0:00

these end users are demanding APIs and when we start hooking AI up to it, I don’t know if you saw some of these

0:05

charts from live, but basically if you look at the API patterns to these systems, they’re very spiky.

0:11

Um, you know, when somebody’s doing something, it’s very spiky. When you hook it up to an MCP, it’s pegged almost

0:16

all the time because it’s going to use every it’s going to use every second of available compute to do something,

0:22

gather information.

0:33

Chris Wade, welcome. Welcome to the podcast. Thanks for having me. Awesome. Actually really been looking

0:39

forward uh to chatting with you. It’s been It’s probably been a couple of years since we’ve talked. I don’t know,

0:45

Chris. Was it like maybe before CO or like during CO, you know? It’s like It seems may seems like it’s been a while.

0:53

CO warped my time. So, um but probably right right around that time. Right. Right before right after, for sure.

0:59

Yeah, that’s crazy. Well, so um so how you been? Good, good, good. Um very busy. I think

1:07

the market we’ll talk about today is going through a lot of changes right now. Um, you know, onoo with with uh

1:14

attaching to AI, there’s just a lot of stuff going on. So, a lot of people are trying to figure out the the today’s

1:20

from the tomorrows. I think I think so. Yeah. I’m Yeah, I’m really looking forward to our chat. But I think

1:25

before we kind of just jump into things, you know, one of the main purposes of these podcasts were to introduce um

1:33

people to other members of the community. And so um so let’s give everyone a little bit of your

1:39

background. So like what’s your career journey been you know uh like and what kind of brought you to like where you

1:44

are today? Cool. So um I guess education wise I

1:50

have a computer science degree. So I think that’s kind of well informed kind of a lot of what I did and I focused in

1:55

networking and databases. So um I feel like my career journey is maybe a little bit straighter than uh than I’ve heard

2:02

but my first job was with a company called Micromuse that built service assurance software. Yeah. So I don’t know if you remember Mike but

2:09

fault management um discovery um a lot of assurance type tooling

2:14

and then that that business got sold to IBM and became part of the Tivly suite and I went to a I went to a solution

2:21

provider um in the assurance space um and then that company got bought by Alcatel Lucent. So I spent four or five

2:27

years doing that running the OSS practice with a lot of third parties. Um so got a lot of understanding of kind of

2:34

product market fit and what people are looking for from large scale deployments

2:39

and then about 2014 we saw you know networking really

2:44

changing so I was at onf I was participating in ODL

2:50

was kicking around the openflow 14spec for people that remember those days and I remember them very well. Yeah,

2:57

exactly. So, we started it to really focus on the programmability aspect of

3:02

of that. So, in the early days, we’ve used a lot of vernaculars, you know, controllers,

3:08

orchestrators, you know, netcom protocols. Um, a lot of people had different, you know, uh, ideas of where

3:14

it was going to go with white box and other things. Our our complete focus was on the ability to have programmability

3:20

of networking infrastructure, um, just like we saw with compute and storage. Um so really really really I

3:28

guess career-wise my first half of my career was really focused on the assurance side. Um I would say at some point you know

3:34

you feel like you’re uh chasing alarms and telemetry issues and uh and wanted

3:39

to be on the um you know kind of the uh the other side the provisioning activation on purpose types of changes

3:46

with the infrastructure. Yeah. Well that’s a great um that’s a that’s a great journey. Thanks for

3:52

sharing. So, and also you’re in Georgia, you’re right. I am. Yeah. Most of our ecosystem in

3:58

this business is Boston, Austin, and the valley, right? So, yeah. Um, interesting perspective maybe being

4:04

in in an alternate city. But, um, you know, if I look out my window, we’re right across the street from Georgia Tech.

4:10

I know. I was going to say like I don’t know like you got Georgia Tech like right there, you know? It’s like Yeah. Yeah. Exactly. You know, it’s a it’s that’s like a

4:17

black hole that sucks in a lot of really good talent, you know. Yeah. For a long time, um, tech was

4:24

exporting a lot of talent to Silicon Valley. I think they were proud they had the most computer science degrees in the valley for a while. Um, but, you know,

4:30

with the distribution of of of private equity and and company building, um, you

4:36

know, there’s a there’s a lot more happening in Atlanta than maybe there was 1015 years ago. Yeah. Well, yeah. Well, I think it’s

4:42

clearly been scattered. You know, um, you know, when you mentioned like Austin, you know, we lost it in Boston.

4:47

You know, it’s interesting too because in Boston it was really AI that um lost it for us, right? Because like MIT was

4:54

really kind of driving our ecosystem around AI in the n early 1990s, late

5:01

80s, early 1990s and it was clearly decades too soon. Um and then just the

5:07

exodus and we missed the PC, you know, uh era and that really kind of went that was the impetus for computing to really

5:13

kind of move out to the uh west coast and um and really be rooted and grounded in uh Silicon Valley. Um it was going to

5:21

be there anyway, you know, but there but I think um Boston was a really large tech hub and I think uh we lost some of

5:29

that, but it’s really been biotech, you know, now in our area. Uh but Austin for sure has uh has has picked up you know

5:36

especially with co we were talking about that a second ago a lot of exodus out of California you know into other areas and

5:42

uh in Georgia and Atlanta you know you got great talent you know there you got great weather you know also so good

5:49

ecosystem it’s a little hot this week but otherwise yeah yeah well it is we are getting close to

5:58

summer you know summer summer time all right so great so like Um let’s chat about um soentual and you’ve been

