Transcript: "Fast Is Slow, and Slow Is Fast": Building Right in the AI Era | Amir Frankel (Zero Networks)

Opening: what occupies Amir most — AI in three directions

Eyal David: [00:00] Great, let's go, how are you? I'm excellent. Good to be here. Good that you're here. It was hot on the way too, right? We talked about it. Always.

Amir Frankel: [00:08] Summer has started. For you? For me it starts early, yeah. Tel Aviv, and then, high humidity in the air.

Eyal David: [00:15] And then fine, enjoying the beach. Enjoying it. That's it, you're very laid-back, we'll say that, you can't see what's underneath here, like Kushmaro, right? You can't see the bottom half with flip-flops and shorts. Come on, I'll go straight in, I won't introduce you too much. We'll basically put some show notes and say that you're basically the CTO of Zero, right? Right. I won't dive in beyond that. But I do want to take us back for a second to our conversation last week. The first question I asked you, beyond "hey, Amir, what's up, how hot is it and..." we basically talked about, okay, what's the thing that occupies you most right now, right? Indeed. So I'd love for you to take us from there, right from there.

What containment and micro-segmentation are — the building analogy

Amir Frankel: [00:56] So I won't be very surprising, and I'll say that like everyone, the thing that occupies me most is AI, but I'll say that in the context of Zero, AI occupies me in three different directions. The first direction is very similar to everyone else. I'm busy with how we adopt AI, in which parts of the organization, what it changes, which ways of working are suddenly being replaced. On one hand, where are the places where you say, wow, how did this junior do in an hour what I used to need a team of four seniors for? On the other hand, where am I not losing quality, not losing control, how do I make sure our organizational culture still stays. We'll come back to that, but I'll move on to the other pillars. The second thing that occupies me a lot is AI in the world of attack. AI has completely changed the cyber world in many ways, because AI gives attackers enormous power. If you think about it, for an attacker there's a super well-defined objective function, and another wonderful thing, they only need to succeed once. So now that we've put things like this in attackers' hands, they just manage to run wild with models, and suddenly an attack that used to take months, the record today stands at less than half a minute to do a full lateral movement in an organization. That's a super significant gap. Whereas for defenders, their objective function is that everything works and nobody breaks in, good luck defining that, and they need to succeed 100% of the time. I really don't see, I think defenders also got a contribution from AI, but they certainly didn't get the same order of magnitude of contribution. Happily, these things connect very well to Zero's approach from the start, which I'll talk about later. And the third part is that organizations are changing right now. Everyone's entire way of working is going somewhere else. You see it from the most old-fashioned offices in the world, how suddenly you hear about how a locksmith uses AI, all the way to giant organizations, and suddenly in this world you once again need new defense methods for understanding how people work, and mainly containment, because you're bringing into the organization, if until now we brought in deterministic things, suddenly something comes in that one day gets confused, and oops, a database disappeared, oops, all your files disappeared. And here too we look a lot at our advantages, because Zero provides containment, lets you close things off and deal with things, so that will connect for us later. Okay, so wait, I'll go back a bit, the locksmith story fascinated me. What is containment? Dumb question. Containment means saying that I'm able to close something off, to guarantee the separation inside the organization, and in the world of attacks, for years the approach to attacks was that you need to detect and prevent them. In the modern world, you're already saying, I'm no longer able to prevent every attack on my organization. Basically isolation into small units, and then... it's a guarantee that even if someone has already attacked your organization, they'll be limited. So that's why Zero, basically zero trust. Zero starts from Zero Networks, from the idea of basically separating the network. We're basically a micro-segmentation company, which is exactly about taking your network and dividing it into small components, but when I say micro, the idea is down to the level of the single computer. Every entity on the network should be treated as an independent segment where I know exactly who's talking to it and who it's talking to, and that creates part of the magic, that from the moment you say, okay, so they got into the building, but all the apartments are locked. Not so bad. Maybe they'll take some potted plant from the lobby, but that's a bit different from them taking the jewelry from your home. And once the building suddenly looks like this, more magic happens, like the thing that... if you're walking around a mall, who knows who's suspicious. You walk around your building, suddenly you see someone who isn't one of your neighbors, because there's order, some things start to be much stranger once someone wanders around, it's easier to detect things. But the world moved from attack prevention to assume breach, assume you've already been attacked, and to containment and uptime. For most organizations it's much worse to be down for ten minutes than for one computer in the organization to be a bit infected, maybe a bit of data got stolen. That's it, the ROI here is clear.

How AI entered Zero: from chatbots to Cursor and Claude Code

Eyal David: [04:59] Okay, so we'll come back to the building and everything. Let me take a step back for a second. So really, let's start simple. So what occupies you? First of all, how did AI come into your organization? How was it? When was it? Look, I'll say that from the beginning, the moment the chatbots came out,

