Zeus Kerravala, Principal Analyst, ZK Research

ZKast with Bob Friday

Industry Voices AI & ML
Zeus Kerravala Headshot
Introduction slide with text that says, “ZKAST, eWeek,” And “This Episode: Zeus Talks With Bob Friday, VP/CTO at Mist.”"

AI is driving a major paradigm shift in networking says Mist founder

If you’re fascinated by the future of networking, you’ll want to watch this information-packed interview with Bob Friday, founder of Mist, which is now part of Juniper Networks.

Show more

You’ll learn

  • Why companies must put the end-user experiences first, second, and third

  • How Mist’s AI-driven Marvis technology now answers 70% of trouble tickets

  • The reasons for closely pairing the customer support and data science teams

Who is this for?

Network Professionals Business Leaders


Zeus Kerravala Headshot
Zeus Kerravala
Principal Analyst, ZK Research

Guest speakers

Bob Friday Headshot
Bob Friday
VP and CTO, Mist, a Juniper Networks company


0:00 [Music]

0:04 uh welcome everybody i'm zeus caraval

0:06 from zk research and i'm here

0:08 for another z cast video podcast as

0:11 always this is done in conjunction with

0:12 my media partner eweek i'm joined today

0:15 as part of a thought leadership series

0:17 with

0:17 bob friday who was one of the founders

0:20 of mist who's now actually part of

0:22 juniper networks who was

0:23 they were acquired a couple of years ago

0:25 now so bob why don't you say hi to

0:27 everybody introduce yourself

0:29 and give us a little background on what

0:30 miss does yeah thank you zeus and uh

0:33 thank you for having me here today you

0:34 know

0:35 my background has been mostly wireless

0:36 most of my career now i sent a thank you

0:38 letter the fcc every year for the

0:40 unlicensed you know

0:41 most my career has been spent building

0:43 mesh networks unlicensed wireless

0:45 controller type of stuff

0:46 you know i did airspace one time which

0:48 was really around wireless controllers

0:50 for the enterprise

0:51 sold that company to cisco and was cisco

0:54 cto for

0:55 six eight years uh and then left cisco

0:58 to go off to start mist

0:59 which is really around cloud and ai and

1:01 you know where i'm here with you today

1:02 to discuss

1:03 yeah and we all we we've kind of joked

1:05 over the years since uh you've been part

1:07 of

1:07 juniper that this was almost a reverse

1:09 acquisition where

1:10 uh well although juniper acquired missed

1:13 missed uh you can see the missed

1:14 influence

1:15 inside juniper now uh you know having

1:18 this ai

1:18 influence across the entire enterprise

1:20 portfolio so that's been kind of fun to

1:22 watch

1:22 i've noticed um the enterprise business

1:26 of juniper it seems like jim was almost

1:28 doubled down on the enterprise i think

1:30 in

1:30 during the last earnings call we just

1:31 had actually a um an analyst call

1:34 with some of the juniper execs and uh

1:36 romney talked about

1:37 enterprise being the growth engine he

1:39 actually predicted that pretty soon it

1:41 would be bigger than service fighter

1:42 which

1:43 i think a few years ago um no one would

1:46 have predicted that

1:47 right for for juniper to have an

1:48 enterprise business that's going that

1:50 fast so

1:50 can you can you expand on that you know

1:54 if you look what romney and the

1:55 executive team general are looking at

1:57 doing you know

1:58 juniper historically has been known as

2:00 kind of a high performance

2:01 routing company for service providers

2:03 you know romney and team are really

2:05 looking to diversify

2:06 juniper and really turn it into more of

2:09 a software company right

2:11 you know so you see a lot of things

2:12 happening at junior as it transforms

2:14 itself

2:15 you know we're extending the businesses

2:17 across sp

2:19 enterprise data center and security so

2:22 organizationally we're getting focused

2:24 around these four different business

2:26 areas and then inside those businesses

2:29 we're really starting to transform

2:30 ourselves from kind of the hardware

2:32 to more of a day zero day one day two

2:35 software company

2:36 you kind of see that because that's

2:38 where our customers are driving us to

2:39 right that's where customers are asking

2:41 for help as these networks come

2:42 become more complex you know they're

2:44 really asking for their vendors to

2:46 actually become

2:47 you know help them with the software

2:49 problem of how they deploy and operate

2:51 software and how we're going to operate

2:52 these networks going forward

2:54 so i think that's where you're seeing

2:55 romney and the team actually being

