#267 Nabil Bukhari: Exploring the Future of AI-Powered Enterprise Networking with Extreme Networks

0:00We think of AI not as a product, but we think about AI as a capability of the platform. You can't really compare a consumer-grade AI agent to an enterprise-grade AI agent. They're very different. This is what I was talking about, the machine-to-machine. Yeah. But the goal there is, before anybody gets too scared of it, is that for us, we are developing it in such a way that the concept of human in the loop is there, the concept of human over the loop is there, and the concept of human above the loop. The ability to audit AI and the ability to explain AI, that is absolutely critical. And I think the systems that are built without it are most probably not going to be successful in an enterprise environment.

0:46The value of AI increases exponentially as it has access to more data. But the risk of AI also increases exponentially as it has access to more data. So, Nabil, great to have you back. Yeah, absolutely. Well, Craig, as always, it is great to be here. Thank you so much for having me on the show. My name is Nabil Bukhari, and you can see that on the screen, but it's good to repeat that. So I work for Xtreme Networks and I play multiple different roles at Xtreme. So I'm the chief product officer, I'm the chief technology officer, and I'm also the general manager of all of our subscription business.

1:29But the role, the way I describe it, I am the person who likes to just get shit done. So whatever needs to happen, you know, I am there. So a little bit about Xtreme. You can think of us as we are probably the biggest pure play networking and security company for the enterprises. What does that actually mean? That means enterprises today at the core of their business is the requirement for secure connectivity. They need to connect their people. They need to connect their devices all across their enterprises. They need to connect their data. They need to connect their applications. They need to connect to their customers and their partners and stuff.

2:10Modern businesses are built around pervasive secure connectivity. And in order to provide that secure connectivity for all of these elements that I talked about, There is a whole set of technology that goes into it from hardware like switches and APs and SD-WAN devices and stuff and a whole slew of protocols and operating systems and then a very massive and ever-expanding management system. So all of them combined together, pretty complex technology that provides that connectivity on which the modern business run. And quite frankly, Craig, it's not just the modern business that runs on that. i think the modern society runs on it because even in our daily lives imagine a world where we can't connect through internet to anything and we can't imagine that world so there it is we are extreme um and we connect when it comes to enterprises people application devices and data no matter where it is and we do it securely yeah um and you were saying how in daily life I mean, we're all connected.

3:16Do you do anything with the public internet or is this all on closed networks? So we typically work on enterprises and then some of our areas are kind of public facing. For example, we're very big in the stadium and entertainment business. So if you go out to any of the stadiums, be it the baseball stadiums or the football stadiums or games or NASCAR, or if you're on the other side of the Atlantic and if you're going to a Liverpool game or a Manchester United game or you're in Germany and you're going to an Olympic stadium to see something, then when you're in there, when you connect to the Wi-Fi, that technology is provided by us.

4:03so so a lot similarly if you go to gosh i mean any of the big airports in the world if you go to those run on our technology some of the biggest airlines they run on our technology so public does connect to our technology a lot but they're always somebody else in the middle we don't really go directly to the consumer ourselves but a lot of the big companies or environments that you and I go to as just general citizens, they run on our technology. Yeah. And you provide both the hardware and the software and the cloud, right? Correct. So full stack solution, all the way from hardware to the software that runs on that hardware.

4:48And then most importantly, these days, the most important thing is the management and the monitoring and the troubleshooting of all of these environments, which is all based out of cloud at this point in time so those are the three layers of the stack the hardware the software and then the cloud software component yeah and presumably this uh you've been growing for a while but at some point it's would seem like everyone's got their solution uh i mean how much more of the world needs to be networked or how much more of the yeah so that's a great question You know, like people like in our environment where we sit, we would think so that, hey, by this time, pretty much everybody must be connected.

5:33But the reality is that's not true. You know, the number of there's still the number of people getting connected is still increasing. And then when you think about from an enterprise point of view, the amount of devices that an enterprise has, that is just continuously increasing because devices are not just your and my laptops and cell phones. Think about IoT devices as well. So everything is now connected. Like you go to an airport and the LEDs that are in the ceiling, even they are connected to the internet. We're getting to a world where more and more and more devices are connecting to the network.