Amir Frankel: [05:16] that was already the first stage of seeing where these things help and where they don't. I think the first leap was, I can say, from my angle as CTO, suddenly things I used to have to wait days for someone to answer me, or any complex SQL, or any complex question about areas I'm not strongest in. I started getting answers. At that time, as we all remember, the first years of AI, it's a bit funny to talk about something three years old as "the first years," but its pace of progress is so high, there was always a good kernel of truth, and then lots of hallucinations, lots of times when you come to someone, you do, you know, they told him he could. No, that doesn't exist, that command doesn't exist. At the beginning it was very much about helping people understand, you go, do the research, go bring the things. Those were the first stages. It also helped outside the software area, I usually talk a lot about what happens in Engineering, but in the end AI helps in all parts of the organization, whether it's marketing, sales, calls in general, or interviews. Where did you feel it most? At that time? At that time, mainly in giving some ability to do research much faster, getting all kinds of corners and things you can get better, that kind of thing, or empowering employees. Suddenly you see that the junior who was previously very dependent, or kept getting tangled every time he had to read some guide, suddenly gets like, oh, that's how you do it. There's some junior story hiding here that you've already repeated twice. There was something like that... I say again, in the end, the way of working at Zero always revolves around development and growth. That's also why we really like sometimes to hire people who are at a stage where we see a lot of potential, and to help them in the first stages with how to think, how to build things. Since it revolves around us a lot, and also our whole way of thinking always talks about what I'm doing today, and about the junior who walks in the door in another half a year. And that's why I'll invest a bit more in code review, and for that I'll say, ah, I understand it, but it doesn't help me that I understand, because it means I'm stuck here and I need to maintain it forever, or in the next problem, or the next big feature, it doesn't even have to be a problem, they'll keep pulling me in, but if I always thought about the junior, my cognitive effort is always very low, and that creates the thing that lets me keep bringing in more people and growing. The second thing is that that's simply where you see the biggest gaps, because if your strongest senior comes to you and tells you he solved a problem in half an hour instead of two hours thanks to AI, that's nice, but it's not the same level as when you suddenly say, wow, I can't believe you did that alone. Those are the things that make you see it. I also always go a bit to the extreme, I'll say, also people, also team leads who suddenly make leaps, also Mid-Level, also seniors, by the way, you see a productivity leap. You asked earlier how AI came into our company, so we started with the chatbots. The next stage was inside Engineering, starting to work with Cursor and Claude Code. Outside Engineering, a lot of note-taking, a lot of people, all kinds of departments, really use it in interviews, they use AI to summarize the interview, and little by little also to get to know the things that will help them decide what was successful, what was less successful, usually they don't use, of course they don't record the interview in order to protect the candidate's privacy, but yes, they take their notes, write them in, and after you train something a few times, suddenly you get... suddenly you have a criterion, and you can do... you suddenly see an interview summary you do, wow, what order was created here, really defined scores, a clear definition, what the pros and cons are, I just saw a candidate like that before, the answer was very nice, the answer was, he has no weak spots, no weaknesses, but there are things he's excellent at and things that can be improved, but they're good, so that was how... Amazing, yes, a great way, very helpful, I mean I assume everyone who interviews and hires knows the insecurity that comes with the process. So you're interviewing a lot now, you're saying. Yes, I'll say it's less in Engineering, we still don't really do this way of working with the notes that I described, but we mainly do it for remote roles. The next stage, as I said, was Claude Code and Cursor mainly to help me, more like a half assistant. Was that also basically your personal journey of adopting the tools? I'll say that in some things, I'd be happy if I had the time to do it in everything, I'll say that at those stages, yes. But there are things I'd be happy to have more time to do, and those are exactly the things we're constantly working on, how in the AI era you also clear the calendar. What does it really look like on the manager's side? So it depends on which level of manager you're talking about. I'll talk about you. Right, right. It's just important to me to clarify that with the middle managers in the organization I see the things they manage and have time to do along the way.

The manager's side: more time, more depth

Eyal David: [10:00] What happens to them at the end of the meeting he has with you... what for example do you see that really makes you happy? Examples of super complex problems they research,

Amir Frankel: [10:08] building big things, and also simply deeper involvement, because suddenly they can do, take a feature on themselves, usually, if you'd said five years ago, I'd explain to a manager, I'd tell him, always take tasks without urgency, without pressure, that can afford not to happen, because your time will get swallowed. And suddenly that changes completely, code review will also be easier because we did a first round of AI and it cleaned up a lot, a lot of your work. Also thanks to AI we allowed ourselves to write a lot more tests, which in the past we'd say, you won't keep up with those amounts, forget it. And also suddenly he does something, because you were dedicated only to planning, building a more defined architecture, the documentation, the architecture there was always very defined, but their documentation is improving. Where do you see it catching you? So for me it catches me all the time in seeing which areas I feel it works well in and which not. One of the things we always watch very closely, AI is an amazing thing, but it's not a substitute for ownership, it's not a substitute for your responsibility for what you put in.

Eyal David: [11:11] If we talk a bit, maybe the time will come for us to talk a little about the context of Zero, to clarify the daily challenges you... so yes, definitely, just tell me a bit about yourself for a second, I'd really love to hear. Where do you really feel it comes into play? You say, I'm able to see where it's good and where it's not good.

Amir Frankel: [11:26] How do you do that? So it goes back to how much faster tasks get done, what their quality is, how things look at the next stage.

The experiment: Cursor vs. Claude Code, and measuring quality after two months

Eyal David: [11:34] Because you just told me, I remember, about how you started running some experiment at the beginning, and you ran very fast, but then you looked back at how the quality was, so can you tell a bit about that? Sure, we basically did it from the moment we reached the stage where AI coding tools started working well,