2:57 driven by

2:58 both the investors and and from the

3:00 customers that are driving us more

3:02 towards this

3:03 diversified company with more of a

3:05 software focus

3:06 yeah now one of the the principles with

3:09 which i conduct my research is

3:11 has always been uh that market you know

3:14 share gains an opportunity to create a

3:16 remarketing transition

3:18 and i think in the enterprise network

3:20 space it seems like

3:21 we're in this big transition where more

3:24 and more networking

3:26 is moving to the cloud right in fact if

3:28 you look

3:29 at the impact clouds had across all of

3:32 ikea

3:33 everything's moved to the cloud except

3:35 networking it's kind of lagged behind

3:37 so can you talk a little bit about

3:39 what's been driving them

3:41 yeah you know if you look at the origin

3:43 story from mist right i mean

3:44 sushi and i were at cisco when we

3:47 acquired meraki right

3:48 you know and that was really the

3:50 beginnings of how cloud was going to be

3:52 able to simplify

3:53 networking uh but what we started to

3:55 really hear from customers when i was

3:57 there

3:58 was really you know we're watching kind

4:00 of these wireless networks go from a

4:02 nice to have

4:03 to a must-have to really they're going

4:05 to business critical and i had a couple

4:07 of very large customers tell me bob

4:09 hey before we put any consumer

4:11 experience on our network

4:13 you know one you had to stop your

4:14 controllers from crashing you know so

4:16 they wanted better software reliability

4:19 two they wanted me to keep up with the

4:21 uh their mobile

4:22 you know digital transformation projects

4:24 you know they were building mobile apps

4:26 every month

4:27 you know the network was still being

4:28 updated every year type of thing

4:30 so they wanted to keep a speed and

4:32 probably third more importantly

4:33 was i call the paradigm shift from we're

4:36 not here to help them manage

4:38 access points or routers which is what i

4:40 did at airspace right you know there the

4:42 paradigm is helping customers manage

4:44 these ap

4:44 switches and routers really the paradigm

4:46 shift is really helping customers manage

4:48 the

4:49 end-to-end user experience what we call

4:51 client-to-cloud

4:53 and it's really they wanted visibility

4:54 you know before they put a robot

4:56 or a mobile app onto their network they

4:59 really want to make sure they have

5:00 complete end in connectivity

5:02 and visibility you know all the way from

5:04 that client device to the internet

5:06 and that was really the big trans market

5:08 transition that you know really

5:10 drove miss you know it's one of the

5:11 reasons sujay and i left cisco right

5:13 and we really took a bet that you know

5:16 networking required both like cloud

5:18 and ai architectural change you know and

5:21 if you were going to do this

5:22 you're really easier to do with a blank

5:24 sheet of paper because it's really an

5:26 architectural change that's happening in

5:27 networking right now

5:28 when people think of mist uh people

5:31 think of you know in fact i'm looking at

5:33 the little background behind you've got

5:35 little wi-fi symbols on there

5:37 and when you launched you were a wi-fi

5:38 company you're part of gartner's

5:40 you know wi-fi you know wired fire

5:43 martin's land

5:44 uh magic quadrant but in fact um

5:48 give the people watching this on on an

5:50 e-week in my youtube channel a little

5:51 more detail on what this is it's not

5:53 really a wi-fi company as much as it was

5:55 a cloud ai company correct

5:58 no that's correct that's what i'm saying

5:59 if you look at the origin admits

6:01 it was really around the thesis of you

6:04 know cloud is this fundamentally a

6:05 better way to develop maintain software

6:08 you know and that's why people are

6:09 starting to move more and more of their

6:11 applications to cloud

6:12 you know one is the innovation speed you

6:14 know the fact that we can build

6:15 software more reliably and at scale on

6:18 the cloud

6:19 and then the second thing is this ai

6:21 transition right

6:22 and that was really kind of the other

6:24 part of mist is you know can we really

6:26 build something on par with a human

6:28 you know and when i saw i know if you i

6:30 don't know if you remember watson you

6:31 remember watson playing jeopardy

6:33 yes yeah you know that that was kind of

6:36 one of the inspirations is hey you know

6:38 if they can build something they can

6:39 play jeopardy on par with

6:41 you know a jeopardy champion you know we

6:43 should be able to build something that

