Episode Transcript
[00:00:00] Speaker A: Foreign.
What's up, everyone? It's KB and I'm on the go at Extreme Connect at the Carib Royale Hotel in Orlando, Florida. This week, some of the biggest conversations shaping networking, artificial intelligence and cybersecurity are all happening at once. This week is where operators, engineers and executives get into the details, what's working, what's not, and what's actually changing inside enterprise environments at scale. We're hearing a lot about AI driven networking, automation and visibility, but the real question is how does that translate into security outcomes, operational resilience and ultimately cost?
Because behind every intelligent network claim, there's still a human team trying to make sense of complexity. Across the next few segments, we'll be speaking with leaders on the ground, unpacking what's real, what's noise, and where the industry is genuinely heading. Next, KB on the go from Extreme Connect, Orlando. Let's get into it.
Okay, so joining me now in person is Marcus Nismel, CTO Amir and head of office of the CTO at Extreme Networks. And today we're discussing autonomous networks and what the future here holds and what it looks like. So, Marcus, thanks for joining me and welcome.
[00:01:22] Speaker B: Thank you. Glad to be here.
[00:01:24] Speaker A: Okay, so I really want to start with networks used to be invisible infrastructure, and so now you're saying they're being positioned as strategic assets. So what fundamentally has changed and why are boards suddenly paying attention to the network layer? And then there's always this saying, out of sight, out of mind. So clearly this doesn't really apply here. So I'm keen to hear your thoughts.
[00:01:46] Speaker B: Yes, absolutely, yeah. Network have been seen as a utility for quite a long time. Piping, plumbing and base connectivity. But if you think about what's happening around AI, networks for AI are going to be critically important. AI is transforming our business. And so the business relies more and more on AI and the connectivity associated with it. It starts with data generation at the edge that needs to be properly secured and made available to agents that are deployed anywhere in the enterprise. And those agents are fueled and basically powered by GPUs that provide inference in the data center. So ultimately, multiple parts of the network that have become strategic, if people work on the infrastructure or use the infrastructure, it's easy to work around challenges. When agents do that, it's going to be challenging for those agents to work around those connectivity issues. So ultimately, network becomes more strategic because more and more business processes rely on AI and agentic AI specifically. And so the business processes will rely and are relying more and more on the network as A strategic asset.
[00:03:00] Speaker A: And would you say, given Your experience last 25 years, would you say that there's been a point in time where like people sort of networking got relegated a little bit? Like people sort of forgot about it? Because we're talking about all these other things now. But now from what you're saying, obviously it's sort of come on the scene again.
[00:03:16] Speaker B: It has changed. So 2022 certainly has changed a lot. Actually, it really started already when Covid hit, where customers recognized the strategic importance of networking technologies and was AI in 2022. It definitely got supercharged from an importance and a strategic point of view.
[00:03:36] Speaker A: So a lot of people I'm interviewing at the moment talking obviously about like AI, you know, agentic AI agents. A lot of that is derived from all the data. So would you say now looking at the network layer and all the traffic and all the data that goes with that, it's really helping inform perhaps the next wave of how people approaching networking,
[00:03:56] Speaker B: for example, it will certainly change the way on how people design the network. So data basically comes in two flavors. You need to be able to transport data reliably and in real time to those agentic systems, because more and more AI solutions are making decisions in real time. So overall, the days where you store your data somewhere in a lakehouse, you wait a couple of days or weeks and then you start processing, provide analytics on top of it and try to make decisions. More and more things are happening in real time. So the real time nature of data has certainly changed. And what also has changed is that enterprises have recognized the importance of data, data governance, data structure and making data understandable by those agentic systems. And that applies both to the transport, but also the way on how you manage your enterprise data overall.
[00:04:48] Speaker A: I want to take a step back for a moment and talk through perhaps Even the last 10, 15 years in this space. So it was at one point in time, it's like, let's capture all the data and look at every single thing. And then people got worried about having too much data, so then they were trying to like get rid of it because of PII and cybersecurity breaches. And now it's like we're sort of seeing the same theme happen again.
So where do you think, given your role and how you see this space, people, are they talking to you about what do they really care about?
