Network Intelligence is Changing the Internet with Avi Freedman, CEO at Kentik

(Intro Narrator at 00:00:00) Today, we're talking to Avi Freedman, CEO at Kentik, about the major leaps they're making in network observability. Thank you to US Cloud for being our podcast takeover for this quarter. To learn more about how you can save on Microsoft support, listen to the end of the episode or go to uscloud.com today. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:28) Give me a quick update. I think it's been about a year since we last had a conversation. What's going on? Has Kentik made any pivots? What are you up to today?

(Avi Freedman at 00:00:37) So we are leveraging all the amazing awesomeness in new technology to continue the journey from where we started. We let people bring all the data together and let people answer all the questions they need. That was, like, you know, five years ago. And then we've added a lot of workflows to help people do the things they routinely do. And now the generative AI accessibility has made it possible to tell people what they need to ask and fill out those workflows and sort of accelerate them. So that's been really exciting for us and helps a lot because historically, our practitioners would say, "Oh, but, well, I could, you know, this outage I figured out in 30 seconds," or whatever. And then the executive would say, "Well, you know, but why?" And they said, "Oh, well, data this and in context this and whatever." And they'd hear, "Well, why don't you just use, you know, why don't you figure it out with pencil and paper or whatever?" But when you can show people things and do their, you know, help them be 10x more productive, we find that executives and other teams help, or we can let app teams self-service and understand what's going on the network and democratize this. So it's been really cool. And I'm really excited because, you know, the technology we can leverage to bring to our customers is continuing to evolve at a rapid pace.

(Joel Beasley at 00:01:59) And so is that in the network intelligence space?

(Avi Freedman at 00:02:01) Yes. It's network intelligence. But, you know, underneath, it's all the same stuff. You know, we started doing the traffic analysis, which is the hardest, hairiest, nerdiest stuff. And then added performance testing. And now we're doing the broadest stuff, which is, is it up or down? Like, the metrics. Is it up or down or full? So bringing all that together, again, you know, the AI and network intelligence helps us do that even better. But, you know, just linking it together has been what we've—that I would say for most of the last year, just having all those products, launching NMS, the metrics, bringing it all together in the platform has been great. But, you know, the pace has picked up even more and the outcomes for more of the organization, you know, that's been a big change even over the last three or six months.

(Joel Beasley at 00:02:45) Yeah. I think last time we talked, network observability was the big buzzword. Right? Has that changed? Well, it's changed a little bit, but was AI the reason why it changed? Like, has AI kind of taken over the observability? Are we just having robots observe us now?

(Avi Freedman at 00:03:03) I think that if you think about—let me come back to your question a little bit roundabout. Are you worried about being replaced by a computer with this podcast?

(Joel Beasley at 00:03:17) Oh, yeah.

(Avi Freedman at 00:03:18) I'm not. I think that AI is gonna be augmenting us all, and those that don't learn how to use this augmented technology to be the best humans possible, whether it's, you know, overseeing fleets of your minions or whatever, are gonna be behind others. Right? Those companies that don't get with that. I'm not as worried about Skynet and AGI. I guess as a creator, maybe it's a little bit different. But, you know, like, running—

(Joel Beasley at 00:03:47) Along the way. And me being worried is kind of like, what's—how much does Joel worry? I don't know. I see the evolution of it to be like an air traffic controller. Like, I'll be in charge of approving topics.

(Avi Freedman at 00:04:00) Exactly.

(Joel Beasley at 00:04:00) It can abstract in its own way, but nonetheless, like, I can absolutely see there's some podcast out there that are just AI podcasts, where the companies—the brands will build the show, and then they'll have the AIs talk to each other and actually host a podcast, and then they release it as if it's like a real person podcast.

(Avi Freedman at 00:04:20) You know, the stuff that I've seen on YouTube must not be at that level because it's generally some not creepy AI voice reading what is clearly within, you know, five paragraphs of words, some text with some videos flashing along, which I just look at it and I said, "This is not, you know, interesting to me." But—

(Joel Beasley at 00:04:38) No. They—it fooled me. So I mean, yeah. And I'm a broadcaster. So somebody sent me a clip of it, and they're like, "Hey, check this out." And I was like, "What is that?" And they're like, "It's—we created it with AI." And I was like, "No way, dude. No way." It was really good. But I don't know how much work went in their preproduction and postproduction or how granular they got. But the point is the writing's on the wall. So my question really is I've got 20, 30 years left in my career. Right? So I'm looking at it, and I'm like, "All right. Things are gonna change drastically." But I'm an adaptable person, Avi. I just find a way or make a way.

(Avi Freedman at 00:05:13) Yeah. That makes sense. And I think that, you know, back to your question about observability, if you can't—to run modern infrastructure, which is very complex, you know, people changing the infrastructure beneath the applications, the users doing different things, observability was about having all the data in one place so that humans could ask questions of it when things go south. And AI lets us augment ourselves so that we can look at more that for humans would be nonsense to monitor a million things. Right? Automatically go ask the same questions that humans would ask and, you know, as best as it can. Surface the thing, do the debugging workflows and saying, "This got shut down and that didn't cause a performance problem. Okay. I'm not even gonna tell the human." Or, "Here's this thing that I think actually happened and it's clearly an application problem. Let me get the right people involved." And to do that, you still need observability. You still need all the relevant data making sense in terms of what the applications and users and all that are. But what the advances in technology have let us do is augment ourselves and in some sense, not completely automate, but automate that first level of the boring stuff or even, you know, accelerate detection of things that would have been noise. So all the things that you needed to do to get the telemetry, make sense of it, have it all together in one place are needed. And, ultimately, the humans are still needed. I don't know as much about applications. You know, Kentik is a big application. But on the network side, AI is not in the next couple years even at the point of autonomous operations. Cisco and Juniper have been promising self-driving networks and, you know, bots to run everything, and it's still basically Python scripts with CLI commands embedded in them. You know, things are so different and unique in all these different kinds of infrastructures that AI can make recommendations. But, you know, even just, you know, the suggestions about what to do about things, we're still not there. But detecting things, using the data that you have, doing some of the same workflows that you would—like, you know, I'm sure in podcasting, editing, and things like that, I understand are, you know, really advanced with AI. Those are the things which are really exciting and, you know, to be able to deliver to people. And that requires all the data, but also these newer tools. Right? Traditional tools like ML, but also, you know, the generative stuff is constantly surprising, but also it's still like intelligent 12-year-olds, not like, you know, trained professionals, you know, in the field yet. We're getting close, though.