6:44 can really

6:46 play jeopardy on par with a network

6:47 domain expert

6:49 and that was kind of that you know when

6:50 i started to realize that you know

6:52 technology was happening right ai was

6:55 really going from this marketing story

6:57 into something that was going to be

6:58 useful for all different industries

7:00 including network right and beyond just

7:02 cars

7:02 you know we're going to see aai actually

7:04 actually make a difference across all

7:06 our verticals

7:07 yeah so how close is it right now if

7:09 you're uh you know if you think of

7:11 the big leap that watson took you know

7:14 winning jeopardy uh

7:16 if you were to put missed head-to-head

7:18 um in a competition with the

7:20 the best network engineers out there is

7:22 it slightly behind is it on par is it

7:25 ahead

7:26 yeah so i would say one thing i found

7:28 also missed doing this

7:30 you know there was the architectural

7:31 change of the us making sure we actually

7:33 could build the right architecture and

7:34 pipelines to actually process this data

7:36 in real time you know build all these ai

7:38 pipelines

7:39 but interesting the other thing we found

7:41 is organizationally

7:43 you know and this is what i tell people

7:44 is you know when you see big companies

7:46 start

7:46 asking their different business users to

7:48 uh work together

7:51 that's usually a sign something changing

7:52 architecturally in the industry

7:54 you know in this case what i found is i

7:56 was able to tie the customer support

7:58 team

7:59 right up with the data science team

8:00 right and that is where

8:02 for every ticket right every customer

8:04 support ticket that comes into miss

8:06 today

8:07 we basically have marvis answering those

8:09 tickets right now and we're

8:10 at about 70 percent efficacy that means

8:13 that

8:14 you know and these are hard tickets

8:15 right these are tickets like you know

8:16 why are

8:17 why is having a connection problem why

8:19 is the zoom not working

8:21 right so right now we're at 70

8:23 championship what i call championship

8:25 level

8:25 you know where i can answer about 70 of

8:27 the questions that come in our way

8:29 um i want to get that to 90. for to be a

8:32 champion my

8:32 criteria is 90 it's like you know we get

8:35 close to 90

8:36 i feel like you know i'm playing i'm on

8:38 par with network jeopardy

8:40 well the good thing about that though is

8:42 that uh when marvis does it

8:43 and as a former network engineer i'm

8:45 saying this uh nicely

8:47 marvelous rolls eyes of the user for not

8:49 knowing what the user is doing right so

8:52 um at least you know it it doesn't uh in

8:54 a nicer way i suppose

8:56 hey um since um juniper required

8:59 yes uh i noticed there was two other

9:02 acquisitions one was

9:03 astra who's an intent-based networking

9:05 company and 128 technology

9:08 um which is an sd-wan company so explain

9:12 how

9:12 i would take missed abstract 128t with

9:15 juniper put those together

9:17 um you know tell me about that story

9:20 yeah

9:20 yeah so you know when you look like when

9:22 we started mist right

9:24 the question we were really trying to

9:25 answer is you know why are you having a

9:27 poor internet experience

9:29 uh it turned out that the access point

9:31 provided about 80 percent of the data we

9:33 needed to answer that question

9:35 and that's one reason when we started

9:37 this that i decided to build an access

9:38 point

9:39 was really to make sure i can get the

9:40 data and needed to answer that question

9:43 so since we've joined juniper our

9:45 mission is really to basically

9:47 extend marvis and ai ops across the

9:50 juniper portfolio

9:52 and we're really starting that with the

9:54 enterprise portfolio so that includes

9:56 the access points the switches

9:58 and the routers and interestingly what

10:00 2128t

10:02 brings to the party is it brings really

10:04 two things one is they have a very

10:05 unique way of doing

10:07 tunnelless uh routing so they bring a

10:09 unique way of actually creating these

10:11 tunnels back to uh

10:12 back to gateways but more interesting

10:15 with their session-based routing

10:16 is they bring more granular information

10:19 about each session

10:21 as opposed to just having visibility of

10:23 the ipsec tunnel

10:24 we actually get visibility into each

10:26 user so what ipt

10:29 128t allows us to do is now really

10:32 allows marvis to answer

10:33 more questions about applications right

10:36 in addition to connectivity

10:37 i can start answering questions about

10:39 why is your zoom not