Amir Frankel: [11:52] and the stage of Cursor and then Claude Code, and we let people decide what they want to use, because we really believe that in the end you need to let people lead a lot of the time, you need to steer as a manager, but you don't really need to tell them how. You want to tell them this is the goal, and how — go execute. One group chose Cursor, another group chose Claude Code, we let them run, and after two months we did a comparison. We saw the places where things improve, we saw the places where, say, too much code was created. We don't want to get to a situation where in the middle of the night when there's an incident, it depends on how many tokens I have, and whether there'll be enough, and then we started to feel whether there's a drop in quality, whether the amount of code went up, whether we lost some standards, and in certain cases we felt that, say, tasks came out too fast, came out really fast, but not good enough for Production-Grade. We constantly examine the state of the number of bugs opened by QA, because if I see that on one hand I'm creating speed at the stage on the first side, but not on the follow-up side, that gives me nothing, it even really connects to one of the company's values, which is Do It Right, fast is slow and slow is fast. We really believe it's always very important to measure the full move, the whole cycle. Because sometimes you can say, wow, look, this shortcut really shortens the first step, but afterwards you pay for it for years upon years. So those are exactly the places that are sensitive for us, and that's why we watch very closely, one of the things that's always important to us is, use AI all the time, but critique it, don't think that you wrote the code, look, do, one of the things we also really demand from this, we still require two reviewers, who's not good and the problem with two, but a lot of times some of the comments in the review will be, do a stronger review of yourself, because you also need to take other people's time into account, because AI will also very easily let you get into a situation of I'm running really, really fast, but who in the end absorbs all the pain? My reviewer, and then my tester, there are some amounts of discussions I've seen online about the... you're not supposed to bring his work, it's yours. The other things we dealt with are privacy, what you're allowed to upload and not allowed to upload, which engines to use, what to pass through the Cloud Provider, and of course in the things I talked about at the beginning, I talked a lot about AI adoption inside the company, of course there's also the, where AI goes into the product, in places where we put in a chatbot to help do things that in the past took lots of clicks, in natural language. So what catches me all the time is what I need to do organizationally, which healthy processes, how to make sure knowledge is shared in the AI world. We now have an AI guild that's constantly busy making sure we create good connections between the groups, that everything people do gets passed on. We have a few very nice stories about performance debugging, about how someone started from telling it, build me the whole test environment that simulates this thing, and then come tell me what, scatter logs in the code, then run, now look at the results, now come tell me what to improve, and after you've done this whole nice story, also fix it, go through it there. Or problems where really some of them, I say, if I looked with all my years of experience, those were problems where I was usually used to grabbing a few seniors, and they sit in circles, sometimes there's a flash of insight, sometimes it can take days, and suddenly the level at which it comes and does for you, look, in a third-party library you're using, in this specific version, there's a bug there in this case with that offset, we won't go into the depth of that bug of course, but the idea is clear. So all the things this creates, and making sure we don't have any security problems, we're a cyber company, it's very important to us to protect ourselves and our customers, we

Eyal David: [15:40] came to improve their quality of life. Okay, that's one of the things, we'll talk about it already, that there's no room for mistakes, and that's something that drives you, or there's a little room for mistakes. Really a little. That's it. I sort of heard a few things. One, how much ownership is a value in your company, right? Like, that's it, we talked about it too, right. At the level of the line of code, but it's much more.

Extreme Ownership and Zero's unique PR structure

Amir Frankel: [16:01] Right, in the end that's what we call ownership, of course we sometimes also want to free ourselves from the lowest level, but the way, how things are done, is very important to us. Low cognitive effort, as I said, that shows in the way we do, say, pull requests, and it's very unique. So today commits go in. One PR can be 40 commits, and each commit needs to be a short story. If you bring me a huge PR, you'll find that the cognitive effort for the reviewer is enormous, and in the end it closes here in the method we work with. I can have a first commit, I did a rename, a second commit, where I was with it infra-something, a third commit I did like this, a revision commit, this, this, this, that, and I'm always keeping my context, just as I want to do for AI, and therefore also for humans. And what that creates for me is that every time I read, usually not me, happily, because the organization grew and more people can do it, you suddenly see there's very strong context, it's really easy to ask the right questions, because you're not looking at everything together. In other things, we're always looking in the end, small startups, one of the nice things about them is that everyone feels like one unit from the very first second, and as it grows you need to preserve culture. I think we do it excellently, so ownership is very important to us, the fact that everyone feels responsible for what they do, and responsible for everything that happens in the company, what we call Extreme Ownership, that doesn't mean you're also responsible for everything, but it does mean you care how it happens, or as I always define it, it's not enough to pass the ball, you need to make sure someone caught it. You passed someone a bug in the middle of the night, call him, see that he's there, maybe he's just on vacation and you did nothing, you didn't transfer responsibility in the first place, but it got solved. That sounds amazing to me,

Eyal David: [17:37] it sounds to me a bit like an analogy to an ant colony or a beehive, right? That it's a factory, everyone has their own ownership.

Amir Frankel: [17:44] That's interesting, but I'd say that there you're taking me to a world of workers, and I want partners. One of the things that's very important to us is that we look all the way through at the fact that we're all together, we're in the same boat, that connects to more of our values, like transparency, that we try to tell employees everything. Everything we can tell, we will, because we want them to know, because they deserve it, they work hard. And a more important thing, that they give us feedback, and if everyone really feels part of it and understands the goal and the why, it's more fun for everyone, because I enjoy working much more when I have motivation and the goal is clear to me and I understand the why, I assume most people are like that, and also that there's a place where I can voice my opinion, and that's how you get this thing together, and then you really see people coming to you about a problem they heard about or ran into, they say, wait, I'm not sure it's like that, I saw this customer's problem, I know it was passed on, but I checked again to see what I can help with to move it forward, and that's how you also slowly break down walls between departments. It feels more, that's it, I also saw where basically you, where you're putting AI in, obviously, let's put code aside for a second, but really as another thinking assistant, I don't want to say it in English, so thinking assistant, and in those places of really analyzing,

Eyal David: [19:00] so a question, now these initiatives you called them, that basically come from the field, are you starting to feel more of them,

Amir Frankel: [19:08] basically, because there's more time, more tools to get to them? I feel it all the time, because happily we keep enthusiastic people,

Hiring: how to spot passion and the "why"

Eyal David: [19:15] who really love what they do. When you live in a world, when you do something you love, then you can do a lot more... How do you examine that when you hire? You talked a lot about hiring, so I believe it's a big deal for you. How do you know how to identify that someone who fits your criteria is coming in?

Amir Frankel: [19:33] So I'll say that first of all, it's always a combination of interviewers. How do you feel? What do you hear about the person? Each with their own interview. My interview is very different from the interviews that happen on the way to me. Usually all I want is for people to tell me how they got to be what they are today. Where did you start? Why? And why? And why? Because that, in my eyes, drives most things. What are you looking to hear in an interview like that? I want to know someone who really has enthusiasm for what they do, who tells me the extra.

Eyal David: [20:02] Who tells me why? It interests him a lot. Do you have some story you remember, from an interview that was really successful?