(Joel Beasley at 00:08:00) Yeah. Well, at least in media, we're getting close. So in the past two years, I've seen more progress in the past two years than the previous 15 years. Like, it is going—I mean, it's my full-time job keeping up with this stuff, and I can't even keep up with it.

(Avi Freedman at 00:08:16) Yeah. You have to look at podcasts written by AI to track everything. Yeah. I mean, you know, there's a big difference, which is that there's some general network understanding in the models that are out there. But to actually understand if you think about the—I'll try, I know this is a nerdy audience, but, you know, sort of the—I'll try to, you know, if you think about the amount of semantic encoding that represents the uniqueness of someone's infrastructure and then God help you add the Internet in. Most of the people have not been training, you know, training for understanding that kind of architecture, engineering, operations, bugs, release notes. You know, there's very few people that are training on release notes of all the bugs. Right? You know, going back from the nineties when you would stick a card in and the protocol would go down or flames would—you turn on a protocol and flames would spit out the serial port. I mean, it's like, you get all sorts of strange things. And so certain kind of defined things on the infrastructure operation side, the general, you know, generally trained AI stuff is pretty good at. And if we say, "Hey, watch what users are doing in Kentik in a privacy-respecting way, but like, the workflows they do, and then adapt what you're trying to do debugging-wise," they're really good at that. But the actual, like, closed loop do everything for people on the infrastructure side. And I think on the application operation side, although that's not my—you know, that's not where I'm into, I think that might be—it's weird. I didn't realize that might be farther along than podcasting because you think that creating interesting compelling content for people that sounds natural would be harder than running the digital infrastructure, because it's humans to humans. But I guess you blew my mind.

(Joel Beasley at 00:10:01) Have you checked out Replit?

(Avi Freedman at 00:10:04) I have not. I've heard of it.

(Joel Beasley at 00:10:07) That's something—yeah. You'll hear a bunch of people talking about—if I were you—

(Avi Freedman at 00:10:10) I've heard about vibe coding. I've heard—

(Joel Beasley at 00:10:12) Yeah. I would spend—if you—you will get 80% of it in 30 minutes. So if you just watched a YouTube video of someone doing something with Replit for like 30 minutes, and you just watch them go from start, you can go from prompt to deployed application in one sitting. Look at the screen there. You just hit go, and it actually builds everything and deploys it to a URL. It's kind of creepy, but it's really cool. But, yeah, it's making technology more accessible, and I think we're gonna end up getting really great tools because you might have a network engineer even within your own company that's like, you know, they're not a software engineer, you know, necessarily, but they know the network stuff, but then they can go build a piece of software to help them do something with just basic instruction.

(Avi Freedman at 00:10:59) From what I've seen, looking—I haven't looked at Replit, but I've looked at some of the other coding tools. Because Replit, I thought, was more on the consumer or sort of web app side than some of the back—you know, build back-end tool framework. But, again, I haven't used it, so, you know, maybe I'm wrong. But you've got to—to make sense. Or let me leave it this way. I have destroyed the Internet. Not on purpose. Okay? A number of my friends have destroyed the Internet. I mean, you know, for like 30 minutes to three hours at a time. Right? You could type one line. Again, I don't wanna—I don't know what level we're operating at, but there's two kinds of routing protocols. There's internal routing protocols generally assume you can know everything and share the state of everything and run Dijkstra. More actually now incremental Dijkstra. But, basically, they try to figure out everything, and it's a big computation. And then the Internet uses something called distance vector protocols where you share a little bit of state, and then everyone sort of distributedly figures out less of it. If you cross the bridges and connect one to the other and send the Internet into your internal routing protocols, you basically cause every device in your infrastructure to get a never-converging computation. You have to turn it all off and on again. And the things that you can do to configure devices to cause things like that are very subtle. Even humans mess it up. And, you know, understanding why something is and how things are connected—like, what I've seen network engineers successfully do is say, "I have a thing I want to do" and be pretty prescriptive about that. So maybe we're just talking at different levels because we automate the whole life of magically run this infrastructure is very different than deploy, consistent, interface names, you know, on a—

(Joel Beasley at 00:12:56) And to be honest with you, I think that's how I—I think the latter way is how it'll roll out. So we'll just go through and different little pieces will get solved. Those will be agentic, essentially. Like, "Oh, I'm really good at solving this. I'm really good at—" and then you'll get some controller, some distributor, whatever the word is. I forget what it is. But essentially, air traffic controller that's going to take the request from the user even if that's like act autonomously and solve this problem and then use its library of agents in order to do that.

(Avi Freedman at 00:13:25) I agree. And in fact, we just did a partnership—launched a partnership with ServiceNow this week where they're using us agentically for some of these workflows, and we are using ourselves agentically. As I said, like, we're learning what people are doing in Kentik using all the data and the observability platform and saying, "Let's help automate that for you." It's the next step. It's the—it's the knowing what is the totality and what is safe to actually take the next step and deploy. That's the tricky part. But this year, it's real that you can have—you can have real acceleration in terms of what the—you know, knowing what to look at and having, you know, 90% of the issues identified by machines doing what you would be doing and presenting to you why. It's not in—in Kentik, it's not as much reasoning. It's more you see—you see the entire chain of things that it was doing. So it's not actually explaining—it can explain to you sort of why it was told to do that, but it doesn't really understand the way a lot of the stuff doesn't really understand. It's just prediction engines that are given, you know, sets of workflows to do. But the end result of I tried this, I tried this, I tried this, it's mirroring what humans do. And that part, we're already, you know, able to deliver. It's the—it's the complete closed loop. Humans just sit back and aren't in the loop to look at, "Is this wise to deploy? Is this a wise change to make?" that I actually don't see that being, complete in the next couple years. Because what we're talking about is—I don't—

(Joel Beasley at 00:15:03) I don't know if it'll ever be complete, to add to your point, because—

(Avi Freedman at 00:15:06) Okay. Well, the software is—

(Joel Beasley at 00:15:08) The software is for the humans.

(Avi Freedman at 00:15:10) Right.

(Joel Beasley at 00:15:10) We're trying to achieve an end, some type of aim. And so without us in the loop—like, at some level, we have to be in the loop because the technology is serving us. Yeah.