6:12And that number is just growing exponentially. And it's not showing any signs of slowing down. And the last part is the data. There is more and more data being produced. and there's more and more data being connected. And then especially with the advent of things like AI and stuff, now more data needs to be connected for AI to do something interesting on top of it. So that requirement to connect data and connecting that securely and with the permissions that you want, that is just getting started. I mean, we've been talking about data in the enterprise sense for like 15 years, but it feels like that was child's play compared to what is ahead of us.

6:55So I think the requirement for this connectivity across people, data and devices, there is no sign of slowness in it, not for another 10 years. Yeah, and when an existing network starts adding sensors or IoT devices or new interfaces for employees. I mean, every employee that comes to my house now has a device. I always ask them who made it, and they never seem to know. But, you know, where they're entering information and it's going somewhere. How does Xtreme, when you build a network, where does that data flow? Is that part of what you guys do, that you connect it to a data lake? Yeah. So that's a great question, Craig.

7:57It depends on what you're doing, right? So there's always a source and a destination for any data flow. Like, for example, right now, if I get on, you know, whatever, I'm on a Mac. So if I get on a Safari and I click the buttons and I'm visiting a website, right, or if I'm using YouTube or whatever, then that flow starts from me and goes all the way to the servers of that company, which is typically all in the cloud. It's all cloud at this point. That's one kind of a flow. So in enterprises, you will hear that enterprises are using SaaS applications more and more. So think them using Zoom or think them using Salesforce Cloud or using Oracle Cloud or whatever, or G Suite or Office Suite from Microsoft and stuff.

8:43Those are all applications that reside in the cloud. So more and more you're connecting people's devices to their cloud. So that's one kind of connectivity. Where do we fly in? We don't own the devices. We don't own the SaaS cloud of those applications. We are the fabric in the middle that takes the data from here to there in the right SLAs and the right security and everything. So that's one flow. The other set of flows are that there's a huge amount of applications in enterprises that are not yet cloud-based. And we call them internal applications. And they run in the data centers of these enterprises.

9:19And I'm going to just use the term data centers because technically data centers can be running in the cloud as well. But let's just not go down into that complexity. let's just say that they run into enterprise data centers so that's the second part connecting you know your users to those data centers and the last one i'm going to categorize that as machine to machine communication now the interesting part is that this is the part that is exploding right now because when we talk about things like ai and we talk about you know all of the new deployments that are happening around AI, we talk about like a trillion dollar investment and stuff.

9:59They technically all fall in the machine to machine data and communication. So think about massive model generation for AI, a huge amount of inferencing and training and stuff. So that is the category where it's just certain machines and system talking to other machines and system. So we are in that networking as well. So very broadly speaking, Those are the three big categories that I can define when it comes to enterprise networks. When you come to a consumer side, then I would say 90 % of that is just going to, you know, essentially websites in the cloud. But on the enterprise sides, those other two are pretty, pretty heavy as well.

10:40Yeah. On the enterprise side, I mean, you were talking about, for example, from my computer to YouTube servers. and i don't know in that case but but in a lot of uh enterprise they're capturing all that data uh so so what do you do on your side is it just depends on what port you're plugging into and and that that's that those are the policies of the enterprises we we don't dictate it one way or another right it's just like our goal is to provide that technology in the middle if you as a business happen to be one of those businesses that want to look at every single thing that your employees do, you can do that.

11:24Or if you are one of those employees where you say like, hey, you know, as long as you're connecting to the work staff, this is what it is, everything else, do whatever you want to do. That's your policies because technology does not differentiate between the two. Technology allows you to do either of those. That's a corporate policy. The beauty of our technology is that you can implement whatever your corporate policies are on that technology. So there's a very clear distinction between what the technology does and what the policies ask it to do. And we don't like to get involved in the policy side of the equation for companies.

12:02And in the case of the government, that's the public policy part. Yeah. How has the introduction of, I mean, you were talking about machine to machine. Now we have agent to agent systems or networks. How do you, do you get involved in that or it doesn't really matter whether it's an agent that's connecting to the network or some dumb laptop? So I'll answer it two ways because one is providing the networking for all of these AI workloads to run on top of it. So that's obviously 100 % networks are. Right now, one would expect that the GPUs are probably the most important component, and they might be right on that.

12:53The compute side is pretty important, but as the scale of it grows, networking starts to become more and more and more critical in that. So we tend to call it that is networking for AI. So, well, how do you create networking? Think about data centers. How do you create massive data centers for AI? And how do you create networks where you can run edge AI and so on and so forth? It's a lot of big words there. But the reality is that's a networking component. But then on the flip side, so we are absolutely in the middle of that. The other side is that we ourselves have management systems that are agentic in nature.

13:27And we push out agents that do multiple different things in terms of network management or in terms of network monitoring or business planning and so on and so forth. It's massive. So in that case, we are the ones that are producing those agentic AI systems. We are the ones that are essentially productizing these fully autonomous AI agents that people will work. So these are two very separated things. Now, in a company like us, both of those exist. So while we are the network for AI, we are also the ones producing AI. And we are kind of in the forefront of both of these. Like, for example, when it comes to the AI itself, think about agentic AI or think about like chatbots or think about like multimodal interfaces and stuff.

14:18Probably in our industry, we are probably the farthest along. because we think of AI not as a product, but we think about AI as a capability of the platform. So I always say when it comes to AI, I don't think products, but think platforms. So it is already embedded in our platform. We call it Platform One. It's already embedded in there. It has huge amounts of capabilities, and you and I as a user can really interface with it in three different ways. We call it the interaction patterns. You can talk to it. That's the conversational interaction. And we are very used to it. That's what we all got exposed to when it came to ChatGPT, right?

14:58You go into a chat bot and you're talking to it and great, that's a conversational interface. The second one is a multimodal interface. And we are also used to that on the consumer side as well. Like, oh, you can ask it to create pictures and videos. Now in an enterprise environment, you're probably not asking it to create videos but there it becomes like reports and graphs and you know all those kind of stuff so that's called a multimodal interface and the last one which we are probably the farthest along than anybody else is fully autonomous agents that do something useful in an enterprise environment and you can't really compare a consumer grade ai agent to an enterprise grade ai agent they're very different because on the consumer side okay you know you can have it send an email for you or you know do things like that and you know it's doing it on behalf of a consumer so you know consumers have very lax you know privacy and security concerns if you will these days uh but when it comes to enterprises these are doing mission critical functions right and they have a huge set of guardrails around it based around what permissions does the agent have, then it is who ran the agent.

16:14If I run the agent versus you run the agent, and if my permission level is much lower than your permission level, then the agents that we run shouldn't be able to do the same thing. Now, these are things that are not even present on the consumer side, but you have to really think through those on the enterprise side, then data access really becomes important. You know, you hear in the news that, oh, somebody just put like your IP into chat GPD and then it got exposed and stuff. So how do you take care of that? And then the last one is, is it going to be all the agents from Xtreme? No, there'll be other companies that will create agents as well.

16:52So what is the structure of agents talking to other agents? There's a lot, that's a very, very new thing. So I know Anthropic have the MCP and I think Google just announced a recent one as well so that's still being developed but all of this space is contained in our products so we are we are we're pretty at the forefront of this and helping the industry kind of make up their mind and really consolidate and settle down on that and that's really the fun stuff. Yeah. Actually, first of all, can you give an example of an extreme network agent talking to another agent, whether it's within extremes?

17:37environment or not, right? So, yeah, so great. So let's just take an example. And I'll take relatively simpler examples because, you know, I want to make sure everybody in the audience can get it and it doesn't require, you know, very specific knowledge of networking. So think, so I'll say one statement and that will help us with this example. When you're thinking about agents, think job descriptions. Agents are very similar to the job descriptions that we are typically used to. For example, if you go into an enterprise or if you're buying something from an enterprise, you are typically used to two different kinds of roles.

18:14One is somebody that is advising you on what to buy and what are the good qualities of it, how do I compare this versus that? So think of it as an advisor. And the other one is when something goes wrong and invariably it goes wrong, you call a service agent, right? And today in today's world, both those two are human, right? Now, when you think about the agentic AI, they work very similarly. So called as the advisor agent and called as the service agent. And now they're not human. They just happen to be AI based. And they work exactly the same way. So in our environment, the advisor agent is present in your environment list in platform one, which is your console.

