks that we're using is amazing.
But I again, it, it is, it's a bit scary to some people when
they don't think ahead on how it's making them a better

(44:58):
network engineer and how they'reactually using it.
So, you know, we, I think it's great that we talked about how
you should be demonstrating those skills when you're looking
for those jobs. And you know, for people who are
wanting to future proof their careers right now, what should
they Start learning? You know, what do they need to
start experiencing experimentingwith today?

(45:19):
So that's a really good question.
That's a a great question. So the first thing rule #1 of
getting experienced club is don't look at all the hype.
Don't start with all the hype. Just stop and and make sure that
you have a few foundational things that you are continually

(45:41):
studying on. If it's network engineering, I
did a lot of BGPTCPIPDNS. Those are important things to
know. Don't take shortcuts.
Learning those things is going to pay dividends over time, I
promise you. And the second thing is, you
know, again, avoid the hype, butstart doing these things at

(46:02):
home. Like you said, I have a, I've
had a few different AI labs overthe years, but right now I have
a few RTX fifty 90s in the basement with and Linux is the
other thing. Learn Linux.
Use Linux as much as you possibly can because that is
like everything runs on Linux atthis point.
It's crazy. So learn Linux, you know, learn

(46:25):
some Python. That's also very good if you can
and then just get to work, startdoing real things.
There's so many different tools out there that make this easy,
AKA container Lab. We have a free tool called
Torero that you can use for automation.
I've containerized all that. So it's really simple to deploy

(46:46):
in tandem with Container Lab. You can load things on it, run
Python against real networking things and you know, you can
insert an MCP server and ask questions and figure out, I
mean, hey B, you know, experiment, ask AI to build you
the OSPF network and manage it, build an agent to manage it.
Say to to the point to where youcan say I need to do

(47:09):
redistribution between this and this, put together the plan,
deploy the config and then buildyour own guardrails.
Like all that stuff you can do for practically free, cause MCP
is something that you can just, it doesn't have to be a public
service. You can write it or build a
private MCP that runs over a command line tool if you want
it. So it's just really interesting

(47:31):
all the different things, but start with the real use cases.
If you think about where the real problems are that you know
about and then start working on small solutions for those
problems. Those problems are probably
personified in in larger networks and larger, you know,
businesses of course, so. Absolutely.

(47:52):
And you know, I kind of want a nerd out for a second because
you know, you guys over there atI Tenshall have built your own
MCP server and we've talked about it a little bit, but your
MCP server is open source, whichkind of blew my mind.
You know, it got mentioned, you know, mentioned to me at Cisco
Live when I was at your booth that you guys are putting this
stuff out open source that anyone can use and start playing

(48:15):
with. And can you kind of give us a
walkthrough on how my intentionsMCP server is actually built?
Because you've had first hand, you've been the one working on
developing it and everything. Yeah.
So the thing about our MCP, that's really different.
So one of the patterns, and there's a little bit of

(48:36):
knowledge that comes along with this and there's a reason why
there's so many MCP servers out there today.
And the reason for that is a lotof MCP servers were created by
basically just taking like an entire open API schema file and
just feeding that into a tool that basically converts your

(48:59):
schema file into an MCP server that you then just put on
GitHub. So that's why like, I mean, it's
kind of like the early days of REST APIs, like you just had so
many APIs that just popped out of nowhere and there was like a
proliferation of them. But now you're seeing it's so
crazy with MCP because most folks have REST APIs and they're

(49:20):
just putting it through the machine and it's popping out the
other end. So there's a few problems with
that. I'd say the first problem is
there's when you, when you do itwith just passing in the schema,
there is a ton of glue code thatis required to actually make the

(49:40):
thing work. And by a ton of glue code, I
mean a ton of of glue code like you have to account for, you
know, the agent loop. You have to account for writing
like very, very customized logicfor, for filtering tools, for
categorization of things like how things are dynamically

(50:03):
selected and launched. And if you pass in like an API
schema, you could be passing in and generating like 100 + 500
tools. And then if you're calling them
all every time you have a, you know, there's so many words for
things. They call it context window
bloat where your context window's just blown up.