6:05

focused on network automation and it was interesting something that you said um you know um spawned an idea. So when we

6:13

started uh on that was back in 2012 um that that was really around kind of the whole openflow stuff and that’s when you

6:19

were probably doing that work with um onf and um you know and and the programmability aspect of openflow um

6:28

and I think so that was a little bit before its time you know because now sonic you know has really actually we in

6:35

the spring like a couple a couple weeks ago uh in May in Dallas we had our largest gathering of the community and

6:42

we had a large uh Sonic pavilion. Um, and the enterprise market is just ready for it. And it’s not just because of

6:48

having an operating system and the white boxes, that entire promise, but now they’re being packaged and now there’s a

6:54

library um of applications that you can use to run various different uh kind of

6:59

life cycle management um oh um I guess tasks and workflows. So I think so

7:06

there’s that one thing but I think the the more important thing that you I think you mentioned was um when we first

7:12

started doing kind of uh white box you know and um kind of um and computing uh

7:19

we were trying to do the programmability aspect of um of networking to kind of catch up to where storage and where

7:26

computing has gone and um and I think you know we’re we’re probably there now um maybe we’re still I was we’re always

7:33

still a little bit behind But I think now um the major focus is that uh we had storage and compute go to cloud um and

7:40

now we’re really now starting to focus on that within networking also with the NAS stuff and even like what even in the

7:46

collaborative uh the owner collaborative we launched a uh it’s it’s called a WAN connect API and uh working group so it’s

7:54

defining an API but what’s but what what’s more important is what that

7:59

enables it really enables kind of networking as services, you know, but I digress. But the only thing only point

8:06

I’m making, Chris, and because you’ve kind of lived this now, is networking trailing behind major compute

8:12

evolutions, um, and storage evolutions. And so, um, so now we’re kind of, um, at

8:18

this, uh, phase in networking where we both have both of these things interacting. We have kind of

8:24

programmability and we have, you know, AI as aiding, uh, and a betting, uh, in that process. And then we also have um

8:32

you know the the beginnings of a of a different kind of market where you can consume networking uh more as a service

8:38

than to actually build it all out uh yourself. So um so maybe you know let’s

8:43

take that as a you know uh potential and your focus now on you know um on the the

8:50

automation and the programmability aspect of of network infrastructure. Yeah. So some of the evolution um you

8:59

know back in 2012 for sure uh when you started was you know we were really using tools and EMS’s um and the idea of

9:08

making the network programmable is we were going to have new interfaces that you know these these were built for

9:14

humans tools EMS’s CLI it was it was very and it’s going to come full circle

9:19

when we talk about AI and we start talking about putting stuff back into English for LLMs but you know these systems were largely

9:24

built for humans and the idea a was that we needed machines to be able to activate networks. Um, and so whether it

9:31

was netcom or whether it was an ODL controller, the idea was we were going to have programmability. So we very much

9:38

focused on integrating with these programmable constructs, mainly controllers. Um, and I think there was a

9:45

large push for a long time going from CLI to API. um you know adopting

9:50

controller infrastructure and then something interesting happened in the sense that you know I think a lot

9:55

I give the vendors a hard time quite a bit but I think most have done a pretty good job with put building programmable

10:00

controllers um and then when they didn’t have that the really the devops community stepped in

10:06

you know namely anible and terraform and python to kind of augment so you kind of have these two worlds you kind of have

10:12

this very devops um IA uh type of world and then you have this controller

10:18

centric programmable world where maybe you know a controller might look more like Zcaler or AWS you know here in the

10:25

near future as you look at like you know Paulo and some other people putting programmable access and you have to marry these worlds up so we didn’t

10:33

really go from CLI to API we went from human centric to programmable but the

10:39

the ecosystem of technologies to make that happen I don’t think anybody would have guessed but um the amount of

10:46

concentration um has really I think helped the enterprise market because now there’s

10:51

not 42 things to pick from. You kind of have a smaller ecosystem of things that you need to work with which I think has

10:57

really accelerated the adoption of of kind of digitizing network operations.

11:02

Yeah. Yeah, that does make sense. you know, because I think, you know, the the main argument that I’ve kind of seen in

11:10

this space was that okay, well, do you kind of like roll your own with like whether that’s Anible, no, Terraform,

11:18

you know, Python, um or you know, do you buy a platform um and then you kind of

11:24

invest in that platform and um and a lot in I know in the new community they went

11:30

to build their own and they regretted it. Um because all of a sudden they have

11:37

to feed that beast and they have to maintain it and keep it up and running all the time. Um and so um some chose to

11:45

to move towards a platform um kind of an automation platform. So I’m not sure

11:51

what you’ve seen like in you know in our customer base and you know and I’m not sure if there’s a typical customer

11:57

journey but like maybe there’s a you know what’s the dominant kind of path that you’re seeing.