What Zero is — fixing the balance of power between attacker and defender

Amir Frankel: [20:10] I'll say I have lots of successful stories of people, and sometimes in the sense of why I went into a certain field. Signs of excellence, the person I'll take things from, even at the level of someone explaining to me why he went to take a retake exam even though he passed, because he simply felt he deserved more. People you see excelled in all kinds of things in life, you see that, say, someone was an individual athlete, that's even harder, I also really appreciate team athletes. When you see a swimmer, a tennis player or things like that, you say this person has a strong internal engine driving him. God forbid it should sound like we only hire athletes, we're open to everyone. You feel it in why he decides to switch roles, in where the impact is, in the stories a person tells, I experienced something inside the company that I felt I could help make better. Sometimes he even tells you a question, and then I sat down one weekend and said, I'll build this thing. I'll say that in today's world it's harder, and there too we really believe, and I'm always afraid of creating some wrong impression. We're not looking for people to work outside working hours, unless it's really important, but sometimes with people with passion, sometimes you come home, I remember it from the beginning of my career, I have to know why the DB behaved like that, or wow, will this improve performance or not improve performance? I have to run the test now. I'll just make the small change, I want to run it, and I'll watch TV in the background, but I have to see this thing, those are the signs of passion. People who basically come and say, wow, I'm in this profession because it's cool all the time. And one of the signs of that, you see AI adoption in the company. You see that people don't look at AI as, oh, they'll move my cheese, why do I need to work in a new way, but the opposite. How cool, look, I used to waste time on this, it was the part of my day I liked least. And now it's a lot more fun. Amazing, so we talked a bit about how you basically measure AI adoption at your company? We basically still deal less with actual measurement, because our first goal is, use it. We, because we're a product that... let's tell, let's talk a bit about what Zero is, and then we'll come back, it'll connect all kinds of things for us. Zero is a cyber company that came to fix the fight, the balance of power between attackers and defenders. We really feel the world has been deteriorating in favor of the attackers. We're a defensive cyber company, so of course it's clear which side we're on. And while at first it was very clear that you protect the single computer, because it was hard to get to it physically, then make sure of something different about bringing up networks, then let's put a firewall around the organization, then let's put detection measures everywhere, EDRs. Endpoint Detection and Response, something that basically sits on every endpoint computer and tries to detect things and respond. We feel that at some point this world created an enormous load on defenders, enormous amounts of alerts all day. Half the time you discover that the person next to you is investigating the same alert, you don't sleep well, you feel you're constantly being chased, being the CISO, the person responsible for information security in the organization, is one of the hardest roles, and by the way, just this week I saw an article about it, that serious burnout is starting there too, it'll probably be a field where in a while it'll be hard to find people. From there in the world you basically reach a situation where the balance of power kept deteriorating, we feel it's time to make a change, and our approach says you can't be reactive in defense, you have to be proactive in defense, you basically have to make sure that the moment something happens, you're already prepared for it, if we compare it, you put bars on your home so that when a burglar gets to the building, gets to your apartment, it says no. You keep a dog so it says there's proactive defense here, I don't do that, I also really believe in dogs, because it's one of the loveliest animals in the world, and there too it sometimes tends to the thing that a dog, often with burglars, the thing is that it often doesn't prevent the break-in, if someone's already come to break in, it makes the burglar say, you know what, let's go one building to the right, it's easier there, not worth messing with him. And for that, the first problem we came to solve is the micro-segmentation problem. That problem comes to say, your network should be completely closed, only those who need to talk to each other, and we're talking down to the computer level. What that means on one hand, once you've managed to adopt this thing, you have wonderful protection on the network. You've made a leap. The problem with it, like a lot of good problems, like a lot of things that sound too magical, is that you're in a hard problem. You need to keep track of everything on the corporate network. You need to know the IPs of everything, just because computers don't communicate with their names, they communicate with IPs. You need to understand why everything happens, because you need to collect, you need to understand which application talks to whom and why, and then you need to start collecting a lot of knowledge about which user, which process, and you basically start getting enormous amounts of information that you need to analyze. And then comes the hardest part of the problem, and it has two parts: one, you always have to work. Because I'm basically affecting how a customer's corporate network works. And your network is a really critical part, it's like in... if we compare it to our home, it's like the electricity. That's what Benny, my partner, always says, and think that someone comes to your home, replaces your electrical grid, when will you be sure, when will you be calm about how it works. So one thing is to guarantee quality, guarantee real time, because the network is a moving thing, you constantly switched to Wi-Fi, got home, you're working from the VPN, and if at that stage you constantly feel that things work slowly, unfortunately Zero won't be there, but happily we always work fast, we invest a lot of effort in it, and the next thing is what needs to be open. Because in the end you have a jungle of applications and things that were built over years, and if you ask most people inside the organization, even someone who was there from day one, nobody knows anymore what's right and what's not right. And here comes our big advantage, the big gap we brought, the news we brought to micro-segmentation, a field that already existed for quite a few years, was, let's make it easy. We want simple things, and the experience we wish for our customer is like Apple, simply segment. Come on, open, there's one very important decision you'll make, buy Zero, and from there your problems are our problems. We talked about proactive versus reactive, you'll sleep much better at night, because you know there's no problem. And if an alert also comes on the weekend, look, you'll look, you'll say, ah, that'll wait for Sunday, I know I'm protected. That the damage, that you can't do big damage. And here comes the part of, one, it's really easy to install us, our installation time stands at a quarter of an hour, 14 minutes to be honest, from the start of a meeting until we're already really starting to work. That's thanks to our agentless approach, meaning we don't install something on most computers. The second thing is our automation, we learn the network for you. Which means that instead of needing lots of people, you simply have something that works in a snap, we did these things before the AI era, and let's say that here we talked earlier about the need for precision in the world of defense, we need to be precise, and AI's small mistakes are either production down or too open. I don't, I believe that in the future AI will also be able to help with that, and we also use internal analysis a lot. So wait,