(50:25):
And this in addition, is going to cause super degraded agent
performance decision paralysis. You're going to get things out
the other end that just do not make sense.
So when you know, and I can't take credit for any of the MCP
design originally this is all Peter Spurgata, the architect

(50:46):
who brought actually brought Ansible for networking the
market. But he went in with a pragmatic
approach of saying, OK, you can't just build something that
exposes everything and cross your fingers.
You need logic behind how this is structured.
You need all the, you know, whenyou think of Oauth and all the
existing value that we provide today that our customers have

(51:09):
built, how do you build on top of that to make sure MCPS
layered on that in a secure way that honors all the guard rails,
all the compliance and all the logic that you've built in?
How do you dynamically launched with tags that the categories of
tools that you want to use and how do you also generate that or

(51:31):
you leverage that on a persona basis?
So like with our MCP, if you were to plug it in to like
clawed desktop or anything LLM or something or another, you
connect the MCP and it's controlling what you have access
to within our platform. And then you just, let's just
keep it simple. Say you have like a an operator

(51:53):
persona or an engineer persona or an architect persona or
something. Well the operator persona may be
able to do like read only things.
They may be able to reset certain adapters, visibility
stuff, health check stuff. So when they leverage the MCP,
it's the only thing they have access to.
They don't have access to push config or do anything that might

(52:15):
be dangerous. It like a a level 1 operations
level, but then when you get a little bit higher to that
network engineer, how do you leverage the power of this MCP
server in tandem with our platform as a network engineer,
when you need to do some more complicate, you know, maybe you
want to back up some ad hoc configs on some stuff that you

(52:36):
don't have in your existing process.
Or maybe you're getting into just taking context from a
partner like selector typing it through with itential and and
doing some auto remediation withcertain things.
So having that persona based approach is something we really
focused on and being able to control, filter and categorize

(53:00):
the tools that are actually exposed in tandem with like all
the industry best practices withO auth with the way we do role
based access control at itential.
So the way that this sort of plays through is like starting
at the least the the least thingthat would cause any sort of
issue. Again, doing the read only stuff

(53:21):
and then slowly giving the AIA little bit more, a little bit
more, a little bit more. And then when you get towards
the end of actually like pushingconfig changes like the other
day with our MCPI actually connected our like our gateway
product to our platform, our gateway manager platform.
And then from there, I just started creating services with

(53:44):
the MCP, like a bunch of Hello world services.
And then I stitched them together and then, you know, I
ran them and then I did concatenation with the different
ways that each thing said Hello world.
And I was just kind of experimenting and it was really
easy to do, you know, building this stuff, you know, through
MCP and through a chat interface, you know, because

(54:06):
it's like, however way that I want to think about doing it, I
can do it that way that fits my the world that I live in.
So yeah, a lot of advanced features, not just your
run-of-the-mill pass through thetool and out pops out an MCP
server design. You know that that is 100%.
I feel like the future. And, you know, Speaking of the

(54:28):
future, I kind of want to pick your brain about what we're
going to see over the next couple years.
But before we kind of go down that rabbit hole, if people want
to connect with you, if they want to check out I Tension, if
they want to connect with you personally, where can they find
you? Where can they learn more about
I Tension, you know? Yeah, absolutely.
So I tension.com and I would go to the GitHub slash I tension,

(54:50):
which I'm sure you can link in the show notes.
I would look at the MCP server again, it's open, check it out.
And then we announced our Flow AI product.
I can give you a link for that as well.
So that's just got announced this month.
And if you want to find me, you can find me at William Hyphen

(55:12):
Collins on LinkedIn. And then I have a link tree
which kind of links to everything else.
It's macro engineered. So like on YouTube, Instagram,
TikTok, all those places. That's my handle, so you can
find me. And then of course, the Cloud
Gamut, if you go to Packet Pushers, you can find the Cloud

(55:35):
Gamut podcast that I host with Yvonne, Yvonne Sharp.
Awesome. Yeah, we'll definitely make sure
and link to all those resources down in the show notes below.
Now, again, like I was saying, Ikind of want to talk.
Pick your brain. You know, you are definitely
born in the weeds than I am. What do you see the next?
You know, I normally I'd ask what the next five years look

(55:56):
like, but I think that's too farout.
I think too much is going to happen between now and then.
What are you seeing the future look like in the next year or
two of, you know, AI and networking?
You know what, what are some newthings that excite you so much?
I know that's a loaded question because we we can never possibly
know how fast it's evolving, but.
Yeah, I'm always like shrewd when I think about these things

