Real AI: A Reality Check Beyond the Hype

0:03 [Music]

0:12 hello everybody Welcome to real AI a

0:15 reality check beyond the hype I'm Jeff

0:17 Aaron VP of Enterprise marketing at

0:19 Juniper and we have Bob Friday GPI

0:21 officer good to be here Jeff yeah

0:23 awesome this is actually a little bit

0:25 strange because Bob and I have had so

0:26 many conversations throughout the years

0:28 but usually there's a bottle of wine in

0:30 the middle for those that don't know Bob

0:31 is a Vintner and actually a pretty good

0:33 one I must say actually I'm not I'm not

0:35 just saying that yeah one one Barrel

0:37 wine a year and yeah we wine and air

0:39 both topics near to my heart so we turn

0:41 this into the wine show and we do work

0:43 that analogy into a lot of our

0:44 conversation which is good but obviously

0:46 real AI is what we're going to talk

0:47 about here and uh you know what I think

0:49 we really want to just peel back the

0:50 curtain a little bit uh you know there's

0:52 a lot of AI washing out there so what

0:54 goes into an AI system how does it

0:56 differentiate how do you build one and

0:58 obviously you're the guy who did it uh

1:01 in fact we we go back and that we worked

1:03 at airspace which was a Wireless company

1:04 bought by Cisco and then we worked at

1:07 Mist which I think some people thought

1:09 was a Wireless company and it was but it

1:11 also was an AI company right we were the

1:13 first to kind of do AI for it so I guess

1:15 maybe we start off how would you define

1:17 AI for it and AI Ops and and why in 2014

1:21 When Miss was founded was that the time

1:23 to really bring this to Market yeah you

1:25 know people ask me that and you know for

1:27 me I look at Ai and AI Ops you know it

1:30 really for me is the next step in the

1:32 evolution of automation right you know

1:34 and if you look what AI really stands

1:36 for it usually means doing something on

1:37 part of the human right and I see that's

1:40 what happening with networking right

1:41 right we're now trying to build a

1:43 solution that can actually do something

1:45 on par with a it expert yeah you know

1:48 and we were just talking about Watson

1:49 right and that was one of The

1:50 Inspirations for missing Jeopardy right

1:52 Watson Jeopardy 2011. that was like you

1:55 know if they can build something that

1:56 can play Jeopardy yeah you know we

1:58 should be able to build something they

1:59 can play networking Jeopardy so what

2:01 happened 24 like why was that the time

2:03 where all of a sudden you you said this

2:05 is something we can build into a system

2:06 you know there's a couple things that

2:08 happened in back in the 2014 time frame

2:10 I mean one was you know I was working at

2:12 a big company you know and we were

2:14 listening to customers back in the early

2:16 days of Wi-Fi when Wi-Fi went from nice

2:18 to have to a must-have but then it went

2:19 to business critical you know and we

2:21 started to hear from big companies that

2:23 before they were going to put anything

2:24 on their Network they wanted to make

2:26 sure that you know controllers stopped

2:28 crashing they wanted to you know keep up

2:30 with their digital transformation

2:31 project you know so they wanted code

2:33 released every yeah every weeks and not

2:36 years and probably most importantly they

2:39 started to want to make sure that when

2:40 they put a critical app on that thing

2:42 they want to make sure the app or the

2:43 user experience was really going to work

2:44 yeah it was a kind of a shift from

2:46 focusing on the network experience

2:47 passing traffic uptime to really

2:49 focusing on the user experience yeah you

2:51 know so we still have to keep everything

2:52 green right APS router switches all that

2:55 but beyond that it was really more

2:57 important to keep the applying to Cloud

2:59 experience up and running and I would

3:00 say the other thing if you look what

3:01 interesting happened in 2014 you know if

3:03 you look at the Google search statistics

3:05 2014 was the year that

3:08 AI really went from kind of a research

3:10 topic you know to a reality yeah it

3:13 became real and I think there was a

3:14 perfect storm of things that happened

3:16 back then yeah it was like Cloud gpus

3:18 you know we got access to low-cost

3:21 compute storage Amazon Google all that

3:24 stuff became real yeah models became

3:26 bigger you know tensorflow the tools to