12:03

Yeah, I usually think history kind of influences how how we behave today. So I think there was a long time when

12:09

networking and network management vendors built very closed ecosystems. Um

12:14

I’m trying to leave names of companies out but you know you might have an you might have a small inventory system, you might have a performance management

12:21

tool, you might have you know this that and the other. Um and then you have a set of EMSs from your hardware vendors

12:26

that aren’t very API or programmable focused. So most organizations had to build a tools team to tie all this stuff

12:33

together. So that’s you know that predates network programmability. So because because the

12:39

ecosystem hasn’t had the most robust set of tooling like if you look at the developer software developer the tooling

12:45

is unbelievable right um I would just say for for networking professionals it was less than ideal you

12:51

know before programmability. So that’s kind of baked into how we operate. Um and you know practitioners are solving

12:57

problems. Um they might not be concerned with building a platform for the whole organization. They’re trying to solve

13:03

their challenges. So the DevOps community I think did a great job stepping in providing you know libraries

13:08

like Net MO anible modules Terraform providers to give us programmable access to all this stuff so we can we can build

13:14

cool stuff. Um and then customers kind of have a decision. and they’re like, “Hey, do I abandon my stuff or, you

13:21

know, and to your point, do I build a do I buy a platform?” And what we’ve tried to strike the balance is really have a

13:26

developer friendly platform that also provides a lot of out of the box functionality. Usually when I talk to

13:31

customers or prospects, they’ll say, “I built this part and I love it like like like I love that part. I had to

13:38

build all this other stuff and I wish I didn’t have to maintain that anymore.” So I I tend to think it’s

13:44

less binary. You know, somebody might build might have built some business logic to roll out firewalls that they’re

13:49

super proud of. It maps to their business processes, supports their vendors. Like, why not leverage that?

13:55

There’s no reason to throw that away. Um, but I just view it through the lens of there is so much to do to digitize

14:01

our infrastructure that um, organizations that adopt kind of an all

14:07

of the- above approach is really are the most successful. Um because when we talk

14:13

about automation a lot of times we’re talking about automation is very domain specific. So how I automate my data

14:19

center like if I pick up like abstra or something super awesome but I can’t use abstra to manage my WAN per se um you

14:26

know and the tooling in the cloud you know you’re not going to get Terraform out of a cloud SR’s hands unless they’re

14:32

using you know CDK or something that’s that’s vendor that’s cloud specific. So

14:38

automation being distributed like that, you want to use the right tool for the right job to help people solve their problems. The question for most

14:44

organizations is how do I take these islands of automation that I’m very proud of some and maybe others need

14:50

help. How do I tie them together so that I can share and take advantage of these things? That’s

14:55

yeah, that’s what a great approach. That’s such a realworld, you know, uh realistic, you know, brownfield kind of

15:02

like, you know, approach, you know. It’s like it’s it’s kind of like both green field and brownfield like kind of combining those together. Um where

15:10

others are really kind of like trying to make organizations make that choice, you

15:15

know, one versus the other. So yeah, congrats. So yeah, so I would think that that’s that message has to resonate

15:22

really well um you know uh in the marketplace. You know, it’s like you’re right. there’s like c certain things

15:28

that you put a lot of time and energy and effort in that is um you know that you’re really not only proud of but it

15:34

really works and you don’t want to you don’t want to lose it. Um um but then there are other things that you’re just so tired of feeding that beast

15:41

maintaining it you know you love to give it up. So, um, yeah, so if you can kind of marry those two worlds, that’s that’s

15:48

a win-win. That’s, uh, yeah, that’s really great. Super. Yeah. and the pressure on these organizations from like a governance and

15:54

enterprise, you know, whether it’s managing secrets or GRC compliance

16:00

issues or security scans of their software. Like these, we’re really moving from a

16:05

world of tooling where I use a tool to do my job on my desktop to enterprise software that has to comply

16:12

with all of the enterprise software requirements of the organization. And as people make that transition um that’s

16:19

really a point where they really evaluate how much they want to invest in different areas um you know tooling

16:26

platforms um whether it’s build or buy. Yeah that’s well no that’s yeah really

16:33

interesting. So that’s great you know kind of a a transformation that that you’ve been kind of like mapping um you

16:40

know across the industry and across your customer base. What about um so let’s

16:45

talk a little bit about now how AI now gets involved in this uh in the automation um of um of infrastructure

16:53

and its life cycle management. So there’s the uh orchestration and orchestration automation and then

16:59

there’s also just day-to-day um you know u management you were mentioning before you can kind of chase alerts and alarms

17:05

and logs and so forth but that’s that’s also becoming automated. So now I guess

17:11

what have you been thinking about how this is going to impact your product set and um and how it may impact your

17:17

customer base. Yeah, things are moving fast. I I think the Linux Foundation got a from uh

17:23

Google yesterday. They they they gave it to the Linux Foundation like literally in the last 24 hours. So

17:29

Oh, they did. Yeah. Well, good for them. That’s um that’s that’s an important piece in the hologenic um you know uh

17:35

journey. Exactly. So um what we announced a little bit ago and our participation I

17:42

think in the first little bit people were building chat bots and different ways of you know uh hooking up LLMs to

17:48

documentation and stuff which is uh you know innovative and interesting but when we think about how to apply this to the

17:54

actual automation orchestration stack you know MCP um I think kind of changed

18:01

at least my perspective on what was possible um because we’ve been building around

18:06

these APIs guys um for the purpose of like we’ve been talking about programmability.

18:11

And if you run an automation 50 times a day or 500 times a day, you know,

18:17

somebody might fill out a form, somebody might open a service now ticket. It’s a very like discrete event. Um it’s a very

18:22

onpurpose type of activity. Um but if you start to think of like building

18:28

instead of services, you build capabilities and you expose them via an MCP.