10:40 behaving properly and it starts to let

10:43 me answer questions with more

10:44 granularity right if you're having a

10:46 connectivity problem

10:48 i can now go down into the lan interface

10:50 i can now figure out

10:51 you know is your problem due to some way

10:52 on interface you know so that is the

10:54 vision of really extending marvis across

10:57 the juniper enterprise portfolio and

11:00 then after

11:01 is really around the data center you

11:03 know that's kind of the ultimate last

11:05 piece of the puzzle right you know

11:06 somewhere you know if you're having a

11:08 problem from the client

11:10 the other end point is in the data

11:11 center somewhere that's pretty

11:13 interesting because that brings an

11:14 end-to-end

11:15 aspect to networking that we've never

11:17 really had before like historically

11:19 like is it prior to being an analyst as

11:20 a network engineer and we thought about

11:22 web engineering

11:23 wi-fi there's data centers campus

11:25 engineers

11:26 but you just have one network now since

11:28 you're able to apply that it's all just

11:30 possible

11:31 yeah and that is really the paradigm

11:33 shift this is you know in networking now

11:35 you know

11:36 in the past it's really been about

11:37 trying to manage network elements

11:39 uh but going forward the paradigm shifts

11:42 you really want to manage the indian

11:44 user experience device experience right

11:46 that is the problem that most businesses

11:47 are dealing with yeah well i've

11:49 described that sort of as a bottoms-up

11:51 approach to networking where we start

11:52 with the elements

11:53 and we try and infer user experience

11:56 right versus the other way around where

11:57 we start with experience and then

11:59 we work plot down to try and understand

12:02 how each individual component does

12:04 so that's uh that you're right that is a

12:06 kind of an interesting shift that's been

12:07 going on now

12:08 now if if um you know missed was one of

12:10 the first

12:12 maybe the first network vendor to talk

12:13 about artificial intelligence

12:15 uh since then if i go to google right

12:17 now and google network

12:19 ai i'm gonna get you know 100 different

12:22 things that come up from

12:24 you know every vendor out there some are

12:26 ai and some aren't so if i'm

12:28 watching this how do i really know that

12:31 your ai versus somebody ai you know is

12:34 better what

12:35 what's ai and what's not at here yeah

12:37 you know what i tell people about ai

12:39 ai is a concept right it's not a um it's

12:43 not

12:43 machine learning is the algorithms ai is

12:45 really the concept of doing something on

12:47 par with a human

12:48 and what you know i tell people is you

12:51 want to look at

12:52 marvis and go how close are we to doing

12:55 something on par with the human

12:56 there's a lot of different fancy math

12:58 that goes under the hood to actually

12:59 make that happen

13:00 you know to make that self-driving car

13:02 there's a lot of map that goes on to

13:03 make that car

13:04 autonomous same in the medical industry

13:06 right when you're trying to do cancer

13:07 diagnosis

13:09 so that's my analogy for people it's

13:11 like when you look at ai

13:13 you really want to answer the question

13:15 have they done something that's really

13:16 close or becoming close on par the human

13:19 in our case

13:20 you know our vision is really can we

13:22 really get marvis to be

13:24 a member of the i.t team you know can we

13:26 finally get

13:27 marvis to the point of it becomes a

13:29 trusted member

13:30 and a system of your i.t team that

13:32 actually can do something on par with

13:34 someone you would actually hire to help

13:35 you help you manage and answer questions

13:37 on your network

13:39 all right now you've um you've mentioned

13:42 marvis a couple of times can you just go

13:43 into a bit of detail

13:44 what marvis is we didn't want to define

13:47 it up front

13:48 i'm assuming that uh you know it's

13:49 planned jarvis like from iron man but uh

13:52 uh you know how what is it how does it

13:54 work does it replace an engineer does it

13:56 augment them

13:57 yeah it doesn't replace engineers i

13:59 think uh best way i would say

14:01 augments engineers and you know if you

14:03 look under the hood of marvis

14:05 what you'll find is a bunch of different

14:07 algorithms

14:09 that are there to answer different types

14:11 of questions right

14:13 you know to your point of you know we

14:14 use marvis ever for every support ticket

14:16 that comes in