3:29 build all the stuff became widely

3:30 available so if you look back 2014 was a

3:33 year a lot of AI startups you know if

3:35 you look at myths you know interestingly

3:38 we didn't really start as an AI you know

3:40 if you talk to sujay it was really more

3:42 of a vision of day two operations and

3:44 making sure we could build an

3:46 architecture Cloud that could handle a

3:48 lot of telemetry data yeah so I mean you

3:51 were and sujay were at the biggest

3:53 networking company in the planet at the

3:55 time and obviously you realized or you

3:57 thought that you couldn't get it done

3:59 there it's kind of a little bit of the

4:00 inventor's dilemma so walk through that

4:04 I mean why did you need a clean sheet of

4:05 paper to do real AI you know why

4:07 couldn't you get there from the old

4:08 architectures well I think you know back

4:10 in that time frame we had just you know

4:11 at Cisco we just acquired Meraki big

4:14 cloud company

4:15 um and you kind of look like what we're

4:17 trying to build the vision was that like

4:19 I said to build something that could

4:20 process a lot of data in real time for

4:23 day two operations

4:25 you know and fundamentally when we

4:26 looked at it it missed is actually a big

4:28 architectural bet

4:30 right we were basically betting that hey

4:31 we really needed a blanket of paper we

4:33 had to change the underlying Foundation

4:36 and I think what people don't fully

4:37 appreciate is moving things to the cloud

4:39 is not as simple as taking controls and

4:42 throwing analyzing and sticking a Docker

4:43 there's really a major software

4:45 architectural change yeah you know so

4:47 when 20 years ago when I was doing

4:49 Aerospace and putting code on Linux

4:51 boxes and shipping software on Linux

4:52 boxes when you move to This Cloud

4:54 architecture you're really moving to

4:56 microservices yeah much more redundant

5:00 much more reliable it actually was

5:01 really interesting to me to again to

5:03 build a real AI engine you guys actually

5:05 looked at like Netflix and Twitter and

5:08 Linkedin how they built a cloud I mean

5:09 that was more the inspiration for for

5:11 the Miss Cloud than than you know

5:12 additional networking I mean outside of

5:14 networking right you know if you look at

5:16 some of the big Netflix you know people

5:18 were already moving to microservices

5:20 architectures continuous integration

5:22 tests yeah you know these are

5:24 architectures where you basically were

5:26 uploading code every day every week with

5:29 reliability yeah and that was part of

5:31 that original customer you know if

5:33 you're going to do digital

5:34 transformation their mobile app that

5:37 part of the thing was basically moving

5:39 very fast yeah and their infrastructure

5:41 was still moving at the speed of

5:43 years months you know they want to be

5:45 moving up weeks yeah so that was kind of

5:47 the inspiration of yes we can basically

5:49 take a new Cloud architecture into mist

5:52 and to be honest that's really hard to

5:55 do in big companies you know if you're a

5:57 big company trying to change the

5:58 foundation of something is something

6:00 better done with a blank sheet of paper

6:02 inside of a big company yeah I can

6:04 imagine that um you mentioned data right

6:06 and I think going back to your wine

6:08 analogy right you know you can only make

6:10 good wine with with good grapes Bob's

6:12 mentioned that a thousand times and it's

6:14 applicable same thing with AI right you

6:16 can only make good AI with with good

6:18 data and I think that is one of the

6:20 things that is part of that architecture

6:22 shift right

6:23 um you know again when I walked in the

6:24 door and missed you know I heard the

6:27 notion of you know rewriting them the

6:28 control plane and collecting 150 user

6:30 States every two seconds and my mind was

6:32 kind of blowing on that and so talk us

6:34 through I mean

6:35 what kind of data in is it a quality

6:38 issue or is it a quantity issue right

6:39 when you talk when you hear some of

6:41 these other vendors out there just think

6:42 oh we have a thousand bazillion more APS

6:44 than anyone else so we're better at Ai

6:46 and I feel like that that story doesn't

6:48 really hold water so well I mean I think

6:50 you know when we started myth right I

6:51 think Suzanne we got a lot of grief on

6:53 you okay why does the industry need

6:54 another access point you know and it

