Bob Friday of Juniper Network on the AI w Maribel Lopez Podcast

0:03 hello and welcome back to the podcast

0:06 I'm marabel Lopez and this is the AI

0:08 with Marbel Lopez podcast and today we

0:11 are joined here with Bob Friday who is

0:14 the chief AI officer and CTO of the

0:17 Enterprise for Juniper Networks hey Bob

0:19 how are you y doing great marille great

0:22 to be here and as I tell people I never

0:24 believe I end my career as a chiefa

0:26 officer but here we are you know Bob

0:29 I've seen you through a lot lot of

0:30 different iterations of the career and

0:31 it's always been something new and

0:33 exciting and it actually doesn't

0:34 surprise me at all that you're in AI

0:36 that you would have found that and that

0:37 it would have found you and that

0:39 probably leads us to our first question

0:42 and that's you know in in my opinion

0:44 you've been at the Forefront of working

0:46 with AI in the networking field for some

0:49 time uh and that used to be an AI always

0:52 spoke about was chips and everything was

0:53 about gpus and CPUs that maybe could act

0:56 like gpus but I thought maybe you could

0:58 spend a moment discussing in what you

1:00 believe has changed in the AI landscape

1:02 since you've been in it from a

1:04 networking standpoint maybe tell us for

1:07 example um what we couldn't do several

1:09 years ago that we can now do today or

1:11 how you see that whole thing

1:13 evolving yeah know you know for me

1:15 personally right you know I started my

1:16 career in the wireless space you know

1:18 and this whole Adventure really started

1:20 back at the during the Cisco Adventure

1:22 we were talking to some big customers

1:24 about putting some customer experiencing

1:27 on their Network and they basically told

1:29 sud he's like we're not putting anything

1:31 on this network until you can promise US

1:33 you know that things are not going to

1:35 crash you can give code to us faster and

1:37 more importantly you know you can

1:39 basically guarantee that the user is

1:40 going to have a great experience you

1:42 know and that's kind of where the change

1:44 started that paradigm shift from

1:46 managing Network elements to really

1:48 managing this in toin user experience

1:50 and that's where AI Ops actually got to

1:52 the thingss you know and what I've seen

1:54 has changed you know the 30 some years

1:56 I've been doing this um you know 30

1:58 years ago I did my Master's and actually

2:00 did it with the neural networks and you

2:03 know I you know so what I've seen

2:05 personally change as these neural

2:06 networks have gotten really big now that

2:09 they could do something interesting

2:10 right and and I think that's what we're

2:13 seeing with AIA you know it's really

2:14 this the next step in the evolution of

2:17 automation right and that is what is

2:19 changing and if you really look at the

2:20 Google data this interesting data point

2:22 you know when did this AI thing take off

2:25 if you look at Google it took off in

2:26 around

2:27 2014 that's when all the MLS searches

2:30 and if you look back in 2014 what really

2:32 came together was a combination of Open

2:36 Source right we saw all the libraries

2:38 come out that let us start to build

2:40 interesting AI things models got big

2:43 enough right that's kind of when all the

2:46 cloud got big enough where he had enough

2:48 compute and storage you know 30 years

2:51 ago I was shipping software on a Linux

2:53 Box

2:55 Bar yeah you know now I've got access to

2:58 AWS Google

3:00 you know there is no limit to Computing

3:02 the storage now it's just a a matter of

3:04 how much your how much your Amazon bill

3:05 is going to be at month so I would say

3:07 that is the big change I've seen over my

3:09 30 years is things have come together

3:12 now where you can actually build things

3:13 that we talked about 30 years ago thank

3:16 goodness and and I agree with you and

3:18 and cloud has been a big part of that so

3:21 what we see happening now is there's a

3:22 great deal of AI Marketing in the tech

3:24 space I mean it's everywhere basically

3:26 everyone is rolling out an AI enabled

3:28 product and it seems hard for Enterprise

3:32 buyers to understand and evaluate the

3:34 various Solutions because everybody says

3:36 they're doing AI so you've been in this

3:39 for a while and you have some

3:40 perspectives of what may be AI what may

3:42 not be AI but just from your perspective

3:45 what are some examples of functionality

3:47 or criteria that companies should be

3:49 looking for an AI

3:51 Solutions yeah so so this is another

3:53 funny story remember when I you know

3:55 when I talked to analysts like you they

3:56 were like hey Bob I we canot tell the

3:58 difference between your PowerPoint and

4:00 your competitor's PowerPoint from a

4:02 PowerPoint perspective it all looks the