18:34

Now you’re basically building a library of assets that the LLM can choose to use

18:40

kind of at its choice, you know, when it decides that’s a good idea. So instead of building kind of large

18:46

CRUD operations, you might build very small capabilities and let the LM piece

18:52

together what activities it wants to do. So the way we’ve talked about it is the workflow of today. Um, and I mean

18:59

workflow in the very generic sense, whether it’s like a GitHub pipeline or whatever. Um, the workflows of today are

19:06

very static. Um, and we think as we get to these MCPS, it’s going to be very much a reasoned workflow. Um, but our

19:15

participation is really taking all of the stuff we’ve built around automating components, orchestrating components,

19:20

building the life cycle of product and NAS services, but now instead of just exposing them to the IT team or the

19:27

application pipeline people, how about I expose it via MCP to an LLM? Um, and let it reason and decide what to

19:35

1. Yeah. Awesome. And then obviously

19:40

people’s first concerns are what what is Gemini or whatever or my llama model

19:45

going to do on my network. So you know well there is that you know yeah the whole security people are like you know

19:51

um like on once they started hearing about agents talking to agents and a directory of agents and uh where the

19:57

agents can consult other agents about doing jobs and distributing them. Uh yeah that’s like you know you know every

20:04

hair on their back goes to their shirt. It’s like they Yeah. Yeah. So I think that’s that’s that’s

20:11

where this will head at least from our perspective is that you are going to want to have um discrete activities

20:18

exposed to the MCP to to to basically constrain what activities you’re going

20:24

to allow. So you could imagine in the early days maybe only giving it workflows that have readonly access just

20:31

making like hey we’re gonna you know we have hundreds of changes in the network we’re

20:36

going to flow those through our standard change management process. We’re going to start off with letting

20:41

the LM have access to all the readonly activities. You know what’s the status of this? How’s this upgrade going?

20:47

What’s this product life cycle? Why did this get turned off? You know whatever’s going on. um so that we can start

20:54

deciding what we want to feed it. And it it it very much mirrors when people first started talking about automating

20:59

infrastructure. They wanted to do readonly things. They wanted to have human in the loop. You know, I wanted

21:04

checks and balances. I want to see the diff before you push it in the network. I want to see all these things. And I think you know that move from

21:11

human to automated or API. we’re going to see the same kind of human reaction,

21:18

governance and compliance requirements before we go from kind of API to a

21:23

reasoned unknown outcome generated by um the model.

21:28

Yeah. So I I think that the whole trust piece there is really going to be uh key, right? And I think maybe that

21:35

process where um where we where we we’ve kind of moved before um as we moved into

21:42

automation where there would be kind of checkpoints and guard rails you know in uh human intervention to kind of check

21:47

before uh you deploy um maybe a new configuration and have review either

21:53

some review or some have somebody kind of like you know in that loop. uh maybe okay organizations have learned that and

21:59

if we can kind of repeat on that you know as you expose these libraries now

22:05

um via MCP to LLM um and have a human kind of in a loop there maybe that’s the

22:11

uh maybe that’s the path to building trust um on these and um you know does

22:16

have kind of a ring of truth you know associated with it especially since we did it before when we had to build trust

22:21

in the systems that u or the automation um you know kind of software that we were kind leveraging and building at the

22:28

time, right? So is is that kind of um do you see maybe another way in which trust

22:33

gets built? So um you know the models of today are

22:40

generic. Um you know they consume the internet they consume all publicly available repos all that kind of stuff.

22:46

Um what is what I believe will happen that will be transformational is that we

22:52

will start getting specific agents with you know very specific knowledge. So um

22:59

you know I’ve been pushing not that they listen to me that you know equipment providers can provide kind of like you

23:05

know a CCI in a box or a JNCIE in a box or a Zscaler expert or an Arista

23:10

validated design agent type of thing. And I think when I can ask via a flow,

23:19

you know, an agent that has all the knowledge of all the tact tickets, all the best configurations, all the

23:25

validated designs, when I can trust that, yeah, then I can start to in interact in my in

23:31

my in my workflow. So I can ask today I can ask it like what’s the risk of me making this

23:37

change? It’s going to give me a pretty interesting output and I can put that in the ticket for a human. But, you know,

23:43

if if my CCI agent knew the exact defects on the version I’m going to and

23:48

understood the configs of my network and understood best practices related to EVVPN deployments, you know, that’s that

23:56

I think that type of knowledge having access to that type of knowledge is going to change just like and from a

24:02

software development life cycle perspective whe whether it’s the vibe coding stuff or claude or whatever that

24:07

the quality is of of the point where people are starting to defer to it. So, we need we need we need some

24:14

we need some agents that we trust technically um to answer those

24:20

questions. And yeah, that’ll be Go ahead. I’m sorry. I’m sorry. To your point, it’s it’s it’s you know the the the metaphors are all

24:26

around us in the sense that we used to automated off of spreadsheets, right? And then we automated off of databases

24:32

and then the discovery of the networks went from discovering of devices to discovering the fidelity of the

24:38

discovery is so awesome these days that I will automate off of it where I would not in the past. So it’s really like the fidelity and the

24:45

quality of the questions and answers I can get is going to be directly tied to the level of trust we can build within

24:52

the organization. Um because obviously our desire is to to to move aggressively

24:57

because AI AI at its core is an extension of automation. Yeah. Oh yeah, for sure. It’s like, you know, this is what’s what I I find

25:05

really kind of refreshing is that one, you’ve not that I wouldn’t have expected this, but you’ve thought it through, you

25:11

know, and you kind of embraced um you know, these protocols uh these agentic

25:16

protocols that are that are coming out and and understand how they can be used, you know, within your platform um and

25:23

how your customers can benefit uh from them. So I think you know it’s like um it’ll be interesting to see you know if