[00:05:16] Speaker B: I think it's all about the right data. And when I say the right data, it's the right data at the right point in time available to those agents for Inference. So we're not talking about training and learning, this is a different topic, so to speak. But inference is really important and the right data is timely, has the appropriate quality and accuracy, which is important that hasn't been focused on by customers quite a lot. And most importantly especially enterprise data is sometimes really hard to understand for those agentic system. So providing the right context, having a semantic layer in place that allows those agents to understand the underlying data and the business relationships in between, is going to be critically important. So when I say it's the right data, it encompasses all of those dimensions, being on time, in real time, being accurate, quality needs to be discoverable, but then also understandable where the semantic layer and ontologies are important.
And that creates the context layer for agents to reason correctly and make the most out of enterprise data.
[00:06:26] Speaker A: Okay, so I want to move on slightly now and talk about autonomous networks. So perhaps describe to me in your words and what's actually driving real investment here versus like hype. And I only say that because like, I mean people like you all the time and everyone's saying, oh, we're doing this and this is the next thing. So I just really want to separate that so it's not just coming across as bravado.
[00:06:45] Speaker B: Yeah, so that's the other side of the coin that is really AI for network operations and broader network lifecycle management. As we look at it, the main driver or the main drivers for that is certainly a shortage of knowledge in the industry. So getting expert engineers on staff for most of our customers is challenging.
Partners are in a little bit of a better place, but everybody is competing for talent and most of the talent don't feel themselves drawn to networking, but other IT and SaaS related technology. So there's definitely a shortage of staff and skillset and AI just helps to basically plug the gap or cover the gap here in terms of getting new employees quicker onboarded to the technology itself, but then also over time more and more take over the operations of the network. So AI is really evolving from an assistant and co worker to almost an operator over time. And depending on what area of the network life cycle we are talking about, that operator capabilities is stronger. In some other areas it's still evolving. And I think it's important for us as a vendor to provide our customers with the ability to move at their pace. Depending on how trust is being built and how comfortable the customers and their employees are with the technology. You have to have the ability to hand over the control to them and they can control what AI is doing. On top of their network to operate it.
[00:08:21] Speaker A: So what would you say Markus, is their pace at the moment?
[00:08:24] Speaker B: Fast their pace. There's a wide spectrum, I would say based on our experience and the spectrum is based on the employee's maturity as well. And I think this is something that people forget. AI in general and also AI for networking is a transformational task in ask and transformation always means you have to have people, technology and process being aligned to be fully successful and across all verticals and across all sizes of enterprises. The maturity level varies dramatically because just having technology doesn't help you. If you don't give your employees the time to get familiar with AI as a technology so they can understand where and how to use it best in their today work, then you're not going to get the maximum out of it. If you don't adjust your processes around change management and how networks are being operated, then you also won't get the maximum out of your investment. So again, people, process, technology, all need to be aligned for customers to maximize the return investment they would get from any AI initiative. And specifically also AI for networking. And we do see around help desk support knowledge transfer, troubleshooting, automated troubleshooting and root cause analysis. Today most of the successes with our AI solution, which is AI Agent one on Platform one that we're going to basically announced this week here and the first generation of that was AI expert Service Agent and think it through being in place and being leveraged by customers. Exactly. For those use cases, knowledge transfer, operational insights, but then also accelerated troubleshooting and root cause analysis which then overall increases the reliability and availability of the underlying infrastructure and can be supported by both junior and senior engineers at the same time.
[00:10:27] Speaker A: So you mentioned before the word trust being the operative word at the moment.
So that probably leads me into my next question for you. Markus would be around moving from manual control to AI driven automation requires maybe like a leap of faith as some people would describe it. But what are organizations doing practically to build trust in systems they no longer fully control like nowadays?
And yes, we can say that there's a human in the loop. That's often what I'm hearing coming back from vendors like yourself. But there's still parts being formed by a machine that you have to relinquish that control.
[00:10:59] Speaker B: Yeah, we always say trust builds over trust. You build adoption and over adoption you, you get success and building trust. Yes, control is one part of it and you mentioned it already. Human in the loop is a component to it and we obviously want to move from human in the loop to human on the loop. That would be also the transition from coworker to operator.
[00:11:20] Speaker A: But.