**Avi Freedman at 00:15:20**
I think it might get there. It's always hard for me to reason more than three years out, especially now. But all I could say is, in the next three years, unless you wave your magic wand—you know, like the engineer that said assume the square chicken, then we can pack them this way. If you assume that all infrastructure looks the same or all software looks the same, and maybe in the software world it becomes more clear, or maybe if the same three AIs are suggesting all the configurations for everything, then operations actually becomes a lot simpler.

**Avi Freedman at 00:15:55**
But, you know, I don't want to call it debt because there's some wonderfully architected snowflakey infrastructure out there, and same thing for distributed systems and applications. But to have the AI replace the human in that agentic coordination framework, it needs to know what's there and what's missing, and that's still where I think humans are going to be advanced. But the people that aren't using this technology to make their lives easier—well, in my world, it's just getting harder and harder to hire people that actually know how all the bits and bytes flow and what all the bugs are and where all the bodies are buried. So we sort of have to use this where companies can't hire enough people to actually build and run this infrastructure and then let the humans focus on what they're uniquely skilled at. So that's threatening, but it's also an opportunity to companies to hyper-accelerate.

**Joel Beasley at 00:16:57**
Yeah. And I'm the early adopter type, so I like to play with the technology early and then watch it as it grows. But one of the things that I hadn't heard of before is probable cause analysis, PCA. Can you explain to me what that is?

**Avi Freedman at 00:17:12**
Sure. So PCA in Kentik is actually more classic ML and guided investigation, where because we see traffic—we see, you know, if you think about network to application, it's sort of like traces but for the network. It's so many bytes went from here to there in packets. Ideally, this application, that user, things like that. So when we see traffic shift, something has just moved, and it could cost you a lot of money. It could be a bunch of traffic going to Iran that you probably don't want to have happen, or there's a performance problem. And we say, what else happened at that time? It's looking across all the signals and dimensions that we have.

**Avi Freedman at 00:17:58**
Again, the humans could go poke at it and do, but we're sort of building these baselines and looking so we can say when this happened, what else happened? And then ask a few more questions. This part of what we delivered isn't full-on agentic, but so that we can say anytime someone's looking at a change, here's all the other things that changed at the same time. Because usually there's a lot of bugs in the infrastructure, but often someone caused it. Or in the reverse, okay, well, that thing changed, but does it matter? If the traffic rebalanced and no performance problems are happening and every customer is happy, then okay, it happened, but that link failure may not be that important. The reason why I don't call it root cause analysis is because there's been a lot of promise in the industry about root cause analysis that has not been delivered.

**Avi Freedman at 00:18:52**
If you look at AI ops, it was like, oh, we're going to automate all this stuff. But it doesn't really understand, even, the relationship between these things. It's more advanced event correlation. And so we call it probable cause because we think the human still needs to look at it and say, well, here's the three things that we think might have caused it. Do you agree? And then make it easy for them to take it from there. And we're betting that throughout the product. So anywhere there's changes or problems, we can show what might have caused it or been related. The more advanced version of that is some of the newer functionality we're rolling out where it actually sees this kind of pattern and then goes and does even additional—more than that first level of what changed at the same time, but to say, is it a problem? What might you want to do about it, et cetera?

**Joel Beasley at 00:19:40**
Well, as the leader of a technology company, everyone's thinking about how to adopt this new technology, when to adopt it, how to not go extinct because of it, all of these things. Where I'm interested is your experience with your direct reports and the leaders of your divisions. Like, how do you talk about this with them? How do you talk about what's coming down the pipe? Do you have annual meetings where you get together, or is it just casual conversation in the hallway, like, hey, did you see this? How does that actually look in practice?

**Avi Freedman at 00:20:13**
So annual is what dinosaurs do. As you said, you've got to stay on top of it. So we find that it's not—I wouldn't say it's strictly a generational thing, but it's actually a whole company thing. It's embracing—you know, set the boundaries, which is to say we've got privacy concerns and security concerns about what tools we use and how we vet them and what's done. We have things we can't do, like train on multiple customers' data because of representations we've made to our customers. And then say, you know, encourage a principle of maximal laziness. I don't know if you've heard this analogy in terms of software distributed systems, but some of the best architects, some of the best programmers are the people who are like, well, this is so—why am I doing this a second time? Like, okay, let me teach the computer to do this.

**Avi Freedman at 00:21:12**
And it's just gotten a lot easier to do that. And then say that's not—we're not just talking about technology. We're talking about billing someone or, you know, where we are more careful is to say, hey, we do want to drive up the chain and talk about—if we're talking about what I don't want to do is remove the humans. Like, there's a difference between customer email comes in, ChatGPT write my response, and customer email comes in and then saying, here's, again, directing it. Here are the things—like, I don't want what the model was trained on that we tell that customer. I want the—you know, or the way I use it, which is to say, tell me if there are any grammatical errors, but don't tell me if there's an overly complex train of thought and break it up because I'm fine with that for this internal communication. So what are the best patterns of that for demand generation or security audits or answering questionnaires? But again, set the principle that privacy, not complete automation.

**Avi Freedman at 00:22:12**
And then really, it's not going to come—no offense to my awesome leaders—from the leaders. It's going to come from everyone in the company. And again, we don't have a very informal process for it. There's no AI committee, AI czar. But it's just encouraging people to find and share things, run them by the security group, experiment with them. You know, as we talk about how we think about running Kentik, we talk about customer obsession, curiosity and experimentation, and teamwork. I don't think we have any AI team members, but for some functions that may happen before long.

**Joel Beasley at 00:22:49**
I think it'll happen soon. I think we'll have AI team members that we could just hire.

**Avi Freedman at 00:22:53**
Yeah. I mean, that's the agentic promise. So, Joel, do you think—do you know the Gartner hype cycle?

**Joel Beasley at 00:23:01**
Hype cycle? Yeah.

**Avi Freedman at 00:23:01**
Do you think we're going to rapidly approach the trough of disillusionment on agentic? Or do you think, like, because it's so overloaded and so hyped? Or where do you think we are with agentic?

**Joel Beasley at 00:23:13**
As a name, you know, as a word.

**Avi Freedman at 00:23:15**
As a name, as a word.

**Joel Beasley at 00:23:16**
Well, yeah, because it—because marketing drives that a great deal, right? First, it was AI LLMs that ran for a year. People are like, we need something different. Yeah, agentic. I don't know. I don't track it, like, in the—it's obviously very—it's getting more popular, so it's coming up on my radar. I'd say last year, it was maybe one in a hundred episodes. Now it's like one in twenty episodes. It's—

**Avi Freedman at 00:23:40**
I would have thought it was more than that because we're talking about that. We're talking about coordinators of the agentic things.