6:56 wasn't because we thought we needed

6:57 another access point in the industry is

6:59 because we wanted to get the right

7:01 technology right yeah you know if you

7:02 look at what we did 20 years ago at

7:04 airspace you know we were sending data

7:05 back every minute to controllers and

7:08 doing symmetrically you know asympt

7:10 symmetric and everything you know when

7:12 we're doing this user experience thing

7:14 you know now we're sending data back at

7:16 every user State change you know when

7:18 you connect authenticate yeah and so

7:20 that was probably the other thing if

7:22 that's happening in the industry is

7:23 we're going from a networking SNP world

7:26 where you're pulling data out you know

7:28 to where these network devices are

7:30 really becoming sensors and sources of

7:32 telemetry yeah so that was the reason

7:34 why we decided you know you have to

7:36 build a cloud-friendly networking

7:39 element if you really want to get the

7:40 data you want to solve user experience

7:43 type of problems and it's jumping

7:44 forward uh you know five years-ish but

7:46 that's also why Juniper Miss came

7:48 together right you know now being able

7:50 to pull Telemetry from routers which is

7:52 security devices firewalls you know

7:54 Wireless you know I would imagine that

7:56 completes the puzzle quite a bit yeah I

7:58 mean if you look when we started we've

7:59 really focused on the access point in

8:01 the edge because we were trying to

8:02 answer the question of you know if

8:04 you're having a poor internet experience

8:06 it turns out that the edge the access

8:08 point has about 80 percent of the data

8:10 you need to answer that question yeah

8:12 you know since we've joined Juniper now

8:14 you know we've started extend that AI

8:16 Ops Marvis framework across the wireless

8:18 the switch the route you know and when

8:20 you get to the router that starts to

8:22 bring in the application layer yeah and

8:24 so that starts to let us start answering

8:25 more questions like you know why is your

8:27 team zoom call having problems yeah and

8:30 so the access point in the edge gives

8:31 you a lot of Layer Two connectivity

8:33 Telemetry the router gives you a lot of

8:35 layers three application Telemetry yeah

8:37 so I mean I actually find that

8:39 fascinating right um you know the notion

8:41 of can you do Ai and silos right you

8:43 know can you have an sd-wan solution

8:44 here a wired Wireless solution here or

8:46 security solution here if they're not

8:48 all together I mean are you really going

8:49 to be able to do end-to-end event

8:51 correlation or we call client to cloud

8:52 and that's obviously one things that's

8:54 impressed me about what we've been able

8:56 or you've been able to deliver at

8:57 Universe I think that's that's pretty

8:58 interesting yeah and I think that's a

9:00 vision and you look at Juniper right

9:01 we've extended that

9:03 Telemetry all the way from the client to

9:06 the wireless AP to switch the router you

9:08 know we're starting to extend that

9:09 Telemetry all the way into the cloud

9:11 application yeah so there's a recent

9:13 announcement with zoom for example yeah

9:15 walk through that because I thought that

9:16 was that was really interesting because

9:17 it's funny we've always been saying for

9:19 years right you know with the client to

9:22 Cloud you can troubleshoot you know

9:24 what's wrong with Bob's Zoom call but we

9:25 were missing kind of a key element right

9:27 we're focusing on the network and not

9:28 necessarily the application side and so

9:29 I think that's an interesting shift

9:31 right yeah I mean so that was really the

9:32 you know you know when you're in the AI

9:35 data Science World labeled data is like

9:37 gold okay you know if you talk to any

9:39 data science who's trying to build a

9:40 model you know if you get label data you

9:43 can train that model you know so what we

9:44 announced that Mobility field day

9:46 recently was basically starting taking

9:47 data from the application layer and the

9:50 application layer knows when something's

9:52 gone wrong they may not know what went

9:53 wrong but they know something so that's

9:55 labeled so now we have label data from

9:58 your collaboration Zoom teams call and

10:00 we're conjoining that with your network

10:02 data yeah right once you've done that

10:04 now we can build models that can

10:05 accurately predict your Zoom performance