[00:11:20] Speaker B: But to achieve that it takes some time. It obviously takes AI literacy on the employee side, but it also takes explainability on our side. So we need to be able to make the decisions that our system is proposing transparent to the user. And with explainability and transparency, trust is being built and then control is relinquished over time. And to achieve that, the predictability of the system needs to be there. So we pay also a lot of attention on how we create a harness around our AI solution, which includes all of the guardrails and policies and controls that are being in place to ensure that an agentic system does exactly what it's supposed to be and becomes more and more deterministic. And that again builds trust and that will drive adoption and then ultimately a different level of autonomy from an operational point of view.
[00:12:17] Speaker A: So we've gone from human in the loop to now your point, human on the loop. So is that an extension then? Is that more an advanced sort of level, would you say?
[00:12:27] Speaker B: That's definitely an advanced level where the user becomes an observer and somebody who is only managing exceptions, but the main tasks are being performed by the underlying agent. And that's the ultimate goal because that is more what most people would associate with autonomous networks, that you basically make that transition from in the loop to on the loop, which means not all of the decisions are being done fully autonomously, but only exceptions are being passed on to the user. And then you get obviously a different level of efficiency gains in the process.
[00:13:03] Speaker A: So then you said before that's the ultimate goal. Why do you think that's the ultimate goal?
[00:13:08] Speaker B: In your eyes, it's the ultimate goal? Because going back to the previous argument that there's no expertise in staff staffing available to run those very complex networks, and they are becoming more and more complex. As we said at the very beginning, networks are become more strategic, they become more complex. There are more agents, more sensors, more actors, GPUs, CPU, cloud on prem hybrid. So it's really, really difficult for a human to oversee and manage that complexity that keeps on growing and growing. And from our point of view, AI is a key technology that allows our customers to keep control and also scale with the requirements just to touch on
[00:13:52] Speaker A: the employees for a moment. Now, this year has been a lot of redundancies across vendors. For example, would you say that you said before that it is hard to find people of this skill set because some of these people who perhaps been displaced in their current Roles. Could this be something they could retrain to learn this, would you say, in terms of being redeployed?
[00:14:09] Speaker B: Absolutely. I mean there was a recent World Economic Forum assessment of the impact of AI on employment. And I think if I recall it correctly, they were talking about 92 million jobs being eliminated through AI, but 170 created through AI. So there's a net benefit of 60 million.
But those 90 million that got eliminated don't really get eliminated. If you think about it and you focus on how to leverage AI as a technology, it is just transitioning and changing the way on how jobs are being done and specifically around it and networking, we feel that the human expertise is still required and retraining is actually key to be successful in everybody's career. And when I say retraining, you want to combine your expertise with the ability to leverage AI. And we have seen that also across the board that the adoption of AI is strongest with very young people and very senior people, which is really interesting because the senior people, they want to get rid of those day to day tasks that are boring for them, the mundane tasks. And they have figured out how to use AI and maximize their efficiency with AI. And the younger generation is obviously very curious.
There's no doubt about it, they are growing up with AI. For them it's a no brainer. It's more like the middle seniority has a harder time to transition, but they have to figure out how to transition and how to leverage AI to become more effective and also valuable in the future.
[00:15:52] Speaker A: What would be your reasoning as to why the people in the middle are having a harder time to transition, would you say?
[00:15:57] Speaker B: I mean, change is always hard for somebody and I think the ones that are mid career have the most to lose. They haven't been as successful as they want to be and they're also not like that young in mind anymore. And I think they have a harder time just in general to make that transition. So that is publicly reported data, but this is also something that I'm seeing even internally because we are also an organization that is using AI internally for efficiency gains as well, whether it's in engineering, marketing, sales, wherever.
And I can say that there are similar patterns evolving for sure.
[00:16:36] Speaker A: Would you say people like companies are perhaps trying to hold on to how they used to do things? Because I'm hearing now from people that I'm interviewing saying we have to do things faster, we've got to be more competitive, we have to get to market quicker. You know, the release schedule is getting the velocity of that is really intense. So would you say that are there still organisations out there that are trying to hold on to the old days?