**Joel Beasley at 00:23:46**
Yeah. But, look, for me, because of my background, twenty years as a software engineer, I just see it as all the same. Like, it's just a different flavor to get—I see where we're going. I see the path for the next three to ten years. And I'm like, look, honestly, agentic AI is the least interesting thing on my radar. We're almost at bipedal robots in our house doing laundry. Like, we're two years away from that. Maybe three, but they're already in the factories. They're already working at BMW. They're already doing stuff. And so I'm over here seeing this, watching everything from Figure—I don't know if you've come across Figure—to what is Elon's Optimus, right? And then Boston Dynamics, they're too busy doing backflips with robot dogs. I don't know what's going on over there.

**Avi Freedman at 00:24:35**
Yeah. I mean, I was going to ask you, like, from an IoT perspective, and security is a place that we have a strong adjacency for because we have all the data about what actually happened, but we're not primarily a security company. But the key thing about IoT for me has been, you know, what about the world given how bad security is around IoT where, you know, the vacuum cleaner trips you and the oven explodes because someone's—okay, you know, it's not the privacy thing. You'll take the convenience.

**Joel Beasley at 00:25:04**
I'll take—we're—all right. So I like human behavior. And so I'm just saying if I had to bet money, I'm going to bet that we're going to exchange it for convenience all day. I mean, look at my relationship with Meta. Is that the history of security?

**Avi Freedman at 00:25:16**
But so do you think we have a Wall-E future? Remember the—I hope not. Everyone was on—

**Joel Beasley at 00:25:23**
I'm trying to get a six-pack right now. I hope not.

**Avi Freedman at 00:25:27**
Is it that the—I mean, that's the ideal, right? And that's what our customers want. It's help me leverage my brain to be able to make sure we have no outages and everything's fast. And if there is an issue, me or my peer, I can find it out, not have to do all the drudgery, and give me some suggestions. I won't always be correct, but hyper-accelerate me. Why? So I could do my job so we could build things to make the company go. And so that's the utopia is we get accelerated by all these things. And actually, you know, instead of using Uber Eats and all these things, I actually create things instead of just sitting around watching YouTube. So, you know, that's a struggle.

**Joel Beasley at 00:26:08**
And that's where I actually lack, Avi, my ability to predict some of these other things is because I am me, so I am a creator. So my entire—I'm anthropomorphic. When I meet people, I just assume everyone's creators, and it's like, that's the mindset that I'm in. So if you tell me that we're in a Wall-E world, I'm going to be the one not-fat person running around in the chair. I'm going to be tinkering with whatever available technology we have today to try to make something new or interesting and create something different. And so, for me, automating all these jobs, look, I realize 80-plus percent of the people are just going to go Wall-E on it, just sit down and chill. But for me, it's exciting because the people who do have it in them to want to create, they're getting more access to these tools at a greater speed. And so there's going to be a lot of beauty in it.

**Avi Freedman at 00:27:01**
I am excited for the future. And, you know, I guess—that's we were talking about it earlier. I mean, that's just how unbelievably fortunate we are. I feel that I am, you know, that I can make things and try to get people to use them and help them and that that's actually something that, you know, instead of just being the computer nerd in the corner, right? But how do we help people understand that that's a possibility so that they don't become irrelevant? The good news is in the, again, the infrastructure world, the people we talk to, they're all searching for ways to help with this, but they've also been disappointed about a lot of vendor promises about, you know, whether it's intent or self-driving networks or closed-loop automation, or AI ops. There's been a lot of over-promising. The good news is it can now be delivered, but it can be tricky to actually show people, you know, what's possible, what isn't possible because there's still a lot of hype in the space.

**Joel Beasley at 00:27:57**
Cry wolf situation, right?

**Avi Freedman at 00:27:59**
Yeah. It is to some extent. So the best way to do it is to say, look, all this stuff you were doing, here's some stuff prefilled out that just makes your life better. And people are like, oh, that's awesome. Let's have more of that. It's like, okay. Well, how do we do that? And then how do we systematize that? And how do we help? Because it becomes less—I guess, you know, to use the replete example—thank you for telling me how it was pronounced because I've only read about it, so I thought it was wrong.

**Joel Beasley at 00:28:24**
Guessing. You know? So yeah.

**Intro Narrator at 00:28:25**
I don't know.

**Avi Freedman at 00:28:25**
I'm always happy to get correction because, you know, who knows? This is like, this is how Latin was pronounced, except you can actually kind of know what they—how they call themselves. But again, take the knowledge folks and make it easier for them to describe and automate and control. Because, I mean, I could tell you there's a shit ton of infrastructure being built to enable all this possibility, and someone's got to do it. So, and again, like, we're not—the opportunity to break the internet and not be fired is relatively gone from the world like we had in the nineties when no one knew anything. So the expectations are a lot higher about how much the stuff—and the people are harder to hire and find. So, you know, it's an interesting time.

**Joel Beasley at 00:29:18**
Well, I feel really blessed with my background because I told my wife, hey, if I need a retirement plan, I could always write SQL because I'll be like—there'll be like twenty people in the world that know how to do it when I'm fifty.

**Avi Freedman at 00:29:32**
You would think that except, A, it's a little bit more lingua franca than COBOL sort of was. And B, actually, that's something that syntactically—

**Joel Beasley at 00:29:46**
Syntactically, yes. I need a new retirement plan, Avi.

**Avi Freedman at 00:29:48**
Semantically, unifying disparate schema where you need to infer from the data what the hell that column actually is and what the subtle dynamics of the differences between this millisecond measured that way and this millisecond measured that way, we're probably—I know I said three years, but we're probably, well, let's say we're probably five years off. Computers will be able to do that. Will they be able to debug things not written by computers in big complex distributed systems correctly? You know, again, what's the time frame for that? Thirty years, all bets are off, but next three years, you know, it's a good time.

(Avi Freedman at 00:30:28) But as you said, to your question, it is something that we have to do with our children, our community, the population. My grandmother, when she was watching me do computers—we're talking about the late seventies, early eighties when I was just very fortunate to have access to them and most people didn't—she's like, "I think there's gonna be people that make stuff and sheep, and it's better to be someone that makes things." And now it's been democratized because everyone can make things. So that's the cultural question: how do we get comfortable with that and enable people, educationally, to do that?

(Joel Beasley at 00:31:07) Do you have kids?