14:18 you know we have different techniques

14:19 like you know if you look over my

14:21 shoulder here

14:21 things called mutual information mutual

14:24 information helps us

14:25 answer one type of question that comes

14:27 in uh we have other algorithms like lstm

14:30 to answer anomaly detection algorithms

14:33 like

14:33 are you having connectivity problems

14:35 beyond normal

14:36 in your network you know and then we

14:38 have different algorithms like graph

14:40 databases and temporal correlation

14:42 that helps us correlate user experiences

14:45 with configuration problems right

14:47 if someone misconfigures a router that

14:49 also

14:50 stops uh mtu packets from you know large

14:53 packets from coming through

14:54 that will screw up your authentication

14:57 those are very hard

14:58 problems to solve so when you look at

15:00 marvis it's really a collection of

15:01 different machine

15:02 learning algorithms that are really put

15:05 together

15:06 to really do something on par with a

15:09 i.t domain expert and then on top of

15:12 that

15:12 there's a conversational interface

15:14 because i think when we look going

15:16 forward

15:17 you know we saw the transition from kind

15:19 of the cli

15:20 you know moving to dashboards i think

15:23 the next transition we're going to see

15:24 in networking is really to these

15:26 conversational interfaces

15:28 becoming a better way to interact with

15:30 the network get data

15:31 and basically avoid you know

15:34 the swivel chair problem having swivel 3

15:37 hundreds of dashboards to get an answer

15:38 to something yeah that's uh

15:40 that that's an interesting challenge

15:41 because it's uh everyone's getting used

15:43 to this for the chair management style

15:44 where i just

15:45 look across the dashboards and correlate

15:47 the information in my head

15:48 now you've um you talked about the the

15:51 seventy percent

15:52 efficacy i'm curious to uh know you know

15:55 how

15:56 accurate is missed and marvelous right

15:58 we i think when it comes to artificial

16:00 intelligence

16:01 there's always a lot of skepticism um

16:04 even with castles and things every time

16:05 they have an accident people freak out

16:07 and things like that

16:08 so you know where was it where's it

16:09 going and where do you think you can get

16:11 to

16:12 yeah you know you know if you look at

16:14 the journey of ai ops and marcus and

16:17 where we started right now

16:18 i would say the first thing is around

16:20 data right we probably spent a good year

16:24 just making sure that we had the data

16:26 necessary

16:27 to answer customer support tickets right

16:30 and so

16:31 and this is the power of the cloud right

16:32 because you want to make sure

16:34 that when you're trying to build this

16:36 that the data you need to answer

16:38 questions is in the cloud you don't want

16:40 to have to go the device

16:41 ssh log in or do any of that so that is

16:44 the first step towards actually getting

16:46 to that

16:47 jeopardy championship level uh marvis

16:50 plane

16:51 that's where we started i would say the

16:53 second thing we started with rounds once

16:55 we got the data

16:56 in the cloud it really became efficacy

16:59 you know having marvelous answer support

17:01 tickets coming in

17:03 and that is really where we started

17:04 working our way and we probably started

17:06 10 20 percent

17:07 you know four or five years ago to a

17:09 point now we're up to 70

17:11 and interesting that's the same parallel

17:13 if you look at jeopardy how long it took

17:14 him to build something to play something

17:16 on par with the championship level

17:18 jeopardy player uh so i think we're on

17:20 par to you know i think in parallel with

17:22 that we're on part of that right now

17:24 we're at 70 percent headed to 90 percent

17:26 right now

17:27 what do you think you get to 90 uh i

17:30 i keep saying as we get closer and

17:32 closer it gets harder and harder you

17:33 know you get close to the last five

17:34 percent

17:35 so i think we're still a year or two

17:37 away from the 90

17:39 range where we will basically have

17:40 marvis answering 90 of all the tickets

17:42 coming into

17:43 coming into miss yeah it's like a golf

17:45 swing you get better fast and then all

17:47 this in your plateau right so

17:49 and by the way i will uh tip my hand to

17:52 the good folks who missed i as a former

17:53 network engineer i'd say there is

17:55 nothing

17:56 harder i believe the speaker there's

17:57 nothing harder in corporate id than

17:59 solving wi-fi

18:00 problems just because there's so many