25:30

the large um kind of networking suppliers actually deliver agents you

25:36

know I wonder how much of that of that would be conflicted um you know with you

25:41

know their their customer base you know I wonder if it would just be required to like maybe third parties to start

25:47

building those kinds of agents that have you know you know high level CCIE you

25:54

know kind of like, you know, backgrounds, you know, or that can really kind of learn um the um um uh an

26:01

environment um you know, in detail. Um just looking at all the configurations, all the changes that have been made over

26:07

some period of time. Um being able to understand, you know, um you know, conflicts that may arise between uh one

26:15

version of software, another version of software, you know, or various different um configuration changes, you know. So

26:22

that’ll be really interesting to see that kind of entity, that kind of agent, you know, come into the marketplace. Uh

26:28

because then you’ll have kind of a group of the these agents working within the flow. Um and you can almost imagine like

26:35

once we build trust in these systems they become highly automated you know

26:40

and and it’s almost like you know it’s like you know Chris like our industry has

26:47

been really good at uh we build these really complex systems um and then there

26:52

would be a company that gets started uh to try to manage that complexity and so we built this one on top of the other um

27:00

and so now if we um are are able to abstract that complexity with uh agents

27:07

that are interacting um and that we trust and that really have a very deep knowledge of the infrastructure. Um the

27:14

infrastructure will get highly complex but we might not even know it more than more than likely we just won’t know it.

27:20

You know the complexity gets totally mashed with the automation. Yeah. I mean whether it’s Google or

27:27

otherwise we usually um talk about these very very automated environments um and

27:34

how impressive they are which they completely are um and then the question

27:39

is why can they do it and why everybody else can’t um and you know a lot of times it’s because you know they might

27:45

refresh all their gear every three years or you know they standardize their protocols or they build their own OS or

27:50

whatever the whatever the anecdotes are. Um, but I think we’ve really been

27:56

challenged with abstractions. You use that word. Um, you know, from a computer science perspective, it’s it’s it’s it’s

28:02

everything. So, we’ve been abstracting human interfaces. And that’s why I kind of keep coming back to that. That’s why I thought, you

28:08

know, 2014 was so pivotal is we started looking at how we can build programmable infrastructure. And if we solve the

28:16

foundational programmability concept, then we can build amazing abstractions. we can build um you know much more

28:24

trustworthy systems because when you build software on a human interface

28:30

that’s been our generic struggle. I mean the CLI was an amazing invention in the 90s right with a stroke with a few

28:36

strokes of a keyboard I can program BGP across a very complicated network. Um

28:42

but as we went to machines um you know the compute world has both the CLI and programmability as as we can talk about

28:49

but we’ve we we need that programmability as a foundational component and and most of these other

28:54

problems can be much simpler to solve to your point instead of trying to build complexity on complexity.

29:00

Yeah. Well it’s interesting too because like um we have almost like kind of industries that are built upon that kind

29:06

of model u or our industry you know built upon that kind of model. Um you wonder like how many companies this will

29:14

displace you know um like the tooling um you know in particular you know

29:20

marketplace. So that’s um you know um as we move more into this uh kind of

29:25

agentic for um infrastructure and life cycle management you would think that

29:31

there would be some rationalization you know um of the market and uh

29:37

yeah I would think there would be some rationalization of this marketplace you know it’s like as we abstract all this

29:43

stuff and the and the agents are able to kind of deal with the complexity and um and manage the infrastructure you know

29:49

um the need for uh kind of tools that try to reduce that complexity go away.

29:54

You know, it sounds obvious. Yeah. Well, we we focused on business logic and technical logic for a long

30:01

time, right? How do I push this config to this box and which OS is it and what does it support and you know like

30:09

super valid complicated things. Um

30:14

but when you think about a future where you know you have a CCI agent sitting next to you maybe your focus is on you

30:21

know productizing your infrastructure life cycling the components um yeah you know the compliance and governance

30:28

that’s related to make sure that the that what we’re doing falls in line with our business outcomes. Um it’s it’s

30:33

definitely transformational. No doubt about it. Yeah. You know what’s interesting too because like there’s always like and

30:38

this has been a kind of an age you know um or kind of a long-term you know

30:44

problem that maybe yeah I guess it is a problem uh in the industry is that it in general speaks a very different language

30:51

um than business managers and kind of executive management you know um so

30:58

whenever there’s like discussions usually with it there’s like you know they’re kind of you know dazzle them

31:04

with like science you know, it’s like until like the board says, “Okay, whatever you need, you

31:09

know, not maybe not whatever you need, but okay, great. You you know, um I don’t totally understand it, but I trust

31:14

you and that um that you’ll go off um you know, and build that, you know.” So in this world you know um can you see

31:22

that you know that business logic informs a policy engine um that uh

31:30

informs the agents and so um you know so now you have a much more direct link and

31:36

tie uh into the business uh at hand and it’s not um kind of constructing this

31:43

infrastructure in support you know of what you think the business needs but now it actually could be directly from

31:50

the business down you know into the infrastructure around configurations and

31:55

change management um new services, new products um you know a real focus on

32:01

kind of service delivery um in support of both customers and uh and markets.