Near-zero churn and the third advantage: network-level MFA

Eyal David: [27:32] we just threw out a lot of information, and I want us to put some order for a second. So first of all, I think the building analogy is good, it's great. I was actually thinking of a water flow and not electricity, but that works too. You understand what's important to you, the shower is more important to you. Water quality, exactly, water quality and that, given that no, in the war I think we discovered that. Tell me, I'm really curious, but what's the average duration of a customer with you? What does it look like? It sounds like, once you're in at a customer,

Amir Frankel: [28:03] like, now... Look, I can tell you that you're very, very right. Our churn usually comes only from one company buying another company, and then usually what happened is that the company that bought it... there's a father who commanded how... no, I mean, really, one of our stories is that one of our churns is a customer that was bought by a bigger bank, and now the bigger bank is our customer. I'll say that actually, if you measure us on these things, we almost never lose customers, you can count them on less than... one hand is still more than the number of customers we've lost. It was usually from external aspects, it was never a bad product. Because we're very righteous, you get an excellent experience. As you said, it's transparent, and with your numbers, like, again, in the end,

Eyal David: [28:49] a bit like insurance, right? As long as it doesn't catch me and you show me attacks that were blocked, like we haven't talked about our third advantage. Go ahead.

Amir Frankel: [28:58] Our third advantage is that we believe that every time a person is involved in the process, we can't give them open access. If I left the IT guy's computer open to the whole network, the project is a waste. You built a really nice gate at the entrance, but there's no fence. So why did we build the gate? MFA says that every time someone tries to make network access, a person involved in the process, we'll pop up MFA, it'll pop up on your phone or on your computer and say identify yourself, say it's you, we'll open it for you. And that makes people also experience us day to day, they feel the granularity, not just like... that's what I meant by blocked, because when an inspector comes and sends me the message, it's pretty much the same thing, and then I feel part of the process. That's passive, I'm offering you something active here. You know that you just accessed some resource in the organization, and then something nice comes to you, and you say, how nice, I feel this is secure. And one of the magic things about a network is that it's one of the only things in the world where if you closed it, you really closed it, nothing will happen, no new zero-day will show up tomorrow. You don't need to chase all the alerts. Now Anthropic's Mythos came along, exposed enormous amounts of zero-days. For most of our customers, well, for all of our customers, a lot of that isn't relevant, because those ports are closed on their computer anyway. It doesn't matter what you can do. So it'll send a message, nobody will answer it. Curious, did you do something marketing-wise around it? Yes, we're always doing around the... classic. Of AI, we have lots of things around Mythos. We also in general, I talked earlier about the different forms in which AI meets them. So in the form of attackers who suddenly advanced, there's basically the ability here to be much calmer. What's interesting is that some of the things that existed from the start of the company, there's even an old article of mine, Stop Chasing CVEs. CVEs are the definition of exposed security vulnerabilities, which have a rating and explain what to do and which patches you need to install. And in a world where you have something like Zero that says, most of it is closed anyway, nobody can get to talk to this computer. You don't now need to clear the weekend to do maintenance on the weekend because you're not allowed to do it during working hours. You simply say, nobody will get there, it's closed. So AI now only made this situation worse when it created two interesting things. One, there are lots of CVEs where you used to need a crazy team of researchers, and now you run something, pay a bit of money, wait a few hours and it all happens. And on the other side, it now lets you, at that very moment when you're now, no, I'm calm. Both vulnerabilities that exist, and vulnerabilities that were known and were hard to exploit and it exploited them fast, and ones that aren't known yet. Once I moved to a proactive approach, where I'm not chasing, what's the problem? I fix the method. One of the things we also talk about a lot is this, I'll say it in English, it's hard to translate, Solve the root cause, not the symptoms. It's exactly like in the world of medicine, a lot of times you need to solve the root of the problem, not the symptoms. We feel that a lot of times when we didn't manage to find the root cause or treating the problem itself was hard, we started treating around it.

Stop chasing vulnerabilities — fix the system (root cause)

Eyal David: [31:57] And once you've solved the root cause, suddenly all the things close up. You can also say that about what you basically did with the company at the end of the day, right? That you basically went with this approach of a closed garden. Did you mean what we do in micro-segmentation? No, no, that's exactly what I'm trying to say.

Amir Frankel: [32:12] Exactly. That's the root cause. The real root cause of most network attacks is that your network is too open and you took something that was yours.

Eyal David: [32:19] In the end you built this network for your own needs, an attacker comes, and it's a playground for him. Totally, I'm saying, but the very fact that you took this approach at the start of the road also made your lives so much easier, that you told me the story, for example, of what you do. It was so clear, okay? I'm saying, and the fact that you get to a situation where you install the thing in 14 minutes,

Amir Frankel: [32:39] like, it sounds like magic. In everything we do, simplicity really interests us. I hear you. We believe that big problems are hard to solve, but a big problem that you've broken into lots of small problems, you enjoy every part along the way and suddenly the puzzle comes together. And we're really looking for how to do things right, as I talked about, and the part of "we came to solve your problem, not to help you, not to improve by percentages, we didn't come to improve the cyber world by a few percent." We came to create a paradigm shift. To come and say, defenders deserve their power, they deserve for it to be easy, and it can't be that we keep transferring more load and more load onto them. Now, we said, what's the hardest problem that will solve the most things? Or more precisely, what is the hard problem that created this? I'd have been happy with, what's the easiest problem that will solve the most things? But it was more what's the problem that will solve the most things, and then think about how you can really make it work.

Eyal David: [33:31] Take us, take us to the future, take us forward. So I'll say like, in which sense?