18:03 variables it could be everything from

18:04 dhcp to radius to people not configuring

18:07 them properly it's on the client

18:09 access point so if you can do that i'm

18:12 fully confident that you can do

18:14 you know you know sd-wan is pretty hard

18:16 too but uh you know i'm pretty confident

18:18 so there's a lot of interest in

18:20 artificial intelligence and machine

18:22 learning

18:22 i i think it's been positioned as

18:26 really the next wave of where i t is

18:28 going and in fact i don't think there's

18:29 any question

18:30 that i think he's going to have a bigger

18:32 impact societally than

18:34 maybe every anything since we saw the

18:36 internet but

18:37 there is a bit of a skills gap here so

18:39 if i'm a network injury and i'm watching

18:41 this and i'm interested in ai

18:43 what do i do to get started you know i

18:46 think you know

18:46 the interesting thing we are putting a

18:48 ton of pressure on the it

18:50 teams right now right i mean you know

18:52 we're asking them to go from cli

18:54 you know becoming you know routing

18:56 switch ap

18:57 experts you know we're asking now that

19:00 for them become python programmers right

19:02 you know because we're basically moving

19:04 into this cloud api

19:06 paradigm uh which is really forcing i

19:09 you know the future i t person really

19:10 become a good programmer

19:12 um as we move to ai ops which is really

19:16 another form of automation right it's

19:18 basically another tool

19:19 now for helping them automate how they

19:21 manage these networks

19:23 we're really really asking the it teams

19:26 to start becoming somewhat data

19:27 scientists

19:28 you know we don't need them to program

19:30 them but they at least need to start to

19:32 understand the basics of

19:33 the different data science algorithms

19:35 what they're good for so those are kind

19:36 of the basic skill sets you know

19:38 and it's really happening with python

19:40 right now right most of those it teams

19:42 are just starting to wrap their arms

19:44 around programming for apis

19:46 and automating with python scripts and

19:48 such uh

19:49 aiops is kind of the next step in that

19:51 evolution

19:52 of taking fancier ai up data science

19:56 algorithms

19:57 to actually actually all help automate

19:58 some of the more complex problems

20:00 they're dealing with

20:01 yeah in fact um this is an interesting

20:03 trend i've been watching because a few

20:05 years ago if you look at what i've been

20:06 writing on you know our network world

20:09 you know sites like that it was it was

20:10 about help urging network engineers to

20:13 become programmers

20:14 but i think i've taken a step back from

20:16 that where it's not really a programmer

20:17 per se

20:18 as much it is as kind of a software

20:20 power user

20:22 and through the research i've done it's

20:23 actually been pretty remarkable that

20:25 three cores of network engineers i've

20:27 talked to have never made a single api

20:29 right so we're trying to advance our

20:32 systems but i think there is a bit of a

20:33 skills gap here so it's a

20:34 it's a good time to be a network

20:36 engineer i think the opportunities in

20:38 networking have never been greater

20:39 you just need to change your skill set a

20:41 little bit and then doing things the cli

20:43 way is probably

20:43 rapidly kind of depends yeah most

20:46 definitely

20:47 i mean i mean you're looking at right

20:48 now right i mean almost every it

20:50 person i know is trying to learn python

20:53 right they're all

20:54 the next step is how to basically uh

20:57 leverage all the apis that they're that

20:59 are becoming available

21:01 all right well bob uh you know thank you

21:03 very much for your time it was uh

21:05 great talking about ai missed wi-fi all

21:07 those things i think

21:08 you know one of the uh you know really

21:12 fascinating things about wi-fi

21:13 is it's become this foundational

21:15 technology right it used to be the

21:17 other network uh but now it's the

21:20 primary network for almost everything

21:21 in-store shopping

21:23 uh you know every hospital schools

21:25 everywhere you go if you don't have good

21:26 quality wi-fi

21:27 you don't have anything so um so you

21:30 know kudos to

21:31 the good work you've been doing over

21:32 there at juniper uh anyways i'm uh

21:35 uh thanks bob uh for joining me in the

21:37 stock this spot leadership video

21:39 i'm uh zeus caravallo from ck research

21:41 and on behalf of my partner ewik

21:43 thank you for watching

21:58 [Music]

22:01 you

Show more