Riverbed Advances Autonomous IT Operations

Hey guys, thanks for the throw. We're here with Dave Dontelli. He's the CEO of Riverbed, and we're having a little chat about the future of autonomous IT operations and the digital workforce, and how this stuff is all coming together.

Dave, welcome to the show. Hi, Mike. Great to see you again.

You guys had a significant launch last month, and I guess I'm hoping you'll put some context around this thing, but it feels like there are at least six different elements to this. So how does it all come together? Sure.

It's been great. We've been talking to each other since I joined Riverbed three years ago, and I'd like to say this announcement is a four-year overnight success. And the reason why I say that, it's actually our third generation of AI we've announced in the observability space.

And each generation, as you can imagine, has built upon itself. Riverbed was the first to ship generative AI in our space. Last year, we announced agentic AI, and we're now moving to really a whole new vision, which is around autonomous operations.

The simple way to talk about autonomous operations is that the whole space was invented to improve employee experience. To prevent problems from happening or to fix them very quickly when they occur. And the whole idea behind autonomy, and it's enabled by AI, is the idea of really fixing problems and taking the human intervention out as much as humanly possible to do with the technology today.

So it's very exciting. I think a lot of people are trying to wrap their heads around the fact that IT operations are basically meant to be done the same way every time, and the outcome has got to be reliable. And then we have gen AI, which is probabilistic, and a lot of the times, in fact, hardly ever is the thing done the same way twice.

So how do you envision all these things being melded together in a way that IT operations folks will trust? Well, the point you brought up is a key point, and I'm glad you brought that up is, as you said, we've moved to a probabilistic world from a deterministic world. And I think most of the general public out there hasn't really come to grips with that.

In the deterministic world, one plus one always equaled two. In our new world, there's a high probability that one plus one equals two, but not necessarily the answer. In fact, we see with generative, if you look at Gartner numbers, about a 25% error rate.

And that's why this announcement that we made in our second generation of agentic is so important because what we learned in the first generation is if you just leave it to an engine, you're going to have errors, and they will approach each problem like it's the first time they ever saw it. And what we've been able to add in our latest generation is an important word, which is context. So by using the MCP protocol, the ability to link generative engines, and we're agnostic, so we use whatever engine the customer feels like using, but we can link that with our knowledge and our context, our deep understanding of how endpoints, networks, and applications perform, and all the knowledge we gathered over the years in terms of knowing where problems are and how to approach them.

And by combining those two things together, you can take what were 75% success rates up into the 90s. That's what we are seeing. And again, it's a combination of agentic skills plus context is what gives you that opportunity to have more trust.

Now, this doesn't supersede your previous investments in AI, which are a lot of machine learning algorithms and things that are a little more deterministic. So as we go forward, is the role of Riverbed evolving, where you're going to provide a lot of different AI models and capabilities, but I, as the IT team, I don't have to master all that. I can just use it through you.

Yeah. And there's a whole lot to unpack from what you just said there, because everything you said is true, and it's actually pretty broad. So if you look at it, some of the prior conversations you and I have had over the years, those products we announced in the AI sphere are getting huge utilization.

So now we're approaching about 400 million AI operations that our customers have done very successfully. So they have automated a lot of their environment already, and that's something we're very proud of. We're past-- You would always worry about with these new techs is too much hype and not enough reality.

And our customers doing 400 million of these automations have seen a lot of great benefit from them. But as you said, also, each customer is different and is going down a different journey. So what we are providing with them is a whole tool set that they can adopt as their needs evolve or what direction they want to move in.

And I'll give you a couple examples. There was a whole discussion in our industry about employee self-service. And from Riverbed's point of view, we think that's not the right outcome, and I'll explain why.

Because if you look at trying to make the employees now interact with AI and fix things themselves, you're pretty much making your entire company a helpdesk. And I think most folks who work want to do their job. They want to have a good experience at their job, and they don't want to be spending their time as a part-time PC or endpoint repair person.

We think what this generation really does and what we're doing with this announcement through autonomous is allowing a lot of really good things to happen at once. One, we take the employee out of the equation, so we automate the problem so they don't have to deal with it. The best problems are the one you don't have to deal with.

But two, we still give IT the things they require. The ability to have an overall view of their entire domain, the ability to control what's happening. We deal with a lot of very regulated customers, banks, financial institutions, governments.

And the thought of them just having agentic agents running freely and changing their whole environment, no one knowing, is not very practical. So we give them the benefit of agentic, but also that centralized management that's very important. And the other thing is what you see in this generation is it gets a lot easier because now, some of the advantage of AI is now we can use natural language interfaces.

So a natural language interface, you can just simply ask questions, as you know, and it get answers back. It also helps with making these automations we talked about, all these automations, the 400 million I mentioned. Now you can design your own, because many of these large institutions want to customize to their requirements.

So they can do that now in a natural language, so it's much simpler and faster for them to implement. And then I think the really cool thing about this as well is it's able to speak the language you care about. And what I mean by that is if you're a deep technician, you're going to want to talk in deep technical terms, right?

Don't waste my time. Give me the information I need. If you're an executive and you just want to know, hey, how are things running in the Southern United States for me?

You want to talk in executive terms. Just give me a summary. Red light, green light, give me a dashboard.

And in that now, you can do that in however you communicate. So, a lot of our customers have Teams or Zoom. You can speak to our software now through those interfaces.

If you're someone who uses Slack, you can use Slack. So the whole idea here is have the machines do more, but do that in a way that the customer still retains the control that's required, so they can decide what they want the machines to do or not do, and then speak to them in the language they're most comfortable with, deep technical or more high level. And that's really what we call our third generation that we announced.

What will be the impact of all these AI agents that people are letting loose in these IT environments? We might be in a scenario where, I don't know, there's 50 AI agents for every human. Let's just pick a number for grins.

And if I'm the IT team that's supposed to manage all this stuff, I have a bunch of AI agents that are trying to accomplish a task, and they don't really have a sense of right or wrong, so they're going to go and do whatever they need to do by any means necessary. And how do I, as an IT team, cope with that? So I'll give you a two-step answer to that question.

So one of the things about AI, first of all, is about accuracy, as we discussed earlier in this discussion. But let me tell you some of the things we do uniquely, and then I'll talk about how we deal with now agentic agents floating around your environment. The first thing is, another analyst likened what we announced to the iPhone of these types of solutions, and I'll explain why.

We cover not just endpoints, not just phones and tablets and desktops. We also cover applications, and we cover networks. So we give you the full view.

You don't run a product to have a network. You run a product to have an application. You're interacting with an application.

So to have the most accurate answer when we talk about probability, you need that total view. Is this a problem on the physical hardware I'm running on? Is it a problem with connectivity to my network, or is it a problem with the application itself?

So first of all, Riverbed uniquely can look at all of those things. And because we can do all that, we also were able to introduce a new product as part of this announcement called AI Assurance. And what AI Assurance does is give end users or the IT department a view to what's happening with AI in their enterprise.

So that's one of the things that's very interesting about the observability business, is that as a new technology is invented, think like zero trust networks, which we've discussed in the past, or now agentic AI. In our industry, we then have to develop a corresponding solution to enable people to understand what it's doing in their environment. So we have a product called NPM Plus, as an example, that was really started about how do we observe network performance management in a zero-trust environment.

With AI Assurance, what we're able to do now is understand all the AI that's deployed in your environment. And as we know, and I've talked to a lot of customers about this, I'm sure you've talked to them about it as well, people have concerns. There is what I call the approved AI, generative engines they can use, and then rogue engines.

People bring in their own stuff that's not certified by the corporation. One of the things that AI Assurance enables you to do is understand what's deployed where. So very, very important.

Second thing it allows you to do is understand who's using what. So, the big rage these days is talking about token pricing. And I can tell you from our own environment, because we use AI to run our business and develop products, you have certain users who are costing you a lot of money, certain users who aren't costing you enough money, and people want to know who's who.

Sometimes that's a good thing, sometimes that's not a good thing, but at least they need to be able to manage that. And the third thing is, as you develop agentic applications, you want to know where these agents are and what they're doing. And that's the third thing of our AI Assurance suite, is that we can understand what agents you have deployed, where they're going, and what they're doing in your environment.

So it gives you that single pane of dashboard that everybody always talks about of what do I have deployed in my environment, both authorized and unauthorized, who's using it and how much, and what are my agentic agents doing? So we think it's a very exciting product, and the customer reaction to this has been really, really enthusiastic because they need a solution like this in order to deal with the AI that they're already deploying. As we move into this era, will we revisit the way the internal IT team is structured?

And I ask the question because we're clearly managing both human and non-human end users at a level of scale that was unprecedented. And historically, we've had these kind of server teams, networking teams, endpoint teams, and that may work For a company with a couple of thousand end users, but if there are now tens of thousands of quote-unquote end users, do we need another approach? I think it's definitely going to evolve.

And I'll tell you the first place we're seeing it is even in our own development here, the products we make. And I'll switch that to then overall management, but on the products we make, product managers, as you recall, historically, would spend a lot of time speccing out products, and they'd do it all on paper. " Now what we see is the product manager, if you look at things like Cloud and everything, they'll develop their own prototype.

Here's the product. Go build this. Right?