18:53And it is constantly giving you tips and tricks. And it's like, hey, by the way, in your network, blah, blah is happening. Maybe you should look at it. Have you considered doing this? Have you considered doing that? And so now think about that as an agent that is constantly reviewing everything and giving you ideas and options how to do this thing better. It's like having a consultant on your left shoulder 24 by 7. And the good thing is that you don't have to pay that consultant too much. So that's one agent. Now, service agent, it kicks in when something happens. Something went wrong. You know, like people in your company calls you and say like, hey, in my building in Chicago, I'm not able to connect or my Zoom is working really slow in Minneapolis or whatever.

19:39Right now, when that happens, that's where the service agent kicks in and it goes behind the scene. It picks up that ticket. It says like, oh, what the problem is. It thinks through it. It goes and does a lot of troubleshooting and then realize, ah, this is the problem. and can actually go and fix that problem. And if it cannot fix that problem, then what do you typically do? You call the tech support and you open a ticket. So it can open the ticket for you as well. Then what do you do? Every two minutes you follow up with the IT person. Is it fixed or not? It can do that for you as well. So that's the way to think about agents, what they do.

20:16Now, in terms of their communication, think about it this way. The advisor agent can invoke the service. agent or the service agent can invoke the advisor agent and when you start thinking in those ways it starts to get pretty overwhelming and pretty crazy but it is full on job descriptions and agents that are talking to each other and accomplishing something and there's a whole concept of agents creating teams dynamically behind the scenes to get something done like For example, in our environment, when a service agent wants to accomplish something, it will go in and it will pretty much pull other agents into its team.

21:01It will create a team, it will plan that team, and then it'll accomplish what it needs to accomplish, and it'll then disband the team. And all of this is happening behind the scenes. All of this is happening into the platform. This is what I was talking about, the machine-to-machine. Yeah. But the goal there is, before anybody gets too scared of it, is that for us, we are developing it in such a way that the concept of human in the loop is there, the concept of human over the loop is there, and the concept of human above the loop. And those are totally different conversations. But the idea there is that you can always insert anywhere human in the loop and you say, before you do this kind of an action, get approval.

21:47human above the loop is all about, you know, whether it is readability of it, observability of it, like, you know, explainability, these are all concepts in AI, right? So the idea is that, yeah, you can observe the AI doing thing, but there's nothing in the AI that is black box, and you cannot understand that, right? And then above the loop is really like where you define the agency of the asians so it's a fascinating world craig and all of this is happening today yeah yeah i mean one we'll go ahead were you going to see something more no no please go ahead yeah yeah the uh this these teams of agents and and uh in the network uh that that all of this data is flowing through uh it's do you have a visualization because it it really increasingly is is like uh you know i mean that's what the that's why we use the term web but this increasingly dense web of connections and one of the things i'm interested in is when data flows through such a network a very complex network and agents are say an agent has a certain permission and it's accessing certain data and using it to get something done with another agent how do you track where that data is flowing.

23:26I mean, how do you know that it's not someone left some switch open and it's flowing to another agent that then... Absolutely. And so the way to describe it is that that's not the job of the network per se. So we don't build it into the network. We build it into the architecture of the agents itself. I see. Yeah. Although we are sitting on both sides, right? So from a network point of view, if you and I are agents let's assume we are ai and you're not humans and if you are talking and if the network is supposed to allow you and i to talk then it will just allow you and i to talk it doesn't really know whether we are talking about uh you know ai or we are talking about you know um how to bake a cake right so it doesn't know that and that's not the purpose of the network so network is there to provide that communication now what we should be talking about what we are allowed to talk about what are the guardrails around our conversations, what is the permission, and what is the agency.

24:26Agency means how much can we actually do, and at what point in time it's like, no, this is outside of my agency. Now I got to go to punt it to another agent or to a human. All of those are built into the architecture of the AI itself. So this is the difference between just going and building a chatbot, which quite frankly, at this point in time, anybody can do. And I laugh about it. Anybody that has 45 minutes to spend on YouTube can actually build a chatbot. There's so many videos out there. Between that versus building these enterprise-grade systems, that's the difference. So when we are adding AI into all of our cloud management and stuff, we are building it ground up with guardrails, with permissions, with RBACs, with agencies.