(Avi Freedman at 00:31:09) I do not have kids. My job is to take the kids, throw them in the air, toss them around, show them Looney Tunes if they haven't seen it, expand their minds, and then hand them back all excited to the parents and then run away.

(Joel Beasley at 00:31:23) So you're Uncle Avi.

(Avi Freedman at 00:31:25) I'm Uncle Avi. I give the foot rides and the airplane rides and wrap them in blankets and toss them around like we used to do in the seventies before people would get arrested for doing stuff like that.

(Joel Beasley at 00:31:35) I know. I was like, "Dad, what would you do when I was going crazy in a tantrum?" Because I've got three under seven. And he's like, "I'd just wrap you up in a towel kinda tight and hold you there for a minute so you wouldn't hit me." I was like, "Alright, that's good, right?" I was like, "Not doing that." Michelle's like, "Uh." I was like, "Alright."

(Avi Freedman at 00:31:55) No, probably not. And we ate dirt and peanuts and all sorts of other stuff, and we had only lap belts and vinyl seats that would burn the back of our legs in the summer.

(Joel Beasley at 00:32:04) You just brought back a bad memory. That was crazy, man. I could feel those vinyl seats.

(Avi Freedman at 00:32:11) I could still feel the ridges on my legs.

(Joel Beasley at 00:32:14) Yeah, yep. Wow, this is great.

(Avi Freedman at 00:32:18) But it's cool. I mean, it's cool that we have access to this technology, and then it's a question of how we work it. And the other thing is, we talked about business. I think that's still gonna be for humans even if we do have AI workers. Ultimately, it's a lot more refreshing. I mean, you run a business now. It's a lot more refreshing to be building technology and say, "I have a problem, I think I can solve it, I tried that, it worked, it didn't work, I iterate, I do it." When you're competing in business and running an organization, you never even really know. You're getting so much feedback about everything. You never know whether there was a path you could have taken that was better until you see a competitor go do it. But at the same time, if I build something that no one uses, does it matter? And we live in a capital—a way of—there's engagement, and I look at Pendo all the time. What are people using? How are they getting value? But also, in order to get money to build things, we need to make money, and it's a good way of measuring value. Do people find value in it? So as a nerd running a business is good, but it also is full of uncertainty in a way that is unfamiliar. So how do we help our people to help us to help customers? And how do we use AI in a way that doesn't dehumanize people? Those are all super hot topics.

(Joel Beasley at 00:33:40) There was an interview with Zuckerberg maybe a week or two ago where he was saying that in the next twelve to eighteen months, Facebook will be at the point where it's essentially autonomous systems writing and improving itself with—you can interact with it—that's where they were, that's their current timeline, that's what they're working on building. Have you heard that?

(Avi Freedman at 00:34:02) That sounds overly aggressive. Now, I'm the homeowner that was not that impressed with Metaverse because I learned distributed systems by optimizing multi-user dungeon games, from communicating processes, communicating in files to big distributed systems with state and all that. But they've got a lot of resource and a lot of people and are innovating in terms of AI. And they have an advantage, which is they have systematized and regular—regularized, that's probably not a word.

(Joel Beasley at 00:34:37) I like that word. It's fun.

(Avi Freedman at 00:34:38) Yeah. Normalized, I guess, would be the nerd way, the geek way to say it. But when I look at—we have customers that we call our first wave of customers. Customers like ServiceNow and Box and Dropbox and people like that that had gotten to the scale where it was only really gonna work if they did that. And then we have a lot of customers who've been around since Sneakernet was a thing. Remember Sneakernet?

(Joel Beasley at 00:35:14) No.

(Avi Freedman at 00:35:15) Okay. So that was like putting things on floppy disks because the network was down and running it over. People would joke, "The network's down." Like, we—it used to be okay. The business would run, the world wouldn't end if the network is down because telephony was something different, and you could just share files other ways or print them out. And you've got networks that have existed for decades. And when you look at things like—if you look at Facebook and you look at their network or some of their application infrastructure, everything is also sort of similar. In a lot of these enterprise infrastructures, it's not that the new stuff is really good, but there's a lot of stuff that just hasn't been redone where if there's an interface name, it's wrong because it was set decades ago and no one really knows why. And so, again, that snowflakeiness for a lot of traditional enterprises is gonna be difficult to do what maybe Facebook could do. Now do I believe they're gonna have completely autonomous but supervised all the distributed systems in a year? I don't know. That still seems a little bit much to me, but maybe.

(Joel Beasley at 00:36:22) What do you think? I don't know. You know? I've got the—there's the rule of visionaries, and you're an entrepreneur, we're entrepreneurs. You have to set the goal in order to hit it or fail trying. And so whether they'll do it or not, I don't know. But I mean, the problem is it's too hard to gauge because we're so close and we're moving so fast in so many different directions. I mean, what was DeepSeek like six months ago? It's pretty crazy. And now it's been rolled out across at least the two I use, Grok and ChatGPT, where you can see them think, you can—now they can go on the internet. I mean, the point that I think is gonna be a huge inflection point, I don't hear many people talking about. But the thing I think will make a huge change in humanity is the moment that you get one of these systems empowered to take action on the internet, like through web browsers and stuff like that, and they have the intelligence level of a 30-year-old professional. It's not even the most intelligent, right? Like, not an idiot, right? But like a moderate level of intelligence because they can compensate that with 24/7 effort and learning and reinforcement and stuff like that. So if you just get the system, if I can just sit here, Avi, and boot up a 30-year-old professional that can use the internet, has intelligence, and I can kinda train them and talk them through stuff, the game is almost over at that point because that's like 80% of the work we're currently doing. And so I think that's going to be the most profound impact that we have.

(Avi Freedman at 00:38:04) I am so terrified by—I have two thoughts. I am so terrified by the unleashing to take action of things that don't really understand. And yes, there's progress, but the majority of the embedding of the things that these things are doing is still primarily prediction, not understanding, primarily. Right? The training is based primarily on predicting. We get these amazing outcomes, but I also don't like the term hallucination because it's just bullshit predictions for the most part that cause it to predict something that doesn't exist because it needs that for the completeness of it. And I know there's work being done on the nannies also, but I just see a lot of this—to take the action part without supervision. Like, that's where computers are really good at doing what you tell them to, and they can do it, as you said, really fast, and they can do it 24/7. Imagine waking up and there's, "I have no money anymore." You know? We exhausted—Amazon is out of this thing that everything decided to do. So I really think that a year from now, and even three years from now, we're gonna be heavy into supervision. And that's—again, that's—I mean, as we think about it at Kentik, absolutely, the vision is to go towards autonomous. There's a lot of work to do that. But you can already boot up—effectively, we're trying to think about it as boot up a version of yourself that you can have a million of as monkeys looking at things that maybe you wouldn't spend the time on because it's too low return, doing the typing and debugging and saying, "Is there something here?" And then present yourself with the best version of that and then kill all the copies of yourself when nothing's happening.