And then the engineers just quickly... There's a difference between prototype and an enterprise-ready product, and the engineers then take over and debate, and then build. So that's clearly changed the way engineering management now works.

And to your point, I believe the same thing is going to happen about overall IT management, right? Even if you look at our product, when we talk about iPhone-ing it, it just means the fact that iPhone combined phone, camera, and iPod, we've now combined application network endpoint. And so that allows you to have more of a universal view versus just a very deep gra-- We give you the deep granularity, but you can do that from a more centralized place.

And it's not going to happen overnight, but we will see more and more change as people deploy more. What's the level of maturity that you're seeing around the phrase observability? And I ask the question because we all monitor stuff for years, and it was a set of predefined metrics, and everybody kind of waited to see if the box lights were green, and they were green, and it meant, hopefully, things are working.

But with observability, it seems like we are moving to the point now where we can do root cause analysis quickly. So do people get that, understand that, and does monitoring essentially gets subsumed by observability? That's the path we're on, but it's a journey.

And the reason why I'd say this is that if you go talk to most large organizations, and this is whether they're a commercial bank or a government, everybody will tell you they have too many tools doing monitoring, too many tools doing observability, and it's too challenging for them to bring all that together. So the big thing out there is everybody's doing some form of tool consolidation in that they want fewer tools that see more of their environment and make things more manageable. And that trend started several years ago, and it continues to gain steam and will gain steam.

The challenge with all that, and this is the challenge of our industry, is that when you look around the industry, those tools were invented for a reason. Everybody collects something. And you want as much of that data as possible to give you the most accurate AI answer.

So, in our case, you see the fact that we're lucky. We started with an APM background. We had APM individual products, we had NPM products, and then we had endpoint products.

Our journey over the last several years in this AI journey has been to combine all these into one. So again, one product that covers more. And I think that trend is going to continue at least for the next five years.

Because again, people need a broader view without all these point tools, because the point tools are just too hard to manage. I also think that the way we view all this stuff is changing, and I ask this question because so many IT teams have dashboards, and we've been trying to get to that single pane of glass for as long as anybody can remember. But seems to me in the age of AI, that's changing now, where whenever there's an issue or an incident, the dashboard just gets spun up in real time for that event, and then that's giving me the visibility that I need that's more relevant than just some sort of dashboard that I've been kind of staring at for the last six months.

Couldn't agree more. And I think one of the things we have with this launch, and it is what I call mass customization. So meaning that I've been doing this a long time.

I developed a ton of enterprise products. And I can tell you, we used to have the interface teams work to build the ultimate interface. Everybody's chased single pane of glass for decades now.

But what AI really enables is really what I call commoditization of the interface. So all of us work differently. All of us want to see different types of data.

You now, as you're aware, with AI, you can have any view you want. And the fundamental thing is the data itself. So as I mentioned, with us, there's really two important things with that.

One, the fact we collect, again, across apps, networks, and endpoints. Two, we have our data store, which you and I have discussed in the past. And what the data store enables us to do is, in a scalable way, collect all that data.

And by collecting all that data, that's the important part now. It used to be years and years of arguments about who had the most elegant user interface and the easiest user interface. Everybody's going to have that.

The people who are going to have the solution the customers really want are the ability to access the data required to give them the answers they want. And we're there today. You literally can access our data with any number of different interface types.

It's whatever you're most comfortable with and whatever you think enables you to do your job better. There are no shortage of vendors talking about observability and everything that goes with it. From your perspective, what's that thing that IT leaders should be looking for that will make a difference in terms of who they select or decide to partner with?

Yeah, I think it gets down to, I'd start fundamentally with real-world results, right? What's really worked, and who's got the track record of doing that? So I think that's a big factor.

I think scalability is a big factor because, again, data is only growing, as we're aware. So there's going to be more and more data, and you need a solution that can scale very high. And then, as I mentioned, the big trend in the industry is they want fewer things that cover more of the territory.

Like how do they reduce? A customer I was recently talking to had over 200 observability tools. And their kind of interim goal, this is why I keep saying it's a journey, their interim goal is to get down to 50.

Which would be a huge change for them. You think about it just on a practical basis, right? Doing renewals across 200 tools, keeping those tools up to date.

This is a ton of work. So for them, going from 200 to reduces a ton of work and allows them to focus more on the success of their users. So that's a big criteria, I think, is ability, how many tools can you replace with your tool?

All right, folks. Well, you heard it here. Everybody's talking the AI game, but ultimately, it still comes down to where that rubber meets the road.

Hey, Dave, thanks for being on the show. Great to see you, Mike. Good to see you again.

Thank you. All right. And back to you guys in the studio.