[00:16:58] Speaker B: Oh, yeah, absolutely. Those are the ones. Going back to the previous discussion that we had around people process technology, those organizations need to start changing their processes to maximize the benefits that they can get out of AI. And one other perspective on how to get the maximum out of AI is not just looking at efficiencies, but what we at extreme did early on as we developed AI for Platform one, we said, hey, efficiencies are great for us, but we as a vendor, we want to make sure that we leverage AI to create differentiated products that give us an edge in the marketplace. And we are seeing that resonating with the market right now. So we built something with AI that wasn't possible to achieve for our customers before without AI, and that allows us to be in a different market position. And this is something that every organization should think about. How can you use AI to build something that differentiates yourself in the marketplace? Either moves you into a new market overall or provides you a key differentiation in your existing market so you can gain market share but also protect yourself with a new existing market against market disruption from new market entrants that are using going to use AI as a key technology to differentiate themselves.
[00:18:17] Speaker A: I want to talk about the risk layer. So, for example, networks, as you've been discussing here today, going to become more autonomous with leveraging AI, et cetera. Do we as an industry risk creating a new class of failure? And so what I mean by that is where AI decisions scale problems faster than humans can intervene.
[00:18:36] Speaker B: Yeah, so that is certainly something where the previous discussion around control and building a harness around your agentic layer becomes really important. It can obviously, technically it can happen. The only way to control this is to put controls and a harness in place that basically describes what an agent or an agentic system is able to do and what it's able not to do, and when an exception needs to be forwarded and pushed up to the user to make a final decision. But that comes down to, let's say, an engineering and architectural design problem that vendors like us have to solve. But this is absolutely a requirement. And if you think about security and AI, obviously attackers are moving much faster. So the only way to respond is also to respond faster using AI. So it's a little bit of a, how do you say, like a hamster wheel that is going on. So you just need to jump on it, but you need to have controls in place. But yes, there's certainly a risk around this.
[00:19:35] Speaker A: Can you talk a little bit more about the harness in a little bit more detail? What does that look like for people, perhaps that are unsure?
[00:19:42] Speaker B: So harness is basically the entirety of controls that you are putting in place. It starts with guardrails. What is the agentic layer even able to accept? From a prompt perspective?
It includes data access control. So what data has the agentic layer access to and what access control are you enforcing specifically? It looks at what tools the agentic layer is able to use and leverage and with what access levels. And how do you monitor tool usage to also detect anomalies, detect unexpected things that shouldn't happen? It's the entirety of all of those controls. Data access tools access how users can interact with the agent, what the agent is allowed to do, and what is considered outside of the boundaries. And then how do you put controls and sandboxing technologies in place to avoid that those agentic systems go off the rails, so to speak? This entirety basically describes the harness.
[00:20:44] Speaker A: Would you still say that companies are in like a stage where they're trying to figure all this out? So I was at recently I was at an event in Alaska and one of the questions was like, what do we do about implementing guardrails around AI?
So do you think companies are still trying to proverbially find their feet?
[00:21:00] Speaker B: Yes. I mean, as always, there are laggers and there are leaders in that space. So I think it's still relatively new. If you think about the AI cycle and compared with other general purpose technologies like electricity, the steam engine, whatever you can think about, those technologies took a while to really get adopted widely, which AI has been already. But making effective use of AI. I think it's still early days in the grand scheme of things. We are just four to five years in the latest generation of AI and we are definitely not done yet. So obviously people haven't figured it all out, but everybody's trying to figure it
[00:21:44] Speaker A: out and really quickly. Rapid fire question to end with, what do you think is going to happen now for the rest of 2026?
[00:21:50] Speaker B: From our point of view, we will see a higher adoption of, let's say, AI as an operator. So we will see the first signals from the market that people are moving from in the loop to on the loop for specific tasks within network operations. That is going to be big. With Agent one that we are announcing this week, we're taking a different approach on how people can personalize and interact with agents. We get inspired by what's happening in the consumer space. There Quite a lot and how you can extend your agent and agentic capabilities with skills. I think this is going to be big. It's not about who's writing the best prompt anymore. It's more about who can write the best skill extensions to an existing agentic framework to expand its capabilities. So that's going to be the next frontier and beyond. 26 I think it's as good as anybody's guess what's going to happen now. But for sure we are not at the end of the innovation cycle again by far.