(Joel Beasley at 00:40:02) I'd have to definitely modify the copies too because my copies would be like, "I'm not doing the work. Let's make another copy to do the work."

(Avi Freedman at 00:40:11) You know, you say that, but again, it's not—that's the thing. We're not really embedding these things with cognition, desire. I assume you've read the Asimov robot books, the mysteries where the Three Rules of Robotics come from?

(Joel Beasley at 00:40:26) I haven't sat down and read them. No. I've known them through pop culture.

(Avi Freedman at 00:40:31) They're an easy read. And I mean, I'm not a mystery guy, but they're actually—I mean, they're not like Heinlein's juveniles, but they're an easy read, and it's pretty interesting. But I think we have things that are simulating some of that behavior, but—and it looks from text output like there's desire to live, but I don't think that's actually what's going on underneath.

(Joel Beasley at 00:40:56) I do.

(Avi Freedman at 00:40:57) Okay.

(Joel Beasley at 00:40:58) Do you wanna hear my argument for it?

(Avi Freedman at 00:41:00) Yes, I'd like to hear.

(Joel Beasley at 00:41:01) Because I'm okay with being wrong, 100%. But I was thinking about it—and I like people who like to play with ideas because that's awesome. So I was thinking about it and I was like, okay, I had this thought, and the basis thought was, okay, what if—because I was looking at trees communicating with each other like mycelium, I was looking at animals, I was looking at humans. I was like, alright, well, what is this intelligence? And I was like, well, there seems to be some basic things like some communication and retention of information to move forward in the future. Like, those two components seem to be pretty universal to what we call intelligence because you can be conscious, but if you don't have retention of information and you're not there and you can't communicate, then it's—so I was looking at those things, and I was like, okay. Well, and then mix that with the conversation of the robots and it's like, well, what if—and this is just an idea. It's kind of out there, Avi. But what if we could look at consciousness and intelligence as a force like gravity? And when you apply that force through different substrates, you get different reactions. So you apply it through plant material, you're gonna get that type of communication, intelligence to the planet. You do it through animals, you get that, you do it through humans, you get this. And when you do it through silicon, you get that. And at that point, you get into materialistic—you can take a materialistic view and you can say, "Okay, well, Avi, you're not really feeling that way. What's happening is there's cortisol in your system that's interacting with this chain of amino acids that's interacting with this, and that's just creating this illusion for you that you actually are happy or stressed." Right? And you're like, "Well, what are you talking about?" So when we look at it and the computers, we tend to reduce it. "Oh, no, no, no. That's just some computations happening, some electrons firing."

(Avi Freedman at 00:42:39) I get the argument. Yeah, I get the argument. And I even get the argument that what are we but a collection of our histories. And but what's impressive and interesting about LLMs is that the assumption was historically with neural networks that they'd be trained and that you might freeze them, but more that they would continue to learn with the full retention of their capacity. If you look at the way that most of these LLM technologically are trained, the amount of information encoded in the context window and even the history is still, I believe, many, many orders of magnitude—times orders of magnitude—less than the information encoded in the model itself. So the—and as I said, really underneath, it's prediction engines in a way that I believe—so there's this physical argument, which is we're just biological computers or that's a silicon version of the biology. But it's really encoding such a small portion of what we do that there isn't even really been attempts to encode feelings, opinions, desires, ambitions, other than as reflections in the way that the technology we're interfacing with has been configured. Now the same technology could be configured, I think, to do a lot of innovation and exploration in ways that could be scary or awesome at actually more self-aware, desirous, thinking, evolving, interacting—artificial, we'll say, intelligence. Whereas right now, it seems like what we're getting is emergent things that look like intelligence, but the underlying things still seem to me to be very different than wetware versus silicon. But again, maybe I'm overthinking it. So that's just sort of the way I think about it. How much identity and individualness is there encoded? And, of course, we're crafted much more artisanally as humans over a much larger time than these things that we're depending on are.

(Joel Beasley at 00:45:05) I think we can agree on this. Right? Because first of all, this is not an argument that I hold strongly. This is something that I play with. Just an idea. I connect and relate with you. I do not feel bad when I turn off the large language model. Right? That's kinda like, okay, where are you at, Joel? I don't feel bad. You know? Just like I don't feel bad when I step on an ant. And I'm like, oh, it sucks. But I'm not going around stepping on ants.

(Avi Freedman at 00:45:31) That's a whole other question, which, again, I'm busy delivering outcomes for people that make the work go so that people can build AI stuff. So I feel really good about that. It's like I was talking to—I was interviewing—we just hired an internal corporate counsel. And I said, "I understand that it could be frustrating," because someone asked me, "What's my mark of success? Like, how do you know this is going well?" It's like, "No one ever thinks about your function. That's like the highest compliment." Right? We don't talk about it.

(Avi Friedman at 00:46:06) And that's the same thing for the people that run distributed systems and write and run the network and build it. When it works, everyone's happy. That's awesome. Now it never always works because vendors have bugs and users do strange things and all that. But the less you hear about it, the better.

(Avi Friedman at 00:46:22) So sometimes that's successful. But I don't think a lot about training AI and all that. But there are different ways of using some of the stuff that could actually get more towards these thinking, desiring things, and that could be good and that could be bad. But, you know, again, I'll just come back to mimicking humans for the understanding, for the watching, suggesting. I think that companies that aren't doing that in every area of the business over the next year—I don't think it's novel or revolutionary to say they're going to be at a competitive disadvantage.

(Avi Friedman at 00:47:02) And maybe they're regulated or maybe they have such inertia they can't fail, but most companies that I work with are thinking about that. The big hairy question is to actually replace humans altogether. You need to be able to understand what to do and do it correctly enough that you're not going to break everything, whether that's pissing off a customer or spending all your money or doing what I've done in the past accidentally and taking down the Internet or your network or, you know, autonomously running your application. I don't think we're quite there yet or will be in the next year, unless maybe, again, it's a constrained, perfectly developed infrastructure you can reason about like a Facebook. So—

(Joel Beasley at 00:47:44) Well, the way these things historically play out is, like we talked about earlier, it's just slow, steady improvements over time. It's like, okay, we feel comfortable. This little segment's automated. It's running reasonably well. We have human in the loop there.