25:16And these are the most important and the difficult part is. So while from a network point of view, we don't care, but from a cloud management or a cloud software point of view, we absolutely care. And all of those things are built into the system. And that's why, if you ask me, the whole concept of like, hey, we're going to put a chat bot in every application out there, that's just so wrong. Wrong in the sense that it's not even useful. It's kind of the hype part of it. And most of it will get through that in the next year, year and a half. And it will really come down to AI becoming the core capability of these large platforms with all of these guardrails and checks and balances and most importantly, the explainability part of it built into it.

26:05Because in the end, we will get to a point very quickly in the next couple of years, if not sooner than that, where the human trust will be at that point where we will allow AI to do a lot more things for us. Humans are just very trusting by nature. So we're already seeing that on the consumer side. It will happen on the enterprise side as well. And that's great. There's a lot of productivity in there. But what we want to make sure is that when needed, we can have the system explain how exactly it did what it did and why exactly it did what it did. So the ability to audit AI and the ability to explain AI, that is absolutely critical.

26:51And I think the systems that are built without it are most probably not going to be successful in an enterprise environment. And I quite frankly think that building AI that is not explainable is probably not a good thing. I'll leave it there. Yeah. And you sound like you're talking about the unified management across all of the different nodes and connections. And do you have, is there an AI layer that's managing that, that's doing that management across all of it? Absolutely. So we have many, many, many, many different things that we do in the cloud. Some of it is managing the network. Some of it is troubleshooting networks.

27:45Some of it is reporting. Some of it is security products. Some of it is just business analysis and stuff. So there's multiple different capabilities. And there's AI embedded in all of them. and AI is always embedded in the following way. And this is my thought process and the way I wrap my head around AI is the following because in the end, it's not about just, hey, let's just stuff more AI into things. You want to accomplish something. So what is it that you're trying to accomplish? And a very easy way of wrapping your head around it is what I coined a couple of years, two, three years ago now, the ARC framework.

28:24So what does that mean? which means that start with what is it that you want to accelerate? You know, it's like, hey, I do this thing every day and I would love to automate it. I would love to accelerate it. I would love to spend less time on it. Okay, that's something that you want to accomplish. AI is very good at accelerating things. Second thing is what are the hurdles and what are the bad experiences that I have that I just don't want to have? I want to just replace them from my stuff. I think, you know, all of us calling, you know, support for some company and sitting on the music hold, that elevator music for like two hours.

29:00Nobody wants to do that. AI can be applied to remove those kind of portions. And the last one, the C stands for create, which is creating new experiences. Things that were previously just not possible. Now think about, you know, what is happening in the medical space, right, where new diagnoses are being done or, you know, in pharma where AI is being used to create new peptides and proteins and stuff. So that's like creating something completely new. So that's the way to think about it. Because in the end, after talking for 20, 30 minutes about all this crazy technology, my fundamental belief is that the phase in which we are right now for AI, AI is there to accomplish something for the human at the end of the chain.

29:46and that human can be a consumer that's using it it could be somebody sitting in the enterprise that is using it it could be the public sector using it but there is a human sitting at the end of the chain and the ai is developed and must be developed and productized to bring some values of that person and you can generally categorize value in this arc framework either i'm accelerating something something good for that end user i'm replacing some pain points and some bad experiences or I'm creating brand new value for the customer. And this is really, after talking about technology, so broad bringing in doubt to earth, because we need to have that check on the technology itself.

30:31Otherwise, it's tech for the sake of tech. And we don't sit in academia. We don't do tech for the sake of tech. We do tech for productization, which has to result in some value creation. Yeah. You were talking about explainability and presumably you have AI driven insights into what's happening on the network. Can you talk about that? So a systems engineer at a big company or a network, I guess, is it the systems engineer, the network? engineer. So it could be the system engineer, it could be network engineer, it could be sec ops person, it could be cloud ops person. And quite frankly, more and more, it could be the CMO, the marketing folks, because they generate a lot of their marketing campaigns based on data.