The future: AI is producing the new legacy software

Amir Frankel: [33:38] In the sense of what Zero looks like? In the sense of Zero, but come take us to the security world. So the security world is going through a lot of changes now and you see it from lots of angles. I didn't detail everything Zero does, we also have a lot of work in the identity world, one of the things we're unique in is that we connect the network world and the identity world. Usually you have companies that are network companies, companies that are identity. Also when we talked earlier about MFA, we basically brought identity to the network world, which usually doesn't have it, I know how to identify you, that you performed an action, and usually it's I need to decide. That computer is open or closed, there's no middle. So we believe there will continue to be consolidation, we also have more protections in different forms, and what we're looking at is that AI takes us back a bit in certain things, in terms of the quality of software,

Eyal David: [34:24] in terms of the things it could do. One of the things written down for me somewhere on the side from all kinds of notes you write on some evening when you sit and think, is that AI today is producing the Legacy Software of 20 years ago, that back then at first Zero was built to protect all the things you can't control, because you have old software you can no longer see updates for, operating systems you're not allowed to change because the FDA will require new approval for this device, so meanwhile they need to work with Windows 7 and Windows XP and things where some of the people probably already working on it were born after the product reached end of life. And then suddenly AI produces new things like that, because suddenly someone who isn't a programmer, and he had the whole defense team around it, he did lots of things, he did a security review for you, and he checked here and maybe ran automated pentests, but suddenly it turns out he missed a small detail. And here Zero comes again with the ability to do containment, because you want containment against the attacker, what we said about closing things off, I should have found a better word for containment in Hebrew, I'll find out next time, and surely a minute after we finish it'll pop into my head, isolation probably. So you need the same thing for AI software. Now basically on one hand, in a period where you say there's an unbelievable productivity leap in the things you can do, like you're doing, right,

Amir Frankel: [35:44] wow. It's reminiscent in many ways of what happened in the leaps of the open source and cloud native world, suddenly the number of packages that came out, and then a stage also comes where you need to start protecting my packages. The same thing is happening to me now with skills and with MCPs, and with the fact that the AI only wanted to help, so it thought it would be really nice if it went to another server it discovered on its own, and then you discover it destroyed all your critical payroll files or whatever it is. So we see a big change in how businesses work, in what's needed. In Zero's world, for now we're experiencing that we still need a lot of manpower, or we were built on a small workforce, so we're really not one of the companies looking at how we'll shrink, but maybe we'll slow growth in the future, but for now we keep growing, we're still managing to see how we're able to do more, and we're managing to shorten timelines, and we didn't lay anyone off, and we're really not in that kind of experience. In the business world we expect to see a change in the type of people working in everything. We expect to see... What do you mean? Suddenly, if the world started with generalists once, if you look at the hi-tech world, say 30 years ago, everyone was generalists, they knew how to do hardware, connect jumpers and set up drivers. And then little by little we moved to specialization, we said someone needs to know how to do front, and someone does back, and this one's good at DB, and that one's good there. And then AI suddenly brought back, suddenly a lot of the need for builders, if we're already connecting to the podcast, people who came to build, who understand more parts. That doesn't remove the need for expertise in certain things, because wow, a design eye helps. If you really understand DB in depth, you'll know how to fix its mistakes too. And in other worlds, suddenly what used to take time moves more and more to thinking. If until now, say, there were worlds where what people sold was their time, people will need to sell, they'll sell their ability more. In many cases this will make a significant improvement to the experience. I can say that, say, our analysts, the leaps they got thanks to AI, before too they had a lot to do and why and thinking about the ideas, but then they'd do a long stage of manual grunt work, or work with scripts they built here and built there, and suddenly it's like... no, it did all of that for me. I see things like the fact that, I'll take Benny, my partner, at some point he wanted to test some concept about all kinds of things related to data and the data we collect and to know how to understand it. He came back after a week with some application he runs on his computer, that shows him graphs of things and lets him do catalogs and consult with AI, and ask a question to a few AIs and get an answer. Now, I really appreciate Benny and Benny is a very technical person, but a programmer he's not. He wasn't a programmer, he started his career once as a programmer and it was a short time. But hey. But hey, suddenly he comes, and the first thing that happened when I saw what he created on the computer was, okay, we have to get this visualization into the product fast.

Eyal David: [38:43] And another super fast thing we do is also, say, what happens to product managers in this world? Suddenly there's, I focus a lot on the hi-tech world, but I think the change is happening everywhere. Feel free to talk about another one.

Amir Frankel: [38:54] Yes, I'm saying, if you look at the law firm, once the intern would do the research, he'd sit for hours on amounts of data, now you come, say, first let AI do it, at most the intern will go over it before it goes to the partner. In our finance world we suddenly see what you can do with AI's help inside the Excel sheets, inside this presentation, inside that calculation, inside BI metrics, how you remove grunt work and don't need to remove employees, but can move forward faster. You can do more things, you can get better results. But, as I talked earlier about ownership, always check it, because if in the end the company is missing a million dollars because it was with it, it won't be, ah, no, no, Claude. So that's it, I hope that in finance they actually do know how to do it, but... I expect that everywhere. I want to ask you, but a question around that actually from a security perspective,

The business: customer types and the cost of an attack

Eyal David: [39:50] all these companies basically protect enterprise. Right. What's the story? We don't protect enterprise, we protect everything. Okay, tell me about that.

Amir Frankel: [39:57] We, we have SMB customers who come through MSPs. An MSP is the Managed Security Service Provider, someone who basically manages all the cyber in the organization for you, and it suits organizations of even ten to twenty heads, employees. And then basically these customers, again, they come through these hubs, so to speak, and such. They come only through the MSP. Yes, exactly, great partnerships, okay. We have SMB, we also have customers of 100 and 300 and we also expect them to come through Channel. We have mid-market customers, which is exactly the stage between 500 and 5000, and we have customers of 5000 and up, which is already enterprise,

Eyal David: [40:30] and the beautiful part is that all of this happens on exactly the same platform. The beautiful part is your business, but I'll put that aside for a second, but whoever wasn't lucky and isn't your customer basically during this period, and is an SMB.

Amir Frankel: [40:43] He's not a customer yet. Yet, you're right, but...

Eyal David: [40:48] But really, like, what chance is there? In the end you have to answer the question of what you're going to build, what you need to do. You basically described to me that on the attackers' side,

Amir Frankel: [41:00] they have super powers, superpowers. Right. In the end attackers, in the end, also need to choose their targets. In the end the world is statistical. How much does an attack like that cost, just so we understand the order of magnitude of something like that? For a large organization, in the best case, only tens of millions. Tens of millions. Well, it probably would cost much more. Here, I have to mention Land Rover, which, say, not long ago had production lines shut down, that's already passing a hundred million dollars. If you think about it, what happens with McDonald's down for one day? No transactions. Well, how much money is hiding there, and we're not even talking about the recovery, about how much needs to be rebuilt. In the Rover example it was actually ransomware that took over the whole factory. The factory stopped, it took days to fix it, and an attack usually, if you're a large organization whose uptime is critical, the amounts are huge sums. Are there segments you identify that don't turn to security solutions of one kind or another? With us we really experience a very unique variety of customers. Like, that's what I'm asking. You have a law firm and insurance offices, and insurance companies and airlines. Companies that make tractors. I can give you a few, probably every segment, in a surprising segment, factories making car parts. We really have government offices.

Eyal David: [42:22] Wow, I now understand how much it actually helped you in sales, and surely in sales it's crazy. All of this was before AI. Basically,

Amir Frankel: [42:29] the field we're in was on the side for many years, because the world assumed it's a problem that can't be solved. And precisely now the field is going through a significant change, I'm also talking about micro-segmentation. As I passed to Zero, there are many other things too, but we'll leave the listeners some room to go discover on their own. The micro-segmentation world is the thing with the most queries today, the most inquiries to Gartner today are about micro-segmentation, and it's one of the things that's also slowly now most getting into the Magic Quadrant and other things, because the world, from our perspective, understood that you need to make the shift from visibility and detection to micro-segmentation and MFA. So all these things, our first customers were customers who were good early adopters, customers who also managed to deal with the fact that the funniest problem we had at the beginning was that customers would come, sometimes hear the pitch, no way this is real, and keep walking. The thing that it can't be that within 30 days you'll learn my network. No way this will work. I remember Benny coming out of all the sales calls and doing yes, I bet him a restaurant, 30 days later, he said, yes, I won a meal from the customer, how fun for me. Benny also sounds like a very good salesperson. Benny is really a combination of many things from many directions, on one hand he's a very technological person, really loves what he does and really understands, he also grew up in the IT world with a very strong understanding, and on the other hand, when you really believe in what you do, you sell it amazingly, and when you love people and care, and if you look, we talked a bit about, say, transparency or things like that in the company, we really care about the people we work with, and in the end, the people are very important to us. We want to be together, we want to come to the office and everyone to enjoy, and to see people smiling, and we really love the friendships that form in the office, we usually in a normal year also bring the whole company to Israel once a year, you really feel, how nice, suddenly someone, two people met in 3D, suddenly afterwards what was a two-week stream of Slack messages, three updates and twenty minutes, something will already move forward. So this togetherness is very important to us, so Benny really loves people, connects really well with people, and he's a good salesperson, and believes in something that works.

Eyal David: [44:38] I hear, it also sounds like you really believe in the human angle basically around this thing, I hear you saying it throughout the whole conversation.

Amir Frankel: [44:46] In the end we really believe, I'll also give Benny's opinion here, if you believe in what you do and in the why, and you care, and you don't look at people as a tool, but you enjoy the friends you made along the way, and the things that happened, and seeing the growth. I talked a lot about juniors, because I really enjoy seeing the journey people go through. How they grew inside the company, how their way of thinking improved, or how it converges onto bigger things, and also what we learned from them along the way.

Advice for juniors: get hands-on, understand why, bring passion

Eyal David: [45:13] I'm glad you said that, so I'll ask a question toward the end, a sweet one. Juniors. Yes. What wrote, you really, it sounds, a, that it really burns in you, b, it sounds like you work with a lot of juniors, you hire a lot.

Amir Frankel: [45:28] What's the best advice you can give a junior today? So first of all, learn things, get your hands dirty, and try to get to a state where you're not trying to make something work. Try to get to a state where you know why something works, where you enjoyed it, and do only what you love. If you're a junior and you don't love programming, save yourself the round of a few years, or find another role in hi-tech to your taste. The PM world is going to grow significantly thanks to AI, or find something that really does you good in life. Come to things with passion, learn extra, do projects for yourself if you see you're not succeeding. I'll say, maybe I told you, I think, in one of the interviews, that just half a year ago we hired a whole team of juniors, we were surprised to see what we manage to do with them thanks to them and thanks to us, about their fast onboarding, and what they delivered in less than three months. As I also told you in the interview, when we talked in the past, that in the first month we don't allow them to write with AI, we only allow them to read with AI. We really encourage them to ask questions, to understand what's going on, and we want them to still learn Zero. And that creates exactly this stage for us where they understand the way, learn the effort, feel how you work with git properly at Zero, through the cognitive effort of the review, and also when they later get to guide the AI, because in the end that's what they do,

Product in the new world — and the human in the loop

Eyal David: [46:54] they'll know what they're talking about. So there, you just connected the thing I wanted to ask you, I'll stretch it out so you can drink. Tell me, so you basically said, and I hear opinions this way and that, so you basically take the take that actually the PM world, this role is going to grow. Do you want to expand on that? Like, like you touched on it just now, but do you want to expand on it? Yes, in the end the new world lets anyone be a builder.