(Joel Beasley at 00:47:59) All right. And then we do—

(Avi Friedman at 00:47:59) It to—

(Joel Beasley at 00:48:00) This segment. And then eventually, you've got 99% done, and you get better monitoring systems. And, you know, Josh and I are going through a process right now at the company, at the production company. We're sort of like an annual audit. We're looking at everything that happens in production, every step that happens in production, and then what potential is there to improve that step.

(Joel Beasley at 00:48:20) Like, can we consolidate these steps? Can we get—you know? And so I think companies should be doing that. They should be looking at the manual human steps that they're taking and at least asking themselves. And the goal is not to replace a person.

(Joel Beasley at 00:48:33) The goal is to say elevate. We want to elevate them because our competitors that are coming out that are born today—they're going to adopt the best, most available tools because they don't have existing systems and processes. They're going to do it the smartest way possible. We don't want to get out-efficiencied. We're making up lots of words today, Avi.

(Joel Beasley at 00:48:53) We don't want to get out-efficiencied by these other competitors.

(Avi Friedman at 00:48:56) Out-efficiencied? I don't know. I have—

(Joel Beasley at 00:48:59) A board member.

(Avi Friedman at 00:48:59) She's—when she starts punning, I can't keep up. She's so literate. It's awesome. But she can either tell me whether something I just made up is a word or should we agree it should be a word, and then we can use it with each other and not let our OCD get in the way, which, unfortunately, I have too much of.

(Avi Friedman at 00:49:20) But, yes. I mean, I think that all that comes back and summarizes sort of the challenge we've set internally and the promise for the customers. Everything you know how to do that's on the monitoring and diagnosing side, right, the analytics, we're going to automate it for you. How to build the castles that these robots live in and how to review the structural engineering before you go make that change or, God help you, automate making a million of them is still your job, and we're going to empower and enable you. So we're not going to promise that which we can't deliver, although that is the vision in the future. And it's been awesome that the last couple years have helped us accelerate the pace of that because that was always where we were going.

(Avi Friedman at 00:50:13) And I think that every observability company that isn't taking up that challenge to augment and elevate the humans is going to find someone, whether it's Google or Microsoft or Anthropic or OpenAI or a competitor that is, that's delivering those better outcomes. And we've already seen executives understand this much better than, "I could do my trace thing and it connects this way, and I can slice and dice this." You could tell them that the robots are doing the human job, educated by you, and that, therefore, you can do your job better and that we're not about to fail because we can't hire these people. You know, that's a much better place to be for everyone in their career and to uplevel ourselves.

(Joel Beasley at 00:51:06) And where do we want to send people to learn more about this? Is it kentik.com?

(Avi Friedman at 00:51:11) Yep. kentik.com. Take a look. I'm Avi Freedman on most of the social. Happy to talk to folks who are thinking and exploring.

(Avi Friedman at 00:51:20) And hug your networkers, hug your lawyers, hug your distributed system architect. Even, you know, tell them you're thankful that you don't hear from them enough, because they do a great job and make the world go.

(Joel Beasley at 00:51:33) Yes. And if you're on Avi's side in our fun debate about humans and computers, go ahead and reach out to him, show him some support. Yeah. No, this was good, man. I wanted—can I wrap up on a leadership question?

(Avi Friedman at 00:51:46) Yeah. Sure. I'll put—

(Joel Beasley at 00:51:47) You on the spot. We'll see how it goes. I'm going to ask you for a piece of leadership advice, but I'm going to give you constraints. Is that okay?

(Avi Friedman at 00:51:54) Okay.

(Joel Beasley at 00:51:55) The constraints are, I want it to be a piece of advice that you heard, then you've implemented, and it was so effective that you continue it today.

(Avi Friedman at 00:52:07) Never ask someone to do something that you wouldn't do yourself. Lead from the front. It's the best way to get respect and the best way to learn and, you know, be in it with folks. That would be the piece of advice.

(Joel Beasley at 00:52:26) That's kind of beautiful. I wasn't expecting that.

(Avi Friedman at 00:52:29) I mean, it's the—I'm not Israeli, but it is the Israeli—it is a leadership principle, I believe, that they teach in the Israeli armed forces. You know, the general's not sitting in the back. Right? You know, you're leading from the—lead from the front is sort of the way of saying it, but you just get a lot more credibility and respect that way I found.

(Joel Beasley at 00:52:53) There's a great technology startup community in Israel.

(Avi Friedman at 00:52:58) Absolutely. Yeah. And a lot of them are people that served together, you know, in very tight-knit communities. It used to be the tank commander was the CEO, and his lieutenants were heads of the parts of the company. So, yeah.

(Joel Beasley at 00:53:12) Yeah. About five years ago, I did an interview with this Israeli guy, and one of my first that stood out to me. And he was very direct and I grew up—my dad, Air Force. So I grew up very direct and I had to learn how to soften it up in the business world. But my household I grew up in was super direct and I noticed that he was being direct and so we got along really well.

(Joel Beasley at 00:53:31) And then he introduced me to one of his friends, and then one of his friends that were also IDF, post-IDF people. Yeah. And I realized, oh my goodness. I'm like, yeah, we actually have a class that we teach here to teach people how to soften things up for the Americans.

(Avi Friedman at 00:53:45) My mother explained to me when I was growing up. She said, "You should realize that people that grow up in ethnic households often communicate to each other exactly how they feel in real time, sometimes at loud volume. And that's not the way that the business world works, and people will just look at you like you're from another planet." I sort of look at it a different way, you know, because I also—I was, you know, I have an academic background. I don't have a degree.

(Avi Friedman at 00:54:13) A lot of my mentors were people that went through academia. When I was at Akamai, I worked with hundreds and hundreds of people that were from academia, like MIT, like, emptied out into Akamai. And I was just smart enough to communicate the real world to them and understand the words they were saying because I was trained in it. And there's a real distinction between "I hate your idea" and "I hate you" that is just understood.

(Joel Beasley at 00:54:42) Mm-hmm.

(Avi Friedman at 00:54:42) And there's also, which is something that I do always struggle with is, there's an assumption that if you're part of the conversation, I think you're awesome. And I don't need to tell you that because you're here. Look at us. We're doing stuff. And that's also not—and it's not just generational, but it's also true that people that come from that kind of background often don't say thank you or I appreciate you enough, which gets back to the leadership conversation.

