This is the first installment of What's Next, a monthly conversation between SDxCentral CEO Matt Palmer and a senior-level executive from the technology industry. Each month, Matt has an informal but in-depth video chat with a fellow thought leader to uncover what the future holds for the enterprise  IT and telecom markets.

Matthew Palmer: Hi, I’m Matt Palmer, the founder and CEO of SDxCentral. Today I’m joined by Bob Friday, the chief AI officer at Juniper Networks. Bob and I have known each other forever. We’ve bonded over our mutual love of winemaking. I invited Bob to talk to us because he’s the founder of two awesome companies — one, Airespace, that Cisco acquired years ago, and the second one, Mist, that Juniper Networks acquired and of which Bob is now the chief AI officer.

Bob, thanks for making time today to join us for our first edition of the What's Next series here on SDxCentral. I'm super excited to talk today. One of the first things that I think will be interesting to touch on is your expertise, which is really about managing network elements and connectivity and a lot of other components within that. Can you talk a little bit about the paradigm shift from managing network elements to managing the client and the cloud experience?

Bob Friday: Hey, Matt. First of all, great to be here. You know my background. Twenty years ago, I had a company called Airespace. And 20 years ago, it was all about trying to help enterprises manage network elements, access points, switches and routers.

When I started this, what I saw was a paradigm shift where, in addition to managing network elements, customers wanted us to manage the client-to-cloud experience.

I had some very big customers tell me they were not going to put any critical applications onto the network until I could stop the controllers from crashing.

They wanted to make sure I could keep up with developing code monthly, weekly — not yearly. But more importantly, they wanted visibility. They wanted to make sure their real business-critical app or device on that network — they wanted to make sure their users were gonna have a good experience. So that's the first paradigm shift that was one of the inspirations for this.

It's more about this AIOps experience and keeping track of that client experience.

Palmer: It's interesting that you bring that up because I was chatting a couple of weeks ago with a friend of mine who's the head of network architecture at Salesforce. And one of his first comments was, “During COVID, everyone became a network experience expert.” He said it started with his wife and kids, who were expecting a network experience they never had before. He said he ran into it with all of the employees at Salesforce, too — people who never knew what a network was, were coming back and asking about the experience.

Friday: That's what we're seeing. With COVID, all of a sudden these problems ended at the house, right? And that's what we're starting to look at as people move into this remote-working environment. We’ve got people at work and we’ve got people at home, and trying to solve IT problems at home is not trivial, because the IT department doesn't own that infrastructure.

And that’s where combining cloud app data like Zoom with network feature data, we're starting to be able to help the IT guys with these problems that have moved into the home environment.

Palmer: How does that relate to AI (artificial intelligence)? Does AI help us solve that type of problem?

Friday: Well, this is an interesting thing, when you look at where AI is taking us. Because for years we've all been doing automation. And what I usually tell people is, “AI is just the evolution of automation.” We're just taking automation to the next level of doing tasks that typically require a human.

This is the next evolution of AI. Specific examples include Zoom, where I can take data from my Zoom, or all my Zoom users, and join it with my network data. I can build models now that can accurately predict your Zoom performance.

And once I can do that, I can start to predict what network feature is causing a problem. So that's an example where AI is letting us solve problems that were very hard in the past to solve. It's solving problems that would take a human to solve.

Palmer: That sounds like it's adding a new tool to the toolbox. Do you have a couple of other examples of these neat new tools?

Friday: I usually explain to people, if you look at what we've done as engineers over the last 20-30 years — we're all very familiar with ML (machine learning) regression. All those algorithms are still applicable. And they still solve a lot of problems.

The thing that is transforming and disrupting the networking spaces is the same thing that's disrupting medicine and cars: these deep-learning models.

For networking, it's about things like LSTM (long short-term memory), transformers and general AI. These models are starting to make things like anomaly detection solvable. We've been trying to solve anomaly detection for years. The problem is ARIMA (autoregressive integrated moving average) and all these other mechanisms and statistical analysis; there were still too many false positives. No IT guy wants to be woken up at three o'clock in the morning with a false positive.

Deep learning started to change that space — anomaly detection of low false positives.

And we’ve all seen ChatGPT. I don't have to tell you that after ChatGPT, the number of AI skeptics dropped in half. Everyone realizes this stuff is gonna make a difference.

In the networking space, you know what I tell people? We're going from the days of CLI (command-line interface), and then we took them into the dashboards. And we had dashboards trying to get rid of all the complexity of CLIs. The next big user interface in networking is going to be around these kinds of conversational interfaces.

This is gonna be the next user interface where you interact with your network in the future. Think Star Trek. This is gonna be more about “Hey, I want to talk to my network and I want to talk to my computer.” I don't want to have to go to the dashboard to find out what's going on.

Palmer: That makes total sense. As we talk about this, one of the other questions I have is the importance of data modeling and what your data models are.

What type of advice do you have for network practitioners? What do they need to do so they can have the correct data to be able to start to have conversations essentially with their network infrastructure?

Friday: People always ask me, “Hey Bob, why’d you build an access point?” I didn't build an access point because I thought the market needed another access point. I built the access point because I wanted to be able to make sure I could get the data I needed to answer questions like, “Why are you having a poor internet experience?” And I didn't trust Cisco and Aruba to give me that data.

So I usually tell people “When you're working, think telemetry.” I think this is another big transition that's happening in the networking space. It's how we're gonna get data — telemetry data — out of these networking devices. In the past, it's always been SNMP (simple network management protocol). We went and pulled data out.

As we move into this AI era, it's gonna be more about telemetry. You're gonna want your networking devices to be streaming the data you need back into your cloud for analysis.

An interesting experience here at Mist — once I got the cloud built and I got data streaming — is, in addition to having to build these pipelines, I had to get my data science team and my customer support team working hand in hand. So, in addition to all the architecture, there's an organizational structure that needs to change.

It took me a year to stop people from SSHing (secure shell protocol, or secure remote logging) into these devices. Because if you go to a cloud AI model, the data you need has to be in the cloud. So we're not going to be SSHing into it. That's probably the other journey to AI — you gotta figure out what data you need in the cloud, and you gotta stop your support team from SSHing into these devices.

Palmer: I'm smiling for two reasons. One is the old Pareto boxes refused to let anybody secure shell (SSH) into them. From 10 years ago! And then today, at SDxCentral, we built our own data collector to collect the data of what happens on the site for the exact reasons you just brought up, which is how can we standardize the data to understand what our audience is consuming? But also, how do we understand what ads or other things are doing across the site? Because unless you can do that telemetry, you're not going to know what's effective. And that's not effective.

Friday: I think that's going to be the next thing in the industry: How do we start to get everyone to some sort of standard telemetry? We want to move to more push models than pull models.

Palmer: That leads really to another interesting question. Regarding your comment about the AP, I remember everyone giving me a hard time at Pareto, saying “Why in the world did you build another router?”

And I said, “Well, Cisco and Juniper and everybody else isn't going to give me that data.” And that leads to a really interesting question about vendor-specific versus vendor-agnostic AIOps. I'd love to get your perspective on that.

Friday: There are two approaches to this. There are things like Datadog. You could try to build an AIOps solution, that works across all vendors. My perspective right now is the industry’s not ready for that yet. We do not have enough telemetry standards to get a vendor-agnostic thing to work across all vendors quite yet.

So I'm much more a fan of vendor-specific. And that's why I'm a fan of “Yes, I need to be able to control the AP, the switch and the router to make sure I can get the data that I need.” Because, like I said, it took me about a year once I figured out what the support team was doing.

There's a whole other step of actually going back and figuring out what data needs to be sent back, which means you have to go back to the device.

Where we are vendor-agnostic now is we all move to a cloud and we're starting to ingest data from all the cloud apps. The good news is you move into this paradigm of client to cloud. Most of our apps now are in the cloud, whether it's Teams, HR or Zscaler — all these apps are running in the cloud. They all have APIs.

So the apps have gotten a little bit farther ahead of us on the cloud side. They've all moved into an API paradigm.

Palmer: That makes a ton of sense. Well, I know we're close to running out of time, and you brought up ChatGPT earlier. I'd love to hear what your take is on where ChatGPT is gonna land in networking.

Friday: As I may have already mentioned once before, ChatGPT is making my life a lot easier. When I started this journey, in mid-2014, eight or nine years ago, everyone was an AI skeptic. That number has gone down by half.

The other piece of the puzzle is, I believe ChatGPT is going to become the user interface of choice for everything. We're going to be replacing our dashboards with these conversational interfaces going forward.

If you're a Star Trek fan, this is that talking computer. I think, besides that transporter thing, we're gonna get to this talking computer thing here shortly.

Palmer: That's cool. How close are you to a beta that we can see?

Friday: That beta is happening right now. We’ve integrated ChatGPT into Mavis now and we’re starting to supply ChatGPT Large Language Model (LLM) answers, along with docs. So you'll get some docs, but you’ll also start getting an answer.

Palmer: Well, I'm going to have to come see it this fall when we start to do some wine tasting.

Friday: You know where I live.

Palmer: Well, Bob, this has been a fantastic chat. I appreciate your time today and thank you for taking the time, and for really a fun conversation.

Friday: It's always fun to chat with you, Matt. I’m looking forward to the next one!