4:04 same now I could tell you one of the

4:07 reasons and people ask me say you know

4:09 why did we decide you know why did suj

4:11 and I decide to build an access point

4:13 when we started Miss right it's not

4:16 because I thought the world needed

4:17 another wireless access point uh it's

4:21 because I really wanted to make sure I

4:22 could get the data I needed to answer

4:24 the question of why are you having a bad

4:28 internet experience

4:30 and so if you look at back to your other

4:32 question what's the difference between

4:33 now and you know 30 years ago 20 years

4:36 ago when I did

4:37 airspace one of the differen is the data

4:39 we're sending back right you know 20

4:41 years ago I was sending back data every

4:44 minute or two minutes back to a

4:46 controller synchronously you know one of

4:49 the Paradigm shifts here I missed now

4:50 I'm now sing data back on the user

4:53 asynchronous you know we're keeping

4:54 track of every user minute on the

4:56 network not the network element but the

4:58 user minute so I think that is one thing

5:01 when I talk to people about data making

5:03 sure you know does the vendor you have

5:05 the right data you need to answer the

5:07 question you're trying to answer I think

5:09 the other thing I've learned you know

5:11 and leaving Cisco to do miss was I knew

5:14 I had to build this real time Cloud

5:18 architecture you know that forced me to

5:20 you know okay I got to get a blank sheet

5:21 of paper uh but the other thing was

5:24 really about I had to organizationally

5:26 change making sure my data science team

5:28 could work with my customer support team

5:31 because when you move to This Cloud AI

5:34 model really the support team is a proxy

5:37 for your customers you know and what I

5:39 tell most customers of people you know

5:42 you should ask your vendor are they

5:43 actually using their own AI op solution

5:45 for their support team if they're not

5:48 they have not started the journey to

5:50 Cloud aops you know so the first step on

5:53 starting that journey is trying to make

5:55 your own support team happy because the

5:58 fewer support tickets that they see is

6:00 the fewer support tickets your customers

6:01 are sending you so that is the first

6:04 step on the journey to Cloud a

6:06 offs so let's just stop there for a

6:08 second because some people that are

6:11 listening to the podcast may have

6:12 varying levels of familiarity with the

6:15 term AI Ops could you define the term AI

6:18 Ops for the

6:19 audience yeah I mean so when I say AI

6:21 Ops usually what I talked about and I

6:24 agree with you people are confused

6:25 because really this is just the next

6:27 step in what they've been doing for 30

6:29 years right you know we've been

6:30 automating networks doing all a lot of

6:33 machine learning regression algorithms

6:35 we've been doing that for a long time

6:38 the fundamental difference that what

6:39 we're doing now is we're starting to

6:41 build solutions that have the cogon

6:44 reasoning skills of a human right we're

6:47 building solutions that can deploy and

6:50 operate networks on par with human it

6:52 domain experts and so the subtle

6:55 difference between the automation we're

6:57 doing now with deep learning versus

6:59 machine machine learning and when I say

7:01 deep learning these are actual models

7:03 that we train with tons of data this is

7:05 like that chat GPT thing right this is

7:09 you know where we're taking zoom and

7:10 teams data and building models that can

7:12 actually predict your user experience so

7:15 I think that is the subtle difference

7:16 when people try to understand the

7:17 difference

7:19 between ml machine learning and the

7:22 difference between these new deep

7:24 learning models and so when I say AI Ops

7:27 we're usually talking about some sort of

7:28 Deep where we're continuously learning

7:31 something about the network and these

7:33 tools that the IT department is going to

7:35 use they're almost like hiring a person

7:37 right you know you're bringing on this

7:39 virtual AI

7:41 assistant that doesn't do the same thing

7:43 every day in the old times we built

7:45 things that were very deterministic you

7:47 buil a model and it did it now we're

7:49 bringing on these AI assistants that

7:52 feel more like an a new intern you know

7:55 someone you have to learn to trust you

7:57 have to learn what they can and cannot

7:58 do and you got to trust they're going to

8:00 get better at what they're doing and

8:02 that's that continuous learning you know

8:04 that this AI assistant model is going to

8:06 learn your network and get better at

8:08 what it's doing well what I loved what

8:10 you just spoke about as well is also the

8:12 fact that it's the network connecting to

8:14 the app it's a full you know OSI stack

8:17 experience right so we're starting to

8:19 see that whole Loop come together which

8:22 I think the app was very disc connected

8:23 from that before so that's another good

8:25 piece of insight so one of the things

8:28 that we're seeing in the marketplace is

8:30 that companies are looking to have

8:32 strategic vendors to build platforms

8:34 that run the business so we have this uh

8:36 sort of pendulum where we go back and

8:38 forth between whether or not we want

8:39 lots of best breed whether or not we

8:41 want platforms it seems like we're in a

8:43 platform phase and I know that Juniper

8:47 Networks has specifically talked about

8:49 building something called an AI native

8:52 networking platform uh that's a lot of

8:55 words together AI native networking

8:56 platform but can you tell us what that

8:58 is and and what it means for

9:01 customers yeah you know the funny thing

9:03 is you know like when sui and I were at

9:05 Cisco you know we're trying to make this

9:06 decision right you know should we stay

9:08 you know do this at Cisco go off and

9:11 start a company and so when you say AI

9:14 native it's really like saying you would

9:16 need to start with a blank sheet of

9:17 paper right and what people don't fully

9:20 appreciate is you know if you're going

9:23 to build an AI native Foundation what we

9:26 realized is we really needed a blank

9:28 sheet of paper to build a new real-time

9:31 architecture right I knew I had to build

9:34 an architecture that could ingest data

9:36 real time and do something with that

9:38 data right that turns out and it also a

9:41 different software architecture right

9:44 you know when you go from kind of

9:46 building software on a Linux box to

9:49 building microservices in the cloud that

9:52 that by itself brings a ton of value to

9:54 customers right just getting data to the

9:56 cloud brings a lot more visibility and

9:58 observability

9:59 but it also brings a lot of

10:01 reliability right that's what's the

10:03 difference between a cloud environment

10:05 and a controller environment is that

10:07 software architecture those blast radius

10:09 get smarter so my location crashes it

10:12 doesn't affect my controller wireless

10:14 piece of the puzzle right so that is

10:17 kind of the thing I T think is different

10:19 when people say AI native um and I think

10:22 you actually look what's going on in the

10:24 marketplace right now I think it kind of

10:26 reflects where we are right now you know

10:29 how did Miss become a leader it was

10:31 basically that bet that there really was

10:34 an architectural change happening in the

10:35 industry right moving to Cloud AI Ops is

10:38 not something you're just going to pull

10:40 onto an existing framework you really

10:42 need to go down to that native

10:43 foundation and start from

10:48 scratch you know I agree that this is

10:50 something that's so important it's

10:52 almost like we're at the opportunity

10:54 with so many parts of this text stack to

10:56 just hit a refresh button and say okay

10:58 let's if we were going to do it from

11:00 today onward what would it look like and

11:02 I think that to me is what AI native

11:05 means just like Cloud native was sort of

11:07 the iteration before this you know now

11:08 we've got AI native so uh Insight

11:12 automation virtual assistance

11:14 conversational interfaces these are all

11:16 really hot topics in the AI space and

11:19 you have a product in the space called

11:21 Marvis um I love that name by the

11:24 way it's such a cool name uh can you

11:27 explain to the audience what Marvis is

11:29 is and how it's evolved since its

11:31 introduction yeah you know so Marvis is

11:34 basically a culation cumulation of all

11:38 the different attributes we've been

11:39 talking about here right you know so

11:42 conversational interface is one of those

11:44 attributes of next assistant and what I

11:46 tell people is if you look back over the

11:49 last 20 years we kind of moveed from CIS

11:53 to dashboards the next really next user

11:56 interface is really going to be these

11:58 natural langu anguage user interfaces to

12:01 really help people interact not just

12:03 with networking I think you're going to

12:04 see it across all the industry right

12:06 this is going to become the next big

12:08 thing and that's what chat gbt really

12:10 brought we always had kind of the

12:12 natural understanding piece what we

12:14 didn't have is a natural generation

12:16 piece right and that is what chat TPT is

12:19 really bringing that natural language

12:20 chat TPT that we actually can start

12:22 interacting you know with people and it

12:24 teams in a very more very much more

12:26 natural way I think the other thing

12:29 attribut to Marvis is this continuous

12:32 learning right and this is that piece of

12:34 you know hey we're moving from machine

12:35 learning and moving to this deep

12:38 learning automation where you're

12:40 continuously learning a network right

12:43 you're continuously building models

12:44 you're continuously adjusting Zoom data

12:47 teams data right so as the network

12:49 changes or something changes the models

12:51 are continuously adapting to what those

12:53 changes are you know so that's kind of

12:56 the Contin conversational inter phase

12:58 continu and finally there is the action

13:00 frame this action framework at the end

13:02 of the day you want to take all these

13:04 insights and turn it into some

13:05 recommendation or some action and I

13:07 think that's the industry change we're

13:09 seeing also is we're moving from that

13:11 SNMP Raw event world where we used to

13:14 send up thousands of events up to

13:17 something above us to where these

13:19 systems are now sending up more AI

13:21 relevant events right we're not setting

13:23 up raw events we're taking these raw

13:25 events from the network and actually

13:26 translating them into something that is

13:28 actionable so I think those are the

13:30 three main attributes of Marvis going

13:32 forward think conversational interface

13:34 think Marvis action framework and

13:37 continuous learning that is the new

13:39 piece that's disrupting everything okay

13:42 so in a way we can think of this is the

13:44 evolution of what we originally talked

13:47 about as self-driving

13:49 networks yeah I mean if you think about

13:51 self we're on that Journey right now and

13:54 the analysis I give to people you know

13:56 when it comes to Ai and uh I don't know

14:00 I suspect you're not a skeptic right I

14:01 would tell you the number of AI Skeptics

14:03 has gone down but I think what people

14:05 are seeing now you look at the healthc

14:07 care right you know I don't think I

14:10 think anybody who goes to their doctor

14:12 they're going to want to know that that

14:13 doctor is using the latest and greatest

14:16 AI for helping diagnose their disease

14:20 cancer right and we're seeing it with

14:22 cars right say we may not be a

14:24 self-driving cars but you're going to

14:25 want to know that that car has the

14:27 latest and greatest AI if you're going

14:29 to buy a car for your kid who's learning

14:31 Drive you're going to want to make sure

14:32 it has all the latest and greatest

14:34 safety so I think the same thing is

14:35 happened in a networking industry you

14:37 know when you connect to a network

14:39 you're going to want to know that it's

14:40 using the latest and greatest AI to make

14:42 sure that we have a great Zoom internet

14:45 experience makes total sense and like we

14:49 said earlier you know connecting it up

14:50 to the app and what you're trying to do

14:51 so we we've been talking about um

14:54 experience I know you and I talked about

14:56 experienced person atw working a while

14:58 ago and it seems like AI actually helps

15:01 us get one click deeper into that so so

15:04 we have Marvis now we've got Marva minis

15:06 I think that's actually quite cool too

15:08 so so so um when people hear marvous

15:12 minis you were just outling U outlining

15:15 the Marvis framework tell me where

15:16 marvus minis fits into that yeah so

15:19 Marvis we started I mean Marvis does a

15:21 great job when you have users on the

15:23 network trying to understand if they're

15:24 having a good experience or not right

15:27 you know and when I you know for those

15:29 who know me I make a barrel of wine

15:30 every year and it's like you know great

15:32 wine starts with great grapes great AI

15:34 starts with great data so the question

15:36 is what do you do when there's no one on

15:37 your network you know you need data

15:41 still have data to do any great AI so

15:43 Marvis minis is our answer to the

15:45 problem of day one networking how do we

15:48 make sure that network is going to work

15:50 before we turn it on how do we make sure

15:52 that network is ready to go at 8:00 in

15:54 the morning when all the users are going

15:55 to come on it so you think of minis as a

15:57 synthetic user user in the network that

16:00 Marvis can send around and gather up the

16:02 data needs to make sure that you're

16:04 going to have a great internet

16:05 connection on day one 8 o'clock whenever

16:08 you're ready to get on that Network so

16:10 that's Marva minis coming coming to the

16:13 market you know it's interesting because

16:15 um there's particularly as we think

16:17 about the AI landscape in general

16:20 there's been a lot of talk about digital

16:21 Twins and this is sort of the concept of

16:23 digital twinning for the network you

16:24 know trying to simulate something see

16:26 what happens I also like theide because

16:29 I think there's a great opportunity to

16:33 take synthetic data and test all

16:34 different kinds of things which is

16:36 basically exactly what you you said um

16:38 Marvis minis is going to do you help

16:40 people simulate or have a have a user

16:43 that simulates what the experience might

16:44 look like on a network before the

16:46 network goes live and you know we see

16:48 lots of opportunities in AI to have

16:51 either this type of experience or if you

16:53 don't have the data you need to actually

16:55 create synthetic data that you can run

16:56 with models to try to create the type of

16:58 model that you want to do so pretty

17:00 exciting work you guys are doing and I

17:02 think it's going to be really

17:03 interesting for uh people as they think

17:05 of Next Generation networks right

17:07 because it's it's a different animal

17:09 than what we've had in the past now we

17:12 say that the market is changing very

17:13 rapidly I know that you see the markets

17:15 changing very rapidly uh maybe you could

17:17 give us a bit of a look ahead you know

17:19 what are some of the things that Juniper

17:21 is looking forward to maybe it's topics

17:24 like AI Ops or other key topics you'd

17:27 like to address

17:29 yeah I mean back to the three attributes

17:30 of Marvis I think you're going see

17:32 conversational interfaces become that

17:35 standard interface going forward you

17:37 know I think it's a slow progression as

17:40 you know like it's hard to move people

17:42 off CLI we've been trying to get people

17:43 off CIS for 20 years so that's anline

17:46 interface will never die folks never die

17:49 but I think we're GNA see you know

17:51 natural language interfaces becoming

17:53 used for both public documents you know

17:56 trying to basically find out what's

17:57 going on you're going see become a u

18:00 even an alternative for business

18:02 intelligence right for exploring your

18:04 network data you know and you're going

18:06 to start to see that natur that

18:07 conversational interface become part of

18:09 your realtime troubleshooting experience

18:12 uh so you're going to see that evolve I

18:14 think what we're going to see evolve

18:15 even faster is this continuous learning

18:18 you know we saw what happened with chat

18:20 GPT when these models got big right you

18:23 kind of saw you know he we start to get

18:25 bigger and bigger models more data

18:26 training uh we we're going to start to

18:28 see that continuous learning part in

18:30 networking get more powerful just like

18:33 we saw with chat gbt and that all

18:35 results into more self-driving action

18:38 Frameworks networks that going forward

18:40 so that's where I see us heading

18:42 networking is that next step in the

18:44 evolution like I said before automation

18:46 right we're just gonna we're taking

18:47 automation to the next

18:50 level so I I love the way that you've

18:52 kind of Capstone some of the changes in

18:55 AI but if there's like one or two things

18:57 that you'd like to to close with it

18:58 you'd like to leave the audience

19:00 thinking about with AI what would it be

19:02 yeah you know when I try to summarize

19:04 more of us I usually try to remind them

19:05 you know like I said with the wine you

19:07 need to have the right data you know

19:09 that leads to the right response

19:12 recommendation and on top of that you

19:14 have to have an AI native infrastructure

19:16 secure

19:18 infrastructure makes perfect sense to me

19:20 we call those uh Righttime experiences

19:22 right information right person right

19:24 time on the right device so very similar

19:27 concept

19:28 now we usually like to leave the

19:30 audience with a new learning opportunity

19:32 and I know that you do a lot of learning

19:34 yourself as do I um is there a book a

19:36 podcast an activity that you'd like to

19:39 recommend to the audience could be Tech

19:41 related or not yeah it's you know so of

19:44 late I've been working on LM and uh I

19:46 met someone named alar antech right and

19:49 he had been working doing great work on

19:51 llm um and for networking type of things

19:54 and he's actually written a book called

19:56 machine learning for network and Cloud

19:58 engineers get ready for the era of

20:00 network automation so that's on my

20:02 latest reading list right now uh and it

20:04 kind of plays into that theme of next

20:06 generation of automation right awesome

20:11 well Bob thank you for your time and

20:12 attention and we always look forward to

20:14 hearing what you're building and what

20:15 you'll build next if anybody wants to

20:18 connect to you where can they reach out

20:21 yeah well you know AI op this is my

20:23 latest topic dear to my heart I would

20:25 just reach out to me on LinkedIn always

20:26 have to chop chat about AI Ops awesome

20:30 Bob Friday Juniper Networks Marbel Lopez

20:33 Lopez research thanks for listening

20:35 until next

20:38 time