[00:22:54] Speaker A: Joining me now in person is Michael Jones, commonly known as nj, VP of AI and Innovation and Office of the CTO at Extreme Networks. And today we're discussing agentic AI in the network and where it actually works. So mj, thanks for joining me and welcome.
[00:23:08] Speaker C: Thank you. It's great to be here.
[00:23:10] Speaker A: Okay, so let's start a little bit high level. What's going on on the landscape at the moment when it comes to like network management? What are you hearing? What are customers saying? Tell me more.
[00:23:18] Speaker C: Oh, wow.
I mean there's so much going on in network management and AI because you know, my focus is on AI and the convergence of those two is a very hot space I think. You know, first of all, networks have never been more important, more integrated into our daily lives. The expectation around network performance has never been higher. So I'd say what we're seeing is networks getting larger and more complex while the expectations of those networks are rising, that they're going to be more stable, more reliable. And ultimately what really matters is that people have a better experience. So we have that on one side, on the AI side, I mean, you have to be like under a rock somewhere not to have heard about AI or anthropic or OpenAI or how AI is fundamentally changing how we approach work today.
And so when you look at the intersection of those two, thinking about networking, a lot of the complexity there, there's so many tools and protocols involved in networking.
Networks need to be available 24,7. So how can we staff those 24,7? How can we quickly resolve issues, identify issues or even prevent them from happening? So it's pretty exciting time there. And ultimately what I see is by applying the right type of AI in your network, you're going to be able to ultimately improve the end user experience.
You're going to be able to do that 24, 7. You know, end of the day, it's just enabling people to be able to do their work and have a great experience.
[00:24:50] Speaker A: Okay, so there's a Couple of things in there, mj, that you said I want to get into. So when you say have a good experience, what does that mean to you?
[00:24:56] Speaker C: Yeah, so we look at experience in a number of ways and I think that's where I feel like traditionally networking has been approached in the wrong way. People always sort of start with let's go look at the device health or let's go look at the rf or you know, they're really starting by going, you know, they're sort of starting on one end of the equation, which is going through the actual devices themselves, the access points, et cetera. And so I think that's the wrong approach. I think what you should start is on the other side of that equation, which is what is the client experience.
So that may be, you know, simply us here recording a podcast and like our, what's the jitter rate on that, things like that. Or perhaps we're all in a big teams meeting, you know, what was the quality of the experience there?
So I, I do think it's an exciting time because we can now start to think more about what is the experience. And then based on the experience, if that experience is degraded, then you go back and look at device health or then you look at RF interference, etc.
[00:25:54] Speaker A: So you mentioned before available 24 by 7. So a lot of people that I'm interviewing at your level talking a lot about this, why more expectations, critical infrastructure, things going to be available.
Also to extend on that, customers, people expect more nowadays. It's like something happens for an hour, people start losing their minds. Talk to me a little bit more about that because I think that this is something that perhaps people know about, but maybe you can shed a little bit more light on this.
[00:26:18] Speaker C: Yeah, the 247 is, you know, talking to some of our, you know, speaking with some of our customers over in healthcare for instance. You know, they're heavily network dependent there. I mean, face it, it's hard to hire people and to have that properly staffed 247 so frequently I think happens is the quote unquote, graveyard shift is staffed by some intern level individual that is probably out of their league.
[00:26:43] Speaker A: So I want to slightly move on now. Talk about there was tool sprawl, heard a lot about that and now we're getting already onto agent sprawl.
[00:26:52] Speaker C: Yeah, that's a good one.
[00:26:53] Speaker A: So how do we get here so quickly?
[00:26:55] Speaker C: Yeah, I love that one because I do the classic saying is, oh, there's an app for that. That same analogy has, you know, come true here. Honestly, we, we Suffered this ourselves here extreme. So we, we launched AI Expert. Soon after that, we Agent. Very quickly after that, customers were telling me like, mj, how do I know when to talk to AI expert versus Service agent? Or like, how do I call service agent? Or does AI expert talk to service Agent? We quickly saw that forcing that decision on the customer was the wrong way to do it. Don't give them more options, make it simple for them. And so to combat essentially this agent sprawl, we're launching Agent One. So Agent One is essentially, you know, your, your single UI into AI and so no longer do you have to think about, oh, is this a service question, is this device health question, et cetera. You'd simply go to Agent One and it takes care of everything from there. So these are what we call multi agent systems. So under the hood there's a collection of agents working, but ultimately it's about the customer and we really needed to dramatically simplify that experience. We didn't want to force the technical constraints up to our customer. We wanted to solve those and make it easy for them to avoid what we all know of that proliferation of apps. And our phones are full of them and we can't find find the one we want.