32:07

So, you know, there, you know, outside of like kind of the technical changes that we’re talking about, you know,

32:13

there’s got to be a bigger, you know, effect, you know, of of having an

32:18

automated uh infrastructure that gets its directions um mostly from business

32:25

logic and policy engines. Yeah, I’ve been pretty impressed with

32:30

the thinking over the past 18 months. Um and I think the result of that is that b

32:38

lines of business teams went directly to the cloud for a bit, right? So um you

32:43

know uh this team builds their apps in a AWS, they build their data links in GCP

32:49

and the those app teams became very comfortable with the cloud operating model. Um and you would hear silly stuff

32:56

like hey we’re just going to go all the cloud and turn off our data centers and shut down it all that kind of stuff. And now we’ve gotten to a more balanced view

33:03

and the regulation and compliance requirements are pushing it to take you know greater responsibility of what was

33:10

kind of shadow IT or whatever words you want to use in the lines of business. Um but those teams have a strong desire for

33:18

a cloud operating model which most of the time we can’t offer internally. Um and I see two two paradigms shaping

33:25

up here. One is more of a shared understanding where the line of business team and it work together to achieve an

33:32

outcome to your point. So maybe the business defines it in a policy engine, but everybody knows what’s going on. So

33:38

we have this super important app. I’m a transportation company where, you know, we’re moving things around. We we it is

33:44

providing some knowledge. Business is probably policy. It’s all tightly coupled. Um I think that’s interesting.

33:52

Um, I think that the most scalable way to do this is to have it be a service

33:58

provider just like you got from AWS in the sense that they provide componentry.

34:03

You know, like I go to AWS, I get load balancers and EC2 instances and storage

34:08

arrays and stuff. Um, and you let the business actually build their apps on top of your

34:15

componentry, your capabilities that you expire. So that way the the app teams and infrastructure teams can kind of

34:21

work independently. Um just like AWS doesn’t call me up when they do maintenance on my infrastructure, but if

34:27

it’s a tightly coupled stack, I might need to call my application for instance and say, “Hey, I’m migrating you to a new F5. I need you to update your

34:33

Terraform plans and re republish your application.” So the maturity really goes from, you know, disconnected where

34:40

line of business was doing their own thing. I really see it lining back up now, which is super exciting. Um, in the

34:46

sense that there’s respect for the business outcomes that need to happen and all the compliance and governance that it supplies. And now we’re thinking

34:52

about how do we how do we bring product thinking and product um development into

34:59

IT organizations to build these the these components that can be consumed by application teams. And that’s that’s

35:04

probably the more the more advanced customers we see. Yeah. Yeah. Well, that makes perfect sense. And I think it’s only going

35:10

that’s only going to accelerate um you know as um as you know we’re very very early on in the agentic you know um

35:16

space and how you know we you know what we’ve been focusing on and talking has been around just kind of the operations

35:23

models and how to automate those you know uh but there’s also too like most

35:28

networks um you know at the last the concept of a data exchange really took hold and so

35:37

because like so many in community are pushing um pabytes of data around you

35:43

know it’s like you know eBay is like you know pabyte a day um FedEx every 30

35:48

minutes a pabyte you know is kind of moved around um you know their their infrastructure and so they’re not

35:55

thinking of the infrastructure as like okay I’m connecting this to that you know this branch office to like this

36:01

regional site you know these desktops to like you know this data center they’re starting to really just think of it more

36:07

as a fabric you know and where its main job is the movement of that data and uh

36:13

and all the rules associated with okay well is that data allowed to move there you know is that user um allowed to gain

36:19

access to that data so it’s it’s a it’s a very different kind of view um of of

36:26

how they’re kind of visualizing and and trying to manage the infrastructure and the other key thing with that too is um

36:34

how do we move that data around so it’s they’re not they don’t want to aband and in kind of routing. Um, but it’s like,

36:41

you know, does does stateful kind of protocols really work in this new world

36:47

or do we need to go more stateless, you know, and that kind of brings us almost like full circle, you know, back to

36:52

like, you know, the whole open flow and controller, you know, pieces because those were stateless, right? And so like

36:58

you would have a controller that would create the path and that controller would get informed um by these agents

37:05

around um conditions on the ground, right? Whether there’s congestion there or the path from point A to point B,

37:13

what’s the latency associated with it? What’s the dollar cost associated um you know with it? Um what’s the roundtrip

37:20

time delay you know uh with it? Can that data move out of that country you know or not? So, so there’s like so there’s a

37:28

whole other you know kind of dimension of this outside of just the operations life cycle management of it but actually

37:34

the movement of this of this data around so I think you know and it’s interesting to see how they’re kind of they and

37:41

thinking right now anyway get married together you know but we are still so early in this and so much is going to

37:48

change um you know around it but it is moving fast so you know you know you

37:53

can’t just like wait and you know and uh and uh and just see how it all works out. You actually kind of jump have to

37:59

jump in just like you did with your products, you know, right now and and how you’re now kind of uh envisioning,

38:05

you know, how um you expose um this abstraction or or expose some of your

38:11

automation into like um AI through MCP and you know, try to leverage the ADA

38:16

a2a yeah ADA protocols and so forth.