10:08 or your team's performance you know and

10:10 once you've actually done that now you

10:12 can interpret that model yeah you know

10:14 once again actually predict your

10:15 performance then that's the power of AI

10:17 right and that's when you're that's

10:19 really AI I mean that goes beyond just

10:20 normal machine learning to really be

10:22 able to you know do anomaly detection

10:23 Predictive Analytics self-driving I mean

10:26 that's that's all that's all becoming

10:28 real so that's that's awesome yeah I

10:30 mean if you look at it you know if you

10:31 look in the history of machine learning

10:33 right there's tons of AIML algorithms

10:36 that have been around for decades you

10:38 know what's really transforming the

10:39 industry in networking outside network

10:42 is really these deep learning models

10:44 right and that's what we saw with chat

10:46 yeah last year right it's like these

10:48 models are getting much more complex and

10:50 bigger and it's really tons of data that

10:52 you're using training yeah and those are

10:54 the disruptive

10:55 models I want to talk about that more in

10:58 a second before we get there you

10:59 mentioned data science right and you

11:02 mentioned MFD um you know I was at an

11:04 MFD once where you know I said we were

11:06 the first with AI Ops and you know data

11:08 science and someone's like does the

11:09 first really matter does being first

11:11 really matter and I kind of felt when it

11:14 came to AI yeah it does right I mean

11:17 you're learning your data science

11:18 algorithms are getting better

11:20 um you know I remember when Alexa first

11:21 first launched right I said who's the

11:23 quarterback of the 49ers and the answer

11:24 was Joe Montana is a famous quarterback

11:26 of the 49ers it's like no that's not

11:28 what I asked and now if you ask it to

11:29 tell you who played last week and what

11:31 their stats were so talk about you know

11:33 the data science toolbox you know what

11:34 kind of you know went into mist and

11:36 again you know why it matters for real

11:39 Ai and how it differentiates yeah I mean

11:41 I think in the industry right now we're

11:43 all using the same underlying algorithms

11:45 you know but the really big thing around

11:47 the data science is really the team you

11:49 know if you look what happened with

11:50 openai right it didn't pop you know last

11:53 year they didn't magically pop out of

11:55 the words that was five years years of

11:56 work that went into going from TPT one

11:58 two three four you know and if you look

12:00 what you don't realize there's gpt3 that

12:02 really broke out of the gate right it's

12:03 been around for a while yeah no you look

12:05 at Mist right you know when we started

12:07 this adventure I had you know took a

12:08 year to get the cloud built you know and

12:10 then it took time to actually get the

12:12 data you know the first mission I missed

12:15 was really basically getting your

12:16 support team to stop sshing into devices

12:18 yeah right because if you're going to do

12:20 AI Ops you got to get that data to the

12:22 cloud yeah right and that by itself you

12:24 know figuring out what data to get to

12:25 the Cloud is a journey yeah and I mean

12:28 I've heard examples where as the data

12:30 science algorithms get better you do

12:32 things like less false positives you can

12:34 do things like more feature sets like

12:36 service levels or finding failing

12:39 clients and things like that so

12:40 obviously it gets more more robust and

12:42 better with time right yeah I mean if

12:45 you look at an army detection I mean

12:46 that's kind of a classic networking

12:48 problem that's been around for years

12:49 yeah uh you know we've been trying to do

12:52 it with arima and other statistical

12:54 approaches and the false positive

12:56 basically generated more noises you know

12:58 no network admin wants to be woken up at

13:00 three in the morning for a crying wolf

13:01 right yeah right you know but now we've

13:04 gone you want to call them only when

13:05 there's a problem you want to get down

13:06 that false you know positive it's got to

13:08 be almost zero yeah and that's basically

13:10 where things like lstm up in that deep

13:12 learning category yeah are starting to

13:14 really transform Network right yeah how

13:16 we can actually build an army detections

13:17 that don't cry wolf yeah and I know