[00:28:10] Speaker A: Okay, so you're right. I want to talk about this more because like every vendor is saying, oh, we're trying to reduce stuff down, but then it's like, oh, they try to claim we've got this single pane of glass, or maybe to your point, Agent 1, but then you still got to log into like 50 other tools anyway.
So do you think that now there is more of an onus now from customers that like, we just want the single thing to log into, we're not going to do anything else. Are you see that coming down the pipe from some of your customers as well? Because there's not just. They're dealing with 50 other vendors that they've got in their tech stack.
[00:28:40] Speaker C: Exactly. Yeah, absolutely. We used to call it a swivel chair experience where our customers had to literally swivel from one console to another console to another console. You know, you use the word tools there talked about tools. And so what's happening is rather than making the user go log into all these tools, we're now providing these tools to the agent.
So all you have to define is the objective. You define the outcome that you want and then let the agent go, select those tools and interoperate with those tools. So put the tool fatigue on the agent, simplify it for the human.
[00:29:12] Speaker A: Do you think so going back to just the AI sprawl, do you think that even the last few years it's really boomed? Right. And like everyone's come out with something.
[00:29:21] Speaker C: No. In the last three months. Two months. One month. One week, yeah.
[00:29:25] Speaker A: Do you think that these companies will probably just go by the wayside now? Because it was like huge boom and there was like, I mean, there's talks around the whole AI bubble and all that. That's separate. But is it just that everyone was just trying to tinker and have a play and maybe create something? Maybe it works, maybe it doesn't. Will we start to see these tools start to diminish or.
[00:29:44] Speaker C: No, I think you're right. I spent a lot of time prior in social analysis and essentially, you know, I'd say we're sort of in the Friendster coming up on early Twitter phase. So it's very nascent right now. I mean, you know, just in the past several months we've seen OpenAI go from literally the top to now anthropic's on the top. And we're going to continue and even
[00:30:05] Speaker A: that's apparently being questioned, as I read this morning.
[00:30:07] Speaker C: And we're going to continue to see that, you know, don't sleep on Google. Gemini is. They have a tremendous amount of data there. Meta's still in the game. We got amazing stuff out of deepseek. I think it's still early innings. And I think what's important is that people need to get in and safely experiment. The old adage of you have to spend money to make money. There's going to be some things that fall on the floor and are not used, but in order us to really innovate and find new ways of doing things, we have to get in there and start taking some actions. Yes, it's early days. Yes, there will be some waste, but ultimately what will emerge will be some pretty amazing technology.
[00:30:43] Speaker A: Okay, this is really interesting. So do you think that it's anyone's game now? So like back in the day it was like, you know, your Cisco is a big dominant player now. They're sort of being, you know, companies now are trying to rattle and then you've got companies that came left of field with multi billion dollar acquisition from Google recently. Very good capability.
So do you think that we might just randomly see a company that no one's ever heard of just to be the top dog? Because like you said, it was all about OpenAI and then like, it's almost like, well, who cares about them right now?
[00:31:11] Speaker C: Absolutely. So I've been in the space for a while. I used to work on the Salesforce AI research team. I still have a lot of friends there.
There's a lot of brilliant people working on startups that nobody even knows about right now. They're doing incredible things. So absolutely there is a ton still to be seen as, you know, the first leader is rarely the one that wins. So we're going to see a lot more coming out. But so what we're talking about right now is more on the frontier model side of things. The reasoning side, I will say. I mean, I hate to call it a commodity, but really anybody has access to those. Anybody can go get a quad subscription or an OpenAI subscription. So what's really differentiating things in AI is it comes down to the context, the data. The context is really what matters. The old adage of context is king. That is more true than ever before.