Amir Frankel: [47:18] An entrepreneur is also a PM on steroids. The entrepreneur is basically the one who comes and says, I have an idea, and I don't just want to define it and understand where it meets the market, I'm willing to bet everything. I'm now going from getting the money, through fighting for it, through dealing with the fact that the responsibility is always yours, and when we're talking about companies with size, you have responsibility for a lot of people, and we talked about the fact that we're not the kind of people who know how to look at it lightly, but we really care. The PM world suddenly can, we talked about the experiment Benny did, Benny is originally a product person, so it's very important a lot of times to see things. We didn't talk about the fact that, say, at our company the PMs, these days, write Chrome extensions to see how things look, not so we'll use them in the end, these Chrome extensions simply shorten the cycle, so that suddenly the PM knows on real data how something worked, tested it in depth, and then he suddenly comes to engineering and it's one shot. There's a lot of the thing that in the end all of us, what do we do with AI? We give it a requirement, give it acceptance criteria, we explain to it how it fits with the ecosystem, we ask it to assemble and we'll give it data, very, very similar to the work of a PM and an engineer. A lot of times the PM gives the essence and the direction, and then where it works, the engineer comes and does back-and-forth, and product engineering is also always very important to us in the company, we always say, you need to know why you're building something, if you understood, you don't agree, keep discussing with the PM. Maybe you have a good idea, of course you need to keep lines at the point where you can't interview all the customers,

Eyal David: [48:48] but then in the new world, suddenly people build, they're the components, that's being a PM. There are two environments here that kind of interest me. On one hand, you say, there are a lot more initiatives you need to think about prioritizing, on the other hand, there's also a lot of precision work that needs to be done. So my question basically, I understand that this, because I can understand that direction, but do you think it's acceptable when AI also decides?

Amir Frankel: [49:15] When AI also knows how to decide? I think we're not there in everything, and if we talked earlier about Zero's world, my quality is critical, my up-time is critical, and I can't tell customers, yes, we wanted to move a bit faster, so the AI broke your network. That's also part of the problem with models that don't also contain deterministic protection, that's why it's a big part of the advantage of our learning. I think there are areas where it can do good work. I think you still need to know how to keep the human in the loop, unless you've managed to define an objective function. If you managed to define something where you say, as long as it's inside this sandbox and these are the things that need to be, that's excellent. In the end, the AI world now produces amounts of things on one hand, on the other hand it also starts to spoil things. If once someone wrote a very interesting text, it was interesting, now you sort of already start to feel you're flooded with texts. The video world, where at first the first people who created video were exciting, and then there were amounts of video before AI, and then AI simply caused you to lose focus and you go like, okay. I feel it's not real. There's this joke that there are lots of people who come every evening and build a whole product in a night, and then they discover at the end of the night that they need to talk to people to sell it, and then they say, okay, never mind, tomorrow I'll start a new project. So you need to see where AI fits. We're constantly working now on seeing where we can bring in agents, where it can help, but it's very important to us that in the last action at the end of the process there's a human in the loop. If we talked earlier about I built something, up to production, it won't go to production on its own. If I want it to debug the bug, I want it to find the solution, I want it to propose a PR to me, I don't want it to do the merge. I still want to make sure I understand what's going on there, so as to be ready also for a situation where one of the AI companies is down, also for a situation where the cost is infinite, and also simply I don't have time to wait for it. Sometimes a person who looks at the area he knows does it faster. So there's this combination of the things you need to be careful about with AI, and we also feel now that the amount of things being built, but then we sometimes lose some of the magic, or it's really easy to start, if I talked about fast is slow and slow is fast, then it's classic. It's really easy to start a POC, really easy to start a story. People often talk about how in the world of literature, starting stories is easy, the hard part is connecting the whole story and in the end coming out with one thing. I think that shipping in the end is the thing you strive for,

Eyal David: [51:46] that in the end comes out, so I agree. It's not just the shipping, it's also the precision and keeping a coherent story.

Amir Frankel: [51:53] Over time you start on lots and lots of tracks, and then the question is whether they keep playing together. And if you have a slingshot product where this thing breaks the other part, or every experience behaves differently, and what tries to save money clashes with what's important to you, and in the end you discover you're sleeping in a tent instead of the place you wanted to get to, instead of a resort, just because you wanted to save a dollar, that's a bit of a bummer. So I'm saying, it's a magical world, I haven't seen a revolution like this since I came to the industry and I've seen a lot of disruption over the periods, I've seen how it happens in every field. This is the biggest revolution I've seen. It contains enormous amounts of opportunities and things that are also simply fun to see. It simply contains risks of, one, being careful, understanding. Not everyone can build something for themselves now that will replace, I hear that all the time about the big companies. We also don't even fully estimate the costs correctly. Sometimes you say, someone gets excited that he saved a manual action he does in three minutes, but he doesn't know that in another half year this thing will cost him 15 dollars. Exactly. This misunderstanding, that you need to know how to use it correctly, you need to deal with it a lot. The world moved from prompt engineering to context engineering, we're very busy with that in the company. I talked about the AI guild, it's a lot of pushing around how we make sure the amount of tokens we spend is right. Not because we're now bothered by the waste, because in the end it's supposed to pay that back in time or in manpower, but because over time it'll get to the point where it gets lost. There are of course enormous amounts of research talking about all kinds of interesting methods and what you need to do in AI, how there are basically approaches of training the model on a small context back and forth and only then continuing to give work to smaller models. And in the end, on one hand something showed up here that does everything, on the other hand in the world of security, risk announced, we hear it from lots of customers. On the other hand they're very happy about Zero's AI segmentation, about our ability to help you make sure the AI won't do the damage.

Eyal David: [53:52] And that's exactly what we deal with. As you asked me at the beginning, we talked about what bothers me, exactly this. Wow, a great closing to the episode, really. I think that's it, there was another bit of best practices that we talked about in our conversation, and I heard them, and we didn't manage to fit them in, and that's okay. Best practices in AI that you've seen, and that, maybe a shame. But you'll come back to us again, and we'll do it. How do people get in touch with you?

Amir Frankel: [54:13] Shall we give LinkedIn? LinkedIn is excellent. Great. Best, and of course we have jobs on the website, people are welcome to apply, and in the end we're always happy to hear about good people who want to come to us, about customers who are interested, and about collaborations, because in the end the world is always a world of people.

Eyal David: [54:35] You've just been listening to the episode, let's go. Amir, it was great, thank you very much. Thank you very much, let's go. Bye bye.