13:19 there's like this Holy Grail out there

13:20 self-driving right you know kind of like

13:22 autonomous vehicles has their stages of

13:25 autonomy and you know sort of us there's

13:27 like a stage of of self-driving where

13:29 you know the network fixes itself right

13:31 do you think we'll ever get there I

13:33 think we're on the way there I mean if

13:35 you look what we're doing right now

13:36 there's all types of cases real Resource

13:38 Management it's kind of the classic one

13:39 that's been it's kind of self-driving of

13:41 adjusting the power control channels in

13:44 your network that's already happening

13:45 which by the way great story there we're

13:47 at a customer Retreat and I'm not going

13:49 to mention the customer but uh you know

13:51 there's a guy who said I will never let

13:53 you know the system automate my RRM and

13:56 we did that's where we basically you

13:57 know ran through things and he chose

13:59 what channels and and outputs and and

14:01 our system did and he was kind of Blown

14:04 Away he's like this is the first time

14:05 ever it actually mirrored what I would

14:07 do so yeah I remember that right yeah I

14:09 mean I think you know six gigahertz is

14:11 probably example guys there's too many

14:13 knobs for the average person to try to

14:14 get things adjusted right you know so

14:16 that's an example where that's

14:17 self-driving I think we're starting to

14:19 see things like vlans missing vlans

14:22 these are things that the AI can detect

14:25 much easier than the human and those

14:27 type of things are going to be starting

14:28 to get self-correcting now and you would

14:30 argue this also then ties back to the

14:31 need for you know a full stack right if

14:34 you don't have the ability to go change

14:36 something on the land or change

14:37 something on the wired Wireless you're

14:38 only going to be able to just give a

14:39 recommendation at best right and so you

14:42 know that self-driving needs that full

14:44 staff to come together yeah in the data

14:46 Science World what we call feature

14:47 engineering right and so when you're

14:49 trying to build these models you know

14:51 predicting your Zoom team's performance

14:53 you know you want features across the

14:55 whole stack right I want features from

14:56 the client the access point and the

14:59 router and all the way to the into the

15:01 data center right because ultimately

15:02 when you're tracking down problems you

15:04 want to get down to what networking

15:06 feature is causing your poor experience

15:08 yeah yeah so the step right before

15:12 self-driving is a VNA a virtual Network

15:14 assistant right like uh in the Juniper

15:16 world we have Marvis

15:19 um walk me through that a little bit I

15:20 know Marcus is your baby uh you know I

15:22 think you even probably named it Marvis

15:23 back in the day

15:27 Jarvis and all that and are we going to

15:30 be sued but that's fine

15:31 um but you know talk you through it like

15:32 what what in your mind is is a VNA and

15:35 what does it bring to the table yeah you

15:37 know I think this is really around this

15:38 conversational interface you know when

15:40 you interact with your network you know

15:42 you're either in a troubleshooting mode

15:43 and trying to figure something out over

15:45 in the self-driving mode yeah you know

15:47 so if you look at what we've done

15:49 marvelous and Ops you know we have kind

15:53 of the action framework for self-driving

15:55 but we also have a conversational

15:56 interface for troubleshooting yeah you

15:59 know and I think you know what we've

16:00 seen with large language models we

16:02 started this interface probably four or

16:04 five years ago did really good at

16:06 natural language understanding you know

16:08 but what llms are bringing is natural

16:11 generation now yeah right and I think

16:13 this is the vision of uh you know my

16:15 Star Trek analogy you know if you look

16:17 at Star Trek almost all those

16:19 technologies have become real I think

16:21 you know talking to your networking

16:22 computer is the next star track analogy

16:25 yeah technology yeah that is going to

16:27 bite the dust yeah I think we're gonna

16:28 be down to teleporters the last piece of

16:31 Star Trek that needs to be brought you

16:33 out there

16:34 um

16:35 I do want to double click on that kind

16:37 of the llm bringing in the journey

16:38 because I think you hit on something