So when you can, for just a few bucks get access to the world's most incredible state of the art model, what's really going to matter is the context. Bringing that context into that model, I think that's the real differentiator is going to be in that. The context I also look at, there's the context layer, but there's also the harness is what we like to talk about.
[00:32:20] Speaker A: You know that I think Marcus Nispel spoke to me about that. Just.
[00:32:23] Speaker C: Yeah, the harness. Okay. So I'd say I've kind of seen three phases in AI. The first one was a basic chatbot where you ask it a question and it give you a response. From there we went into thinking, so you'd ask it a question, it would think it would take you a few minutes and it come back with a response. What's happening now is we're giving these agents tools and skills and memory and context. So we're surrounding this reasoning model with access to memory on like how you like to work, or what your preferences are, or what you do well at, or what you don't do well at.
Then we're giving it access to skills. You know, skills is procedural knowledge is very focused on a specific space. So suppose we have a whole collection of skills that are focused on healthcare, that are aware of HIPAA and all those constraints there.
So ultimately what we're seeing is sort of this third big wave around AI in which we're building a harness around it and it's bringing in the tools, the skills, the memory and the knowledge along with that reasoning and context. That's why right now is such an exciting time.
[00:33:23] Speaker A: Okay, I Want to talk to you about.
We've gone from co pilots. I was just talking to Marcus before about the whole Human in Loop and now extremes version is human on the loop. So I've just discovered that. So what does that sort of shift actually look like a network environment day to day in your eyes?
Because we're sort of going more towards fully autonomous agents.
[00:33:45] Speaker C: Like we just discussing AI adoption is going to move at the speed of trust. If people don't trust AI, they're not going to let it go run wild. They're not going to say, yeah, go do that for me. I don't think it's just this binary choice of either we do everything or we let it do everything. It's going to be a gradual progression of building trust together. And so Human in the Loop is one such mechanism there where it would ask you before it takes effect. Ultimately what we're doing is building the right governance model around it.
We sort of look at three tiers there, we look at the deployment scope. So what scope of your network does it have access to? We look at the rbac, what tools and permissions do you have? And then finally we look at Human in the loop. Is that on or off? And so we do have a model in which we believe customers will be able to build trust with these systems. And as you start to trust something more, you will defer more to it.
[00:34:38] Speaker A: Would you say though, given your experience, people generally do trust these systems.
[00:34:43] Speaker C: I mean, there's countless examples out there around how these systems count the number of Rs in the letter strawberry or can do some rather basic tasks. So there's this concept called the jagged edge of intelligence. And this jagged edge of intelligence essentially speaks to the fact of these AI models can do such incredible things like PhD level things, but then on some of the most basic tasks they just fall apart. And so that jagged edge of intelligence is why it's going to take time to build that trust. So I think over time people will become more comfortable with it. But no, I think right now there's countless examples of things hallucinating or not working properly. And when you're in a mission critical space, it's going to be a gradual over time with the right amount of governance and then observability. You want to be able to go back and look at what did it do and why it's through a lot of these features that we're releasing that customers will be able to move at their own pace. It's not a binary choice of you do it or it does. It's going to be a gradual path
[00:35:43] Speaker A: just to add to that. Would you say that's more of a capability problem rather than a trust? So it's like, if I'm a customer, I'd be like, oh, I want to trust the system despite whether it can count how many Rs are, in a word, for example.
[00:35:54] Speaker C: I think if you just kind of go walk out on the street and you ask people like, what do you think about AI? You're going to find some people that absolutely love it and they're like, oh man, it's done this for me and that for me. And other people are like, yeah, it failed for me here and I didn't trust it. I think many people tested these AIs six months ago. And the thing is, the AI that you use today is the worst AI you're ever going to use in your life because tomorrow it's going to be better. And so I think the people that really frustrate me are people that are like, oh yeah, mj, I tried that last year and it didn't work, so I'm not going to do that again. And I'm like, oh my gosh, you realize like one year ago is like 20 years ago in AI and things have gotten much better. So I think public opinion is very mixed at the moment. And it's through the experiences we have with it, it's through the transparency that it provides. And ultimately nothing sells like success. So if people taste test, try it, see success, are able to repeat their successes, then they will continue to hand off more to these systems.
[00:36:48] Speaker A: So I'm curious to know what specific workflows and network operations are being fundamentally reimagined with agentic AI. So to clarify, I mean, not just like optimize, but like completely like redesign. Can you talk through that?
[00:37:01] Speaker C: Perhaps with our group, we rethink everything, really everything. And that's where I do like to come back to. It comes down to the experience. So rather than start at that with a collection of diagnostic tools, which really come down to what is the experience, and then from there you can ladder into the the correct solutions for it.
[00:37:21] Speaker A: So what do you think moving forward we're going to see in the market? I know it's hard to answer. It's just more curious given, you know, the caliber of person you are. What do you expect, whether it's about the industry in general, from customers, what do you think about.
[00:37:36] Speaker C: I'm techno optimist. I think it's going to get better. I imagine back to some of our healthcare customers that have a hard time staffing at the proper levels in the evening, I think having a coworker there with this junior staff person to a system is really going to be beneficial. I think it's really going to change the way we work. I think a lot of people start their day off by checking a variety of dashboards and logging into different systems and a lot of rather menial tasks to go check on things. So rather than having our customers have to go to all these different systems, with that swivel chair experience, you're going to, you know, sit down with your cup of coffee in the morning and get a recap, and your agent's going to say, hey, I. I checked everything this morning and this is what I found. And these things you've told me I can take care of, so I went ahead and took care of them, just letting you know these are the things that you want it to review, so these are ready for your review. And these are some things that I'm anticipating are going to come up in the future. So if you have extra time, you might want to look into that as well. I think from that initial experience and everything beyond that, I think you're no longer going to be tied to a desk. You'll be able to use your mobile rather than having to type. You'll be able to talk to it. I'm pretty excited about it. I think from the experience it's going to be radically better. And at the end of the day, it's going to be about those users on your network.
[00:38:53] Speaker A: And then that sort of brings me back to the trust. So do you think it'll just take time? Do you think people will just never be on board with it, would you say? But if they're not, I mean, so many people are saying competitive edge. Like, now, customers expect more. So that means that they have to deploy things faster. They have to leverage AI to give them that competitive edge.
[00:39:10] Speaker C: So it's roughly the bell curve. Like, when I talk to customers, some of them sort of on one side of the curve are like, mj, I know you like AI, but I don't want anything to do with it, so you keep that to yourself. The vast majority there in the middle are what I say, are AI curious? They're like, hey, I find this interesting and that interesting. Can you tell me more? And then there's that other edge that just blows my mind, where they're like, MJ, we took the APIs and threw it into here, and I gave it access to this and this is what happened happened. And I'm like, wow, that's pretty amazing. I definitely see a bell curve of adoption out there. I think one of the big drivers here though is like we used to talk about the consumerization of it and what I mean by that is what you experience in your consumer life, you start to expect in your work life.
[00:39:51] Speaker A: Give me an example.
[00:39:53] Speaker C: All these applications that we have grown to know and love on our phones, like our WhatsApp and chat at enterprise level, that's teams and Slack.
And so I think what we're going to see is remember OpenAI is a consumer application. So there's a massive population that is using OpenAI every day now, doing grocery lists, helping plan things, helping research things, etc. So people are using AI and getting very comfortable with it there. They're going to start to expect that same experience at work that they no longer have to go do so much that they can define the outcomes that they want and the agent will figure out the way to go make that happen.
[00:40:30] Speaker A: So what you're saying is going to be a behavioral thing. So if they're using it in their day to day life, it's going to start to be like, spill over into their work life to be like, hey, why aren't we just using AI? So even if they're not on board, they probably will be by default.
[00:40:42] Speaker C: Exactly. They're like, I do this at home, why can't I do this at work?
[00:40:45] Speaker A: Well, I guess time will tell. So just to conclude today, mj, any closing comments or final thoughts?
[00:40:51] Speaker C: I just encourage people to get out there and experiment. Go out and safely experiment, go spend some money on some of these models, go run some experiments. I think the more people utilize this, the more we're going to learn. I see this as this is a global phenomenon. We're all in this together.
And so I just encourage people not to fear the future, but to create the future. And the way we do that is we get out, we experiment, we share our knowledge and we learn and move forward together.
[00:41:22] Speaker A: And there you have it. This is KB on the go.
Stay tuned for more.