(56:18):
and I usually underestimate, butI think with AI that's going to
be a little bit different. So I think MCP really making it
to Main Street now has really because AI, this AI agentic
stuff was kind of Catching Fire a little bit and gaining some
momentum. Pre, pre MCP, people were

(56:39):
talking about agents, it was kind of a thing.
But now MCP has actually rapidlyaccelerated the ability to do
that. So what I see in the short term,
and I've actually worked with customers and talked with
customers that are actually doing this today in production,
some of them are ISPs, but bringing in and integrating

(57:02):
something like MCP into their process with what is known as.
So you have like human completely in the loop.
So basically human running the loop, doing all the things
completely manual. And then you have something
that's human on the loop, which you're bringing AI in to do
things, but the human is really involved in the whole process,

(57:22):
watching each change, vetting stuff before it happens.
Like the AI has control to do some stuff, but not all the
things. And then you have this human out
of the loop, which is everybody's dream.
Like, hey, you know, we don't need humans anymore.
Get rid of your knock and you know, blah, blah, blah.
Which for some things way out inthe future, I think it's going

(57:43):
to be a thing not for. There's many parts the tech
stack that AI will never touch, especially with certain
verticals and certain technologies.
It's just reality, let's be real.
So I think short term, we're going to see more human on the
loop use cases where you're doing a lot of read only stuff,
you're testing it out in your lab in the data center and

(58:06):
companies are trying to figure out how to do it.
And then there's also going to be a proliferation of customers
or I mean, companies like Itential that are rolling out a
lot of good innovation and productizing a lot of what AI
can do for the enterprise into apackage that you can buy, use,

(58:27):
and, you know, have some sort ofinfluence on the road map.
And then you're going to have start-ups coming to market that
are doing brand new things that are, you know, productizing AI
and packaging it up to where an enterprise can buy it, use it,
get an SLA of, you know, all thethings that enterprises do.
So really short term, a lot of human in the loop use cases,

(58:48):
small things. And then going into the future
like 5, maybe three years and beyond, I bet you will probably
see more agentic human out of the loop use cases as well.
Yeah, and I feel like having that human in the loop makes it
easier for organizations to adopt, you know, there's a
little bit less fear of, you know, you know, Skynet at that

(59:11):
point. Just put it in layman's turn,
you know, because. You got to build trust.
Yeah, right. Absolutely.
You have to make sure you know you're doing the right things
because, yeah, it is scary when you kind of think about what AI
could do with things went off the guardrails if you didn't
build those guardrails properly for it.

(59:33):
So having that human loop definitely makes it a more
palatable thing for organizations to, to to adopt.
And I definitely see that becoming the future.
That's what that's what's comingdown the pipe now.
That's what we're seeing as of right now.
I feel like slowly but surely it's becoming more in
mainstream. But yeah, IIA, 100% agree with
your your predictions there. I, I really appreciate you.

(59:56):
You've dropped so much knowledge, so many so many
useful tidbits that I feel like some people are gonna have to
rewatch this episode again to get them all.
But I really appreciate you taking the time to join us today
and just offer so much great information.
Absolutely. Thank you.
I had a blast to get into chat and nerd out on these things.

(01:00:17):
Yeah. So thanks for having me.
Absolutely. And again, I'd absolutely love
to have you back in the future because I feel like there are so
many more things we could have dived deeper into that we just
can't even cover it in the amount of time we have.
Wide topics, yeah. Yeah, exactly.
Everyone, I really hope you enjoyed this episode.
I hope you took some some usefulknowledge away, some motivation

(01:00:38):
to go out there and start practicing and learning these
things because this is the future.
This is how you're going to advance your careers nowadays
and just go out there and Start learning, start tinkering with
things and start messing with things.
Everyone again, until next time,keep learning.

All Episodes

Episode Transcript

Popular Podcasts

Dateline NBC

Stuff You Should Know

The Bobby Bones Show

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}108: Don't Let AI Replace You—Learn This First

Episode Transcript

Popular Podcasts

.css-r6mb8g{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:1;overflow:hidden;}Dateline NBC

Stuff You Should Know

The Bobby Bones Show

All Episodes

108: Don't Let AI Replace You—Learn This First

Dateline NBC