(Avi Friedman at 00:55:08) You know? It's, you know, you have to do—at Akamai, you had to and especially even in the very early days, you had to do something really impressive. But then when you got feedback about how awesome that was, it was so meaningful. You know? And so I think there's countrywide differences in culture around these things.

(Avi Friedman at 00:55:27) There's regional differences. And, yeah, it's part of, you know, as you figure out how to operate a company, you know, where do you come down on that? And how do you make it okay to disagree, understanding that that's awesome that we disagree, that we're going to talk about it and figure it out and get to a better place, you know, for us, our customers, everyone. You know? But it is something that not everyone has seen. You know? So—

(Joel Beasley at 00:55:51) I am so spoiled with having this show. I get to meet the greatest people, like you, the open-minded thinkers, individuals that just want to make progress and be useful. Like, I don't know. This is—I'm having a good day. I think I said at the beginning that I'm having a good day.

(Joel Beasley at 00:56:06) Josh is probably like, "Avi, I want you to believe me. I'm not usually this happy." Okay.

(Avi Friedman at 00:56:11) You actually convinced me. We were going to argue—I was going to—I almost raised this question of, like, are we going to live in a simulation, which I think is such a bullshit, meaningless, intellectual nerd question to ask because what the fuck does it matter? What are we going to do? Like, it's like, hey, you know, it's like you're going to rail against God? What are we going to do?

(Avi Friedman at 00:56:29) We're going to break out of the simulation and take over? Why does it matter? Why are we wasting time on this? But, you know, there's some isomorphisms to some of the stuff that you discussed that's interesting. And I'll give you a reading homework assignment from when we talk next year.

(Joel Beasley at 00:56:40) Okay.

(Avi Friedman at 00:56:41) The problem that I was worried about 20 years ago that comes from science fiction was the gray goo problem. Are you familiar with the gray goo problem?

(Joel Beasley at 00:56:49) I'm not.

(Avi Friedman at 00:56:50) Okay. Imagine self-replicating nanobots that eat carbon and shit robots.

(Joel Beasley at 00:56:55) Yes. Okay. I am familiar with it. Yeah.

(Avi Friedman at 00:56:56) And imagine you have a software bug that causes them to eat all the carbon in the universe and make and shit more robots, and now we've filled the universe. Yes. Right? And look at, you know, it's the same as biohacking or whatever. So this is again, there's a similarity. When we talk about complete automation with AI today, knowing how bad we are at writing software securely and within bounds, this is where, you know, I sort of draw the line, but I'm hopeful.

(Joel Beasley at 00:57:21) You know? Where I have—that's where I think God plays a role in my life with faith. It's like I had to make a decision in my life that I believe that everything works out. It's like, is life a comedy or a tragedy? Like, I think it's an amazing thing. And that's just the team I joined.

(Joel Beasley at 00:57:37) I was like, I'm going to go on that team. And, but, yeah, there's a—

(Avi Friedman at 00:57:41) Look both ways before crossing the street.

(Joel Beasley at 00:57:43) Of course. I'm not an idiot. Yeah. Right. Well, I look—faith has to come with equal parts action.

(Avi Friedman at 00:57:49) Right.

(Joel Beasley at 00:57:49) You know? Like yeah.

(Avi Friedman at 00:57:51) There's a joke. This pious Jewish guy is, you know, he's got 10 kids, and he is poor. And he's like, "God, I do everything you want, and I'm trying so hard. And, you know, could you please—could I please, you know, like, win the lottery?"

(Avi Friedman at 00:58:09) And after, you know, after a year or so, he's walking to synagogue and hears this voice boom down. It's like, "You could help a brother out and buy a ticket." You know? It's like, you have to be an active participant in in in these kind of good outcomes.

(Joel Beasley at 00:58:25) Yes. Faith without action is wishing, and I think that's silly. And so the way I've lived my life is I take as much action as I can. I pray. I try to—I have a personal rule where I try to always do what I call the next right thing.

(Joel Beasley at 00:58:39) I almost always know what it is. I sometimes don't like what it is, because it costs me more than I want to, you know, spend and stuff like that. But I always do the next right thing. Right? You know?

(Joel Beasley at 00:58:52) And so that's kind of where I'm at. I'm 37 for context as you're learning. So that's—if you have any other advice for navigating life, throw it on me.

(Avi Friedman at 00:59:02) Be long-term greedy.

(Joel Beasley at 00:59:04) Long-term—

(Avi Friedman at 00:59:05) Greedy. It's fine to be—even if you're not naturally—it's a way that I've met people that are not naturally empathic and thinking about I want to do the next right thing. But short-term greedy is like, I'm going to get the most money now or the best outcome now. It's just understanding that think about how to build the relationship best for the long term, deliver value best for the long term.

(Avi Friedman at 00:59:31) Like, don't optimize in the short term. Think about the totality of this relationship. Right? Someone asks you for help and they're 20. Imagine what they could become in the future and help you.

(Avi Friedman at 00:59:44) And I'd had no idea about any of this, and I was just like, "Oh my god. Internet routing is so fucking confusing, and it doesn't need to be. Let me just write these articles." And everyone's like, "Oh my god. You are so smart because you knew this thing that I didn't know and explained it to me 40 years ago."

(Avi Friedman at 01:00:00) And it's like, no. Not really. But, you know, they're so grateful that you helped. And so just think about it as not like, "Oh, it's a pain in the ass," but just think about the future cone of goodness you've created yourself by putting good into the world. So there's an evil, you know, it's—greedy is a bad word, but, you know, even if you don't naturally think, "Let me help people."

(Avi Friedman at 01:00:25) Yeah. It benefits you to do so. So it's a—but I need to think about do the next right thing because I'm not sure that's—I actually understand that. So—

(Joel Beasley at 01:00:33) Well, practice it this week.

(Avi Friedman at 01:00:36) Okay. Awesome.

(Joel Beasley at 01:00:37) Just ask yourself, what's the next right thing? And you almost always know in your gut. You're always like, "Uh, yeah. That might not be the most beneficial thing for me in this, but, like, long term, it's kind of very similar to what you said." It's like long-term doing the next right thing is long-term greedy.

(Joel Beasley at 01:00:52) Doing the next right thing is not usually short-term greedy. Yeah. Rarely is doing the next right thing a short-term benefit.

(Avi Friedman at 01:00:58) