Guy Nadivi: Welcome everyone. My name is Guy Nadivi and I’m the host of Intelligent Automation Radio. Our guest on today’s episode is Bob Friday, Vice President, CTO and Co-Founder of Mist Systems, a Juniper Company. Now for those not familiar, Mist Systems is in the business of providing wireless networks with AI built in, to provide automated incident remediation. We haven’t spoken much on this podcast about that kind of intelligence being baked into an enterprise’s network, so we decided to bring Bob onto the show to gain further insights on how AI-powered networks might be part of an AIops-driven future. Bob, welcome to Intelligent Automation Radio.
Bob Friday: Guy, thank you for having me. This is a topic dear to my heart, so happy to be here.
Guy Nadivi: Bob, let’s start then with asking you if you could just share with us a bit about what path you took that led you to co-found Mist Systems and the AI work that you do.
Bob Friday: For me personally, this path to Mist and AI really started almost all the way back into, for people who remember, the eighties. When the FCC first came out with this unlicensed spectrum rules, that’s when I really got into the wireless space. For those who remember back then, I was doing Metricom Ricochet. This is building a nationwide packet wide network, and that’s where I really got my first taste of wireless. From there I really went off and co-founded a company called Airespace and that was back in the early 2000’s. And for those who remember back then, that’s when wifi was just becoming, going from a nice-to-have to a must-have. And that’s when I started really working with enterprise customers and really started trying to help them figure out how to manage these wireless networks that were coming into the enterprise space. From there, I actually sold that company to Cisco and it was really at Cisco where I became the CTO of mobility that I really started working with some very large enterprise customers. And there we saw wifi and wireless kind of go from a nice-to-have to must-have to really becoming a business critical. And that’s where I really saw the paradigm shift from enterprise customers wanting help managing these network elements, to where they really wanted help managing the end-to-end user experience. It wasn’t good enough to tell them that the EP or the switch is up and running, but they really wanted to know if they were going to put an app on that consumer device, that they can ensure that that consumer really had great internet connectivity. And that’s how I got into AI. You know, from there, it happened to be the convergence with AI was becoming popular. We finally had the technology. When I started Mist back in 2014, that’s really when AI kind of went from a marketing thing to a reality thing. We really had the compute and storage that we could actually use to solve interesting problems. So that’s how I got to Mist, that’s how I got to AI and that’s how I got to here today.
Guy Nadivi: 2020 was a challenging year for so many organizations. Bob, when it comes to AI for IT, what are the biggest lessons that enterprises learned in 2020?
Bob Friday: Like I said, I think one of the big lessons that they’ve learned in 2020 is that really AI is really becoming more than just marketing hype. I think for a lot of IT departments and enterprise businesses, AI has been kind of a marketing thing and not a reality thing. And I think when we look back on why is AI becoming real now? I think it really kind of started, like I said, back in the 2014. 20 years ago when I did my masters, I actually did neural networks and masters. But 20 years ago, the problem was we really couldn’t build neural networks that were big enough to do interesting things. Somewhere around 2014, we got this perfect storm of compute storage costs going low enough, datasets getting big enough, open stores, where we find we’re able to start building AI that can actually solve real problems. And I think that that is one of the things that they learned in 2020. I think the other thing they’ve learned is really a difference between AI versus ML. We’ve been using ML and machine learning to solve problems throughout most of my engineering career. Really in 2020, people have started to learn that AI is really doing something on par with a human. Whether it’s learning to drive a car, interpreting a medical MRI or x-ray, and really in networking it’s really about can we really build something that can do something on par with a network domain expert? And that was one of the inspirations for Mist was really for those who remember Watson playing Jeopardy. When I started Mist, when I saw Watson playing Jeopardy, I was like, if they can build something that can play championship-level Jeopardy, we really should be able to build something that can actually play networking Jeopardy. Do something on par with real network domain experts.
Guy Nadivi: Okay. So with 2020 thankfully behind us, can you talk about the role of AI and AIOps in the future of work?
Bob Friday: Yeah. So I think what we’re going to see in the future of enterprise IT work, we’re going to start to see these AI assistants actually start to become part of the IT team, right? And so I think what we’re going to start to see is IT administrators and businesses start to free up their IT teams to do more strategic things in the business. Case in point is right now we’ve got to the point where we can actually build these systems that can actually detect bad ethernet cables. That’s a very hard thing for a person to go detect. That’s an easy thing for an AI assistant with machine learning to find bad ethernet cables. So now you don’t have your IT team busy trying to basically go track down that bad cable. You can have AI assistants join the team. So I think what we’re going to start to see coming forward is really around IT starting to adopt these AI assistants onto their team as kind of a trusted member and start bringing that and start training those AI assistants like an employee coming on to their team.
Guy Nadivi: There are different shades of AIOps, but Bob, can you explain what the difference is between domain agnostic and domain specific AIOps?
Bob Friday: Guy, I think you always hear people talk about kind of narrow AI or general AI. And usually when they use those terms, they’re talking about kind of “Terminator AI”. And I think most people agree today, most AI is narrow. We’re teaching AI to drive cars, interpret medical records, images. When we talk about domain agnostic, domain specific, specifically we’re talking about in the IT space about different platforms that were designed specifically to solve a specific IT problem. And when you look at AI, it really starts with a question. You got to kind of answer, what questions do you want your AI assistants? So like when I started Mist, the questions we wanted to answer was really around why is the user having a poor internet experience? Why is the user having a poor connectivity experience? And that is really around trying to figure out what data you need. So when you think about domain specific, it’s really starting with that specific question or that specific problem. Domain agnostic is more about we’re going to take a platform, a generic platform, and try to train it to actually solve a problem. And right now I think if you look at the Gartner who kind of came up with these terms, the general consensus is that most enterprise businesses will get to an ROI quicker if they start with a domain specific platform versus a domain agnostic platform. When you look at having to solve these AI problems, there’s a lot more than just putting the data into the platform. It turns out, after doing this for the last six years, a lot of work goes right into the feature engineering. Even after you know the question you want to answer, you may spend months, weeks, making sure you have the right features you need to solve the problem. And that’s one reason when I started Mist, that’s the reason why I built my own access point. It wasn’t because I thought that the world needed another access point. It’s because I wanted to make sure I could actually get the data I need to actually answer that specific question of why that user experience is poor. Why are they having a poor internet experience? So when you look at domain agnostics and domain specific right now, I think most enterprise businesses are going to find that they will get to a quicker ROI with the domain specific platform where all the feature engineering has been done for them. Where the data has been cleaned up, they don’t have to worry about preprocessing the data. There is a specific solution that’s actually processed that data and has that solution up and running to answer a very specific question in cloud. As opposed to kind of do it yourself. So you kind of break it down to domain agnostic is kind of the do it yourself approach. Domain specific is you’re actually finding a solution, it actually solves a specific problem that needs to be solved.
Guy Nadivi: What market transitions, if any, have you seen Bob that are driving business and IT to adopt AIOps?
Bob Friday: As we kind of started the discussion earlier, when I started Mist, I was the Mobility CTO of Cisco and I was working for some very large retail customers. Some very large enterprises customers. And back then I was building these controllers with these kind of embedded software architectures. And what I heard from these customers was, you know Bob, before I put any of your stuff on the network, I need to make sure that your controllers don’t crash. I really need to make sure that you can keep up with my mobile development cycle. They were developing mobile code every week, right? The mobile developers had these very agile development environments. Whereas we were releasing code maybe once or twice a year type thing. And then as I mentioned before, they were really wanting to make sure they had end-to-end visibility. So I think one of the big market transitions is really going from this networking becoming business critical, where they were putting some sort of critical applications, whether it’s a robot in a distribution center or a consumer app on top of a consumer device, it was really around going to that business critical, networking was becoming business critical at that point of view. So that was one of the key market transitions is business going from that paradigm of managing network elements to really managing the end-to-end user experience.
Guy Nadivi: What are the biggest motivators and barriers that you’re seeing for AI adoption?
Bob Friday: I think as we mentioned, the biggest motivator is really these networks are becoming very much more complex. I mean the other big transition we’re seeing out in networking right now is watching the workflows move from on-prem, behind the firewall to public clouds. So if you look at Salesforce, Microsoft Office 365, we have people in workflows working from home now in addition to working from the office. We have workflows scattered all the way across from the private data center inside the enterprise to the AWS, the Googles and the Azures out there. So that is one of the biggest motivators is really how do you handle the complexity? And we’ve gotten to a point now where when we transitioned from CLIs to dashboards, dashboards were kind of the one way to deal with that complexity. Now to the point where there’s just too many dashboards. We’re getting to the point where a network IT person cannot deal with the amount of information and log files that need to be dealt with. And that is why we’re starting to see them move to more of these AI assistants. Because AI assistants and these conversational interfaces are what’s really helping our IT departments be able to get the data quicker. Instead of having to remember the hundred different dashboards you need to find something, you can now go to your AI assistants and basically just simply ask, “Please tell me, why do I have unhappy users right now?” And the AI assistant can do the work of basically aggregating the data necessary to answer that question. So that is kind of the motivator. The barriers is really around, as we mentioned before, the adoption of AI. When you look at these AI assistants, whether you head down the domain agnostic path, there’s barriers there, and that’s why it’s hard to get to the ROI quickly. If you do it yourself, you find that there’s a lot of feature engineering. Getting the data you need to answer the question is a barrier in itself. And that’s why we’re seeing more enterprises move towards these domain specific, where the domain expert is actually helping them bring a solution to this table that can actually solve an immediate problem. So I think from a motivator point of view, it’s complexity that’s driving people to adopt these AI solutions. From a barrier point of view, it’s basically a knowledge base. We’re starting to ask our IT departments, we’re asking a lot out of the enterprise IT department nowadays. First of all, we asked them to move from the CLI paradigm into these dashboard paradigms. And then we’re asking them to become Python programmers. We built all these cloud APIs for them to help them automate and get data out of their networks quicker and easier. And now we’re asking our enterprise IT departments to start to wrap their heads around the data science and the AIs. We’re asking them to become a little bit of a data science expert enough so they can evaluate all the different options out there. So that’s probably the biggest barrier right now is this knowledge. Bringing our enterprise IT departments and educating them around the different data science options that are out there to help them solve their problems.
Guy Nadivi: Bob, with 2020 behind us, what do you expect AI for networking adoption will look like in 2021?
Bob Friday: I think the one word I would use here is acceleration. If anything, what we’ve learned over the last hundred years is the adoption of technology seems to be accelerating faster and faster. And I think this is going to be totally true with what we’re seeing with the adoption of AI. It’s becoming very clear that AI is going to be valuable in helping enterprise businesses basically deal with these complex networks going forward. I think we’re going to see the adoption accelerate. We’re going to see the technology definitely accelerate. We’re definitely seeing AI become, we’re in that kind of exponential hockey point of view of the AI adoption thing. That started back in 2014. Every year we’re starting to see more and more AI solutions show up in the marketplace. We’re starting to see more and more open source AI solutions on top of which we can build. So this is becoming easier and easier for startups like Mist to actually add value, because we’re building on the shoulders of giants right now. The mountain of AI open source code is just accelerating faster and faster every year, making it easier for us to actually bring value to customers. So acceleration is the word I would stick with for 2021.
Guy Nadivi: So with adoption of AI accelerating, as you’re seeing, what’s at risk for companies who don’t develop an AI for IT strategy in the coming year?
Bob Friday: I mean, interestingly, I think there’s two risks. There’s kind of the ultimate customer experience. And this is what we’re seeing with our big B2C customer, hospitality, retail. Anywhere where there’s a business-to-consumer experience, that’s the first thing. The risk is, hey, if you’re in that business and you’re providing experiences to your consumer or your employees, you are going to need AI to start to manage this end-to-end client-to-cloud connectivity experience. The more implicit type of risk is it’s really subtle that unless you’ve been doing it is really what I call the “vendor to customer support”. The interesting thing as we start to move to a cloud AI paradigm, your big, large networking vendors, they now actually have the data to do much more proactive, even on your support models, your networking support models. And this is the one thing I learned at Mist interestingly is what we did organizationally. There was the technical, architectural issues of building real-time pipelines for AI, but there’s actually an organizational component here really around combining customer support with the engineering team at Mist. And that was the key to success, to really bringing a new support model into the enterprise. Where we as a vendor now can actually help our enterprise customers proactively send broken hardware now. If there’s a broken piece of hardware or software in a network, the customer doesn’t have to send us an RMA ticket, we know it. So we can be very proactive on helping, saying, “Hey, we know there’s a broken AP out there, a broken switch out there. There’s one in the mail for you.” That’s a total paradigm shift of what they’ve had to deal with in the past of arguing with their vendors about networking problems. Which usually turns into multi-day, multi-week discussions about sending log files back and forth to each other. So that’s probably the other big thing that we’re going to start to see around the risks for companies who don’t start to really wrap their arms around, and their heads, and start to internalize where AI can help their businesses.
Guy Nadivi: Interesting. Staying with 2021, what are your biggest 2021 AI for IT predictions that you’re most excited about?
Bob Friday: Yeah, for me personally right now, my big focus in prediction for 2020 is trust. I think it’s become, I think people are starting to become aware of AI/ML. You know, they’re starting to understand that AI/ML is more than just marketing hype. They’re starting to see it actually solve real problems that are relevant to their businesses. I think 2021 is a year of trust. How does an AI assistant earn the trust of the IT department to be a trusted member of that team? So when I look at 2021 and AI, where we are in the journey right now, it’s about conversational interfaces and bringing that AI assistant into AI as a trusted member into the IT team.
Guy Nadivi: Bob, for the CEOs, CTOs and other IT executives listening in, what is the one big must have piece of advice you’d like them to take away from our discussion with regards to implementing AIOps at their organization?
Bob Friday: As the saying goes, every journey starts with the first step. So my words of wisdom for a CIO, a CTO who hasn’t started the journey is to take that first step. It seems daunting sometimes, and probably that first step really starts around the question and the data. If you’re just starting this journey, first start with the question you want to be answered. What do you want to leverage AI to do? What human characteristic, what human task do you really want AI to take on? And that’s back to the point of really what is the difference between AI and ML? And I try to highlight to people sometimes, AI is really about building solutions and software that actually does something on par with the human. So for that first step is really thinking about what are you asking AI to do on par with human? For me personally at Mist, it was really about building a solution that really was on par with networking IT domain experts. Can we build a solution that really can answer questions and manage networks on par of network domain experts? So I would say that’s the first step I would recommend to CIOs, CTOs is really look at the question, what human behavior are you trying to mimic in your business that you think AI can help you with, and then move on to the data. Once you’ve got that figured out then start working on the data, making sure you have the data that kind of answer that question.
Guy Nadivi: Interesting. A new way of thinking for IT executives. All right. Looks like that’s all the time we have for on this episode of Intelligent Automation Radio. Bob, it’s very interesting to hear about AI-powered networks when we mostly only hear about AI as an add-on to an existing network. So I’m very much looking forward to seeing how your approach to AIOps plays out in the near future. Thank you so much for coming onto the podcast.
Bob Friday: Guy, thank you for having me and it’s been an honor.
Guy Nadivi: Bob Friday, Vice President, CTO and Co-Founder of Mist Systems, a Juniper Company. Thank you for listening everyone. And remember, don’t hesitate, automate.