38:21

Yeah. So yeah, I don’t uh I’m not sure if you’ve seen that like within your own customer base like you know how you know

38:28

the thinking around infrastructure is starting to change in terms like what does it do and and how does it um

38:35

um what is the value that it delivers because you were just talking about the cloud side of things. It’s like when you

38:40

have AI and you have cloud um it’s you know all of a sudden like you know the

38:46

um patterns are very different processing you know is different you know latency is different and it’s

38:53

almost like the um you know kind of you know networking is

38:59

becoming inverted you know in terms of like what we would normally expect networks to do and deliver and so um I

39:06

think we’re at like you know we’re at a place right now in the industry where we’re all starting to like reimagine um

39:13

what does this infrastructure kind of look like? What does it do? And how do we do it, you know, and the value that

39:19

it creates, you know, uh for the organizations that we work for. Yeah. So, I think there’s probably two

39:26

two things you said that that uh I think are interesting to talk about for sure is, you know, ATA and the concept of AI,

39:34

I mean, we’ve seen this in the world, right, is highly highly concentrated. I saw a map the other day of like where

39:40

the AI capable data centers are globally. Um, highly concentrated. Um, so what that means most likely is as we

39:46

start to adopt this technology, it’s going to be highly centralized. You know, you’re you’re going to have a strategy of where your training and

39:53

inference models live. You know, one interesting way to think about it is how do you wire that up to your

39:58

infrastructure? Like how how does that happen? Um, and you know, people would

40:05

have all sorts of comments of we we the the honest answer is we don’t know how it’s going to happen. Um, MCP is a great

40:10

uh great first step. Maybe it sticks. I hope it does. Um, but how do you bring the data to your point and how do I

40:17

bring the capabilities and expose them in a controlled trustworthy way? So, that was one thing. And then um I think

40:25

admitting uh I mean I think everybody’s admitting this that you know this is going so fast that we’re not sure where

40:31

it’s going to go. What that tells me we need to do is we need to build small componentry

40:37

um that can be used in some interesting way in the future. Um and I think from a networking

40:42

perspective that doesn’t feel natural to us. I think usually we want to go gather the requirements of of the you know what

40:49

sort of network do we need? We’re doing a high frequency trading situation. I need this that and the other and we go

40:55

build the solution for that problem set. Um yeah, if you think about the how the cloud folks build things, they build

41:01

componentry, they build S3, they be build EC2, and they’re saying go do something innovative with this. So when it comes to the data exchange and some

41:07

of these other concepts, it’s like we build sometimes I say we like to

41:13

build products nobody uses from a network services perspective. And when I ask people why did we build this, they’ll say because I was trying to make

41:19

it easy for the app team. I was trying to make it easy for the business. But what what but the disconnect there is

41:25

the business wants pieces that they can put together without the the cycle of innovation um that’s required. So how do

41:31

we build small componentry that can be consumed by MCPs or or this future ADA interface and how can I let my business

41:38

stakeholders take advantage of that that componentry and build what they want on top. Um, and I think I think that

41:45

product mindset and that cloud operating model are kind of like macro trends underpinning um, you know, a lot of the

41:52

stuff we’re talking about today. Yeah. Well, that’s interesting. It’s almost like the movie What about Bob? You know, it’s like baby steps, right?

41:58

You know, it’s like do the do the componentry, you know, first and then, you know, kind of baby steps into like,

42:04

you know, uh, what the new design will look like, you know, and then it it, you know, it starts to build upon itself.

42:09

Yeah, I think that’s that’s that’s a as good approach as anything else, you know, right now as we start to try to

42:16

think about how um you know, provide thought leadership around how um you know, others who are kind of like now

42:22

operating these infrastructures uh kind of move forward. So yeah, I don’t know Chris, it’s like you know one heck of a

42:28

time you know in our industry. So where do you think you know I maybe I want to come back to one thing that you said you

42:34

said you know very concentrated around uh around AI data centers for sure you

42:40

know um but there’s another thing too like if we extrapolate out like to like 2030 um David Sax did a really good job

42:48

talking about this in one of the all-in-one podcasts you know where he was extrapolating you know kind of the

42:54

growth rate of AI so he talked about the models he also he talked about uh the chips and also the data centers and I

43:02

think you know so I’ve kind of listened to that and then I actually I kind of built the equation and if you kind of

43:09

map that out like in 2030 or so you know we’re looking at some AI data centers

43:14

that are just absolutely faking massive you know whether they’ll be at one location or a cluster of a group you

43:21

know is to be seen but you know we’re talking between 5 and 10 million GPUs

43:27

you know um you know 10 100 you know um you know gigawatts you know of power

43:32

maybe a terowatt if we can actually unleash you know kind of the uh power generation you know uh in the country um

43:40

these are these are massive massive compute centers and you would think that

43:46

no IT application will escape that right so it’s a either it’s going to get sucked into that um you know that

43:53

capacity um or it’s going to feed you know all these other you know kind of

43:58

private um you know um AI data centers and so I think there’s so there’s one there’s there’s this um massive gravity

44:06

that is that is going to be created and then the other stat you know that um

44:13

that was really kind of talked a lot about you know in in Dallas you know at the summit was

44:20

the crossover um of AI spending between hyperscalers and the enterprise uh will

44:27

happen in the next few U where in 2030 there’ll be more spending on in the enterprise on AI

44:33

infrastructure than the hyperscalers. So that basically says is that we are

44:38

going to be in a hybrid world uh for the foreseeable future and that we really need ways in which any of the thinking

44:45

around infrastructure has got to be can’t be isolated that relying upon just hyperscalers for your AI or only you

44:53

know your you know your own kind of infrastructure for AI. It’s going to be a hybrid you know it’s so we have to

45:00

start kind of factoring that into the thinking you know as well. And so in our discussion around automation, how do you

45:07

kind of bridge between that gap, you know, it’s like all the cloud providers have their own tools around automation?

45:13

Um they don’t really expose them, they kind of take care of them, you know. So um like in in your work, you know, has

45:21

you know has the product set um um been mostly focused on the enterprise piece

45:27

of it or are there kind of extensions that you can have into the cloud providers?

45:33

Yeah. So, um, some decisions we made early on, um, were that we integrate

45:40

with technologies, not vendors. So, kind of the previous generation in

45:46

our space would build like a Cisco driver, a Juniper driver, an Arista driver, and you kind of get in this like

45:53

arms race all the time. Yeah. And I kind of learned it this with my time at Micromuse. I would say, no

45:59

offense to my micro muse friends out there, but we would spend half our R&D on all the integrations to HVAC units, 5

46:06

ESS switches, you know, all sorts of stuff. Um, and we had a I think an above

46:11

average fault management platform, but we won because we had an integration to everything like every every possible

46:17

device you have. So, I could consolidate all of your alarming for root cause analysis across all of your infrastructure. That’s how that’s

46:23

how we won. And then it was kind of the same thing on the on the automation configuration management side in CCM

46:29

market was you have a driver for all this stuff and you’re on this like you spend a significant portion of your R&D

46:34

on building integrations not building the best automation and orchestration platform. So it was it was tough at the

46:40

time but we decided we’re going to integrate with technologies. Um so that’s why things like ODL at netcom

46:45

were very interesting tailf at the time um anible terraform python um and then controllers swagger specs open API graph

46:53

APIs you know these types of things um so by doing that we kind of don’t

46:59

understand the nuances of the technology we’re integrating with we’re integrating um at at at an interface level versus a

47:06

specific box level. So I I say that background to say that like as these new technologies come out um they have

47:14

programmable infrastructure. So it’s very natural for us to be kind of domain

47:20

agnostic in how we attack these problems. So um we’re just as happy with uh VPCs and security groups as we are

47:27

EVVPN in the data center. And I think that’s how how most enterprise IT

47:32

organizations want to think about it is they want a strong set of capabilities that they can expose to their

47:39

application teams as a set of products. And those those business owners are

47:44

probably less concerned with the vendor you chose um or the 5,000 capabilities

47:50

it has. It’s more of like um you know to do load balancing at this pharmaceutical

47:56

company or this insurance company. this is how we expose it so we can get mass you know the business benefit and the

48:02

productization of either the AI data centers or these capabilities should follow a very similar similar path which

48:08

is really this this whole path of programmability um domain based automation tying it

48:15

together and then building products for somebody else to consume through automation um which is something else we haven’t

48:20

talked about but like for a long times we expose things in tickets and forms these end users are demanding APIs and

48:26

when we start hooking AI up to it. I don’t know if you saw some of these charts from live, but basically if you look at

48:32

the API patterns to these systems, they’re very spiky. Um, you know, when somebody’s doing something, it’s very

48:38

spiky. When you hook it up to an MCP, it’s pegged almost all the time because it’s going to use every it’s going to

48:43

use every second of available compute to do something, gather information. So

48:49

the systems that we have if they’re brittle, if they, you know, aren’t wired in to scale and other things, as soon as

48:56

you get into the MCP, you know, the power like, you know, so, you know, the the the other interesting

49:03

thing is the focus on these amazing AI innovations are actually doubling down

49:09

the requirements of kind of the core enterprise software um authorization models, um, you know, secrets

49:16

management, auditing logging um traceability through the systems kind of these these core platformy features

49:23

I think become even more important as we start to unleash um you know digital

49:29

agents uh doing what they see fit um yeah you know with with the guardrails we

49:35

provide. Yeah, Chris, awesome. You know, um I want to kind of like um we’re getting

49:41

kind of close to end of time. Sure. I thought maybe what might be good to end on is kind of a a future and a

49:48

vision. So it seems like you know what what is really focused on is exposing uh

49:53

when automating the infrastructure uh but also exposing it um to whether those are development teams you know um and

50:00

now to LLM through uh through MCP um and there may be one or two other ways in

50:05

which you’re exposing you know kind of that automation and the infrastructure too. So, um, how do you see this

50:12

industry moving over the next couple of years? Like, you know, do you, you know, what’s your extrapolation? You know, if

50:18

we’re looking if we look back and it’s 2030, what do we see?

50:23

I mean, I think every day that goes by, networking resembles compute and storage more than um, you know, a unique bespoke

50:31

set of of infrastructure. Um, so I think the trend towards networking,

50:37

you know, probably said programmability too many times on this call, but um, you know, the the ability to treat it like a

50:43

computer, a storage instance, um, obviously it’s a super complicated distributed systems, mad respect for for

50:50

what’s going on underneath the covers, but from a user perspective, the more we can make it universal. Um, and I think

50:57

you’re seeing this with acquisitions and the macro sense of, you know, equipment providers. If we look at the types of companies they’re purchasing and the

51:03

investment profile of what they’re doing, I think networking starts to look a little bit more like infrastructure.

51:08

We saw this with kind of the hybrid, what do you call them, the the like the nutanics of the world that start

51:14

smashing everything together. I think we start to see more of that kind of on a macro scale. And then, you know, uh how do we expose

51:23

this infrastructure to these agents and then how do we build products for our business owners? I think maybe that’s

51:30

only a couple years away, but uh I I think that’s that’s the core focus. A lot a lot of the other stuff is noise.

51:36

Um I think that’s the northstar uh that we need to be driving towards. Yeah. Awesome. Well, Chris, hey, thanks

51:43

so much. This was actually a lot of fun. um one I learned a lot you know also especially about your focus and uh what

51:50

your background also but also the um kind of like what you’re guiding you um

51:56

and the decisions that you’re making on products uh for your customers attention so uh so thanks so much for the time

52:02

Chris absolutely thank you very exciting times awesome great thank you thanks everyone for watching