16:39 interesting you know our perspective is

16:42 it's just it's it's another

16:43 conversational interface it's another

16:45 add-on so where do you see it providing

16:48 you know a lot of value to what we're

16:50 doing where do you see it kind of being

16:52 tangential like how do you see it kind

16:53 of living together I think you know in

16:54 the troubleshy I think it's going to

16:56 make it a lot easier for it Network

16:59 admins in the future to actually get

17:01 information from a you know a much

17:04 growing complex network no I think you

17:07 know we have to look at most Network I.T

17:08 admins they start their careers with

17:10 clis we've slowly moved them from clis

17:13 to dashboards to help and make things

17:15 easier to manage I think this next

17:17 transition is going to be moving from

17:18 these dashboards into these

17:20 conversational interfaces

17:22 um I think open AI I mean we are another

17:25 step closer to you know talking to Kim

17:28 you should be talking to your network

17:29 and asking you know what's wrong today

17:31 you know why do you have a stomach ache

17:32 so AI killed the UI star you can quote

17:35 me on that that's our next webinar so

17:37 we're gonna use that one

17:38 um

17:39 um so what do you see coming next right

17:42 I mean for starters I think llm took a

17:44 lot of folks by surprise it created a

17:45 lot of Groundswell around this industry

17:47 a lot of you know folks figuring out how

17:50 to how to do that and

17:51 um you know even then I have some

17:52 concerns of you know our competitors

17:54 will launch an Ln Solutions say this is

17:56 a VNA and and obviously it's very

17:57 different but I'm curious on your take

18:00 on that I mean I think it's clear that

18:02 with llns right we're going to see

18:03 conversational interfaces become real

18:05 yeah you know we're going to start to

18:07 see them helping actually to think I

18:09 think you know next thing is really

18:10 around troubleshooting you know and how

18:12 we really solve real-time

18:13 troubleshooting loms are not going to

18:15 solve that real-time troubleshooting

18:17 problem it's a time stamp um they're

18:18 definitely a model at a point in history

18:19 right so it's good for knowledge base

18:21 but not for real-time questions I mean

18:23 so we'll definitely see llms basically

18:24 help with the knowledge-based stuff you

18:26 know basic questions you'll see my LED

18:30 blinking or you know what is EVP and

18:32 vxlan right questions like that and I

18:34 think we'll even see llm start to help

18:36 the uh network data in databases you

18:39 know where we start to build text to SQL

18:42 translators oh interesting no making it

18:44 much easier to get data out of the

18:46 database so I think that we're going to

18:47 see happening uh the real-time

18:49 troubleshooting is still going to

18:51 require domain expertise and data

18:54 scientists nice so to bring it all

18:56 together you're still obviously need the

18:58 right software architecture need the

18:59 right data need the right data science

19:01 you know you can't just overnight say

19:03 hey we're going to work with open Ai and

19:05 and and and deliver everything you need

19:07 in a networking solution no I mean I

19:09 think you know the journey to building

19:10 AI Ops you know starts with like I said

19:12 this Cloud Foundation but I think the

19:15 other piece people don't fully

19:16 appreciate you know there's the

19:17 technology piece and there's also this

19:19 organizational piece yeah right and I

19:22 think that's one of the things I found

19:23 you know another reason why I we left

19:26 Cisco to start missed was Oregon State

19:28 it's hard to do that in a big company

19:30 it's hard to change organizations inside

19:32 a big companies and I think where you're

19:34 referring to that is like having the

19:35 data science team sit with you know the

19:37 the support team you know uh and the

19:39 engineering team right so they're all

19:41 kind of looking at the problems coming

19:42 in feeding it back into the system

19:43 getting better I think that's that's

19:45 what we're alluding to in terms of

19:45 organization yeah I mean I think I

19:47 always highlight the people it's like

19:48 once we move to these Cloud Solutions

19:50 your support team has really becomes a

19:52 proxy for your customer right I mean

19:54 once I get the data to the cloud your

19:57 support team is the ultimate customer

19:58 you know if you make your customer

20:00 support team happy you know the fewer

20:02 tickets they see that means the fewer

20:04 tickets your customers are generating

20:05 and and we've I think every event we've

20:07 been at we've shown that Marvis efficacy

20:09 slide why don't you describe what that

20:12 is what are what are our goal is on that

20:13 for those that probably haven't seen it

20:15 yeah I mean I think that's the other

20:16 part of the journey it's like hey you

20:17 know if you're going to go down the

20:18 eiops you've got to eat your own dog

20:20 food you know so since we make your own

20:22 champagne Bob drink your own wine

20:25 so anyway I think that is the other big

20:27 aspect of this is hey if you get your

20:29 support team to actually use your AI Ops

20:31 you know and that's where I said it took

20:34 a year to figure out why I had my

20:36 support team SSA 18 into these devices

20:38 right because if I'm doing Cloud AI I've

20:41 got to get that data back to the cloud

20:42 and and so we would collect charts on

20:44 you know is the answer in the AP is the

20:46 answer in the VNA is the answer in the

20:48 cloud do we not have the answer at all

20:50 yeah we'll walk through that because I

20:52 think that's really a very foundational

20:53 element on how you build real AI right

20:55 to get to the root of can you solve the

20:57 problem yeah so that was an example

20:58 where the poor team actually used Marvis

21:00 to try to answer every support ticket

21:02 coming in and the data science team

21:04 reviewed those tickets with them every

21:06 week to figure out if Marvis didn't get

21:08 the answer why didn't we get the answer

21:10 is it because the data is still an AP

21:11 and so I was basically making sure we

21:13 can get the data from the EP that was a

21:15 year-long you know and that's part of

21:16 the it takes time to actually build

21:18 these AI Solutions because you have to

21:20 get the data you know and then you have

21:22 to figure out what features you actually

21:23 need and what I found interesting is you

21:25 know during that time you know the

21:26 number of tickets didn't necessarily go

21:27 up um so things were obviously getting

21:29 better but

21:31 um it yeah yeah it's just um there's

21:34 just it's very interesting in terms of

21:35 the data you can pull from that and and

21:37 like you said it just takes time it

21:39 takes time to build that out and what's

21:41 what I find interesting is that now it's

21:43 like the easy problems have been

21:45 answered so now it's the hard problems

21:46 that you're kind of focusing the

21:47 problems that are worth waking someone

21:48 up for right well I think that's part of

21:51 the journey is you start to solve the

21:52 low-hanging fruit problems right you can

21:54 see the graph go up quicker but as you

21:55 start to get towards parallels because

21:57 that last 10 gets harder and harder and

21:59 I think that is where if you look in the

22:01 data science

22:02 it's hard to hire data scientists who

22:04 are really networking experts yeah and

22:06 so that's why you've got to get you know

22:07 get your data science team but your

22:09 support team is the domain expert your

22:10 support team knows how these networks

22:12 work yeah and that's why you got to get

22:14 the right that's marriage between

22:15 because I know I've seen you in accounts

22:17 where you said you want to know if

22:18 someone's using real AI ask them how

22:21 their support team interfaces with their

22:22 data science yeah I my take right now is

22:25 you know Miss is the only ones actually

22:27 using their own AI Ops in their support

22:29 team yeah you know if you go to a vendor

22:32 and ask him it's like

22:33 what is your support team using the

22:35 answer support tickets if they're not

22:36 using their own AI Ops solution they're

22:38 not on the journey yet yeah

22:40 um one other thing I want to talk to you

22:42 a little bit um kind of what's next that

22:44 we're starting to see here even at AI in

22:45 action is

22:47 um integration like between AI systems

22:49 like you know servicenow for examples

22:51 here we've done a lot of integration

22:52 with them you know where do you see that

22:53 going and what what excites you about

22:55 that yeah I know this is an interesting

22:57 one because I think what we're seeing in

22:58 the networking World

23:00 um in the past we all used to send

23:01 networking events up to the systems

23:03 above us you know solarwinds some other

23:05 it's Splunk or something we're starting

23:07 to see networking move to AI events

23:11 right we're seeing a bunch of

23:12 distributed AI systems is that

23:13 standardized at all or I think we're at

23:15 the very beginning of telemetry I mean

23:17 we've seen standards go on to try to how

23:19 to configure network devices Telemetry

23:21 standards are just starting of you know

23:23 how do we make sure we start

23:24 standardizing what data is going to be

23:26 sent back from these networking elements

23:28 but what we're seeing what happened in

23:30 the AI stuff is we all start to filter

23:32 these raw Network events through AI

23:34 events we're seeing a much more

23:36 intelligent events above us and I think

23:38 that's what the industry is dealing with

23:39 is you know how do you deal with a bunch

23:40 of distributed AI events AI systems

23:43 across across the network yeah and I

23:45 know for for folks that are in Vegas

23:47 servicenow is going to talk about it but

23:48 for folks that are live on stream why

23:50 don't you just give a quick snippet on

23:51 what we're doing with servicenow because

23:52 I think that's really interesting

23:54 I mean if you look at servicenow they're

23:57 becoming kind of that consumer of all

23:59 these AI events yeah right and so

24:01 they're becoming kind of the next I

24:03 always call a generation of

24:05 AI above us right you know as we send AI

24:08 events they're taking AI events across

24:10 different systems and now it's starting

24:11 to correlate those events together to

24:13 help solve even bigger problems across

24:15 the network got it pretty cool

24:18 closing out a little bit any advice

24:22 um you know any advice for someone going

24:23 down this journey starting out saying

24:25 you know I want to build real Ai and not

24:28 go down a path a dead-end path you know

24:30 what would you give them well I mean I

24:32 think at a high level I always you know

24:33 people are starting the journey whether

24:35 startup or new adventures

24:38 this is the ultimate team sport you know

24:40 if you're going to start a you know a

24:41 New Journey whether it's a startup or an

24:43 adventure you got to realize that it

24:45 takes a team yeah you know and it takes

24:47 a team of data scientists support people

24:50 and organization sales marketing I don't

24:53 think people fully appreciate it

24:55 or or even we're not going to go down

24:57 here because that's a whole other

24:58 session of

24:59 legal ethics uh you know obviously AI

25:03 touches on a lot of points now that I

25:04 think are pretty interesting right yeah

25:06 I mean I think you know when you go down

25:07 that path you know there will be

25:09 compliance you know if you're headed

25:10 down the AI path you have to be aware of

25:12 the new compliance laws

25:14 um ethics you have to be aware of you

25:16 know how you're using AI that means

25:18 those are all aspects of AI that you

25:20 have to actually take into account when

25:21 you start to start your journey very

25:23 cool so I would imagine at some point we

25:25 won't be having this conversation it's

25:26 just be the the chat is doing it I'm not

25:30 even sure we're having this conversation

25:31 yeah I'm sure we are it's a simulation

25:32 everybody

25:34 um so with that I want to thank you Bob

25:35 I enjoyed the conversation

25:37 um and thank you guys all for joining us

25:39 if there's anyone that wants more

25:40 information you can check out the

25:41 comments and also there's a chance to

25:43 win a free AP

25:45 um I always feel like test driving the

25:46 product is not a free bottle of wine not

25:48 everybody

25:51 usually when we're in a small setting

25:53 says is that if anyone's in Los Gatos

25:54 they can come to his house for wine I

25:56 don't know if you're prepared to do this

25:57 over here or not but give me yeah you're

25:59 in the Alaska let me know Mississippi

26:01 State wine tasting room is open the

26:03 Vista State wine tasting room I thought

26:04 it was Friday Estates the name changes

26:06 all the time now it's a day of the week

26:08 depends on the day of the week and how

26:09 drunk you are

26:10 um so again with that please check out

26:12 the link and thanks everyone for joining

26:13 us we really enjoyed it thank you Bob

26:15 thanks dude

26:19 [Applause]

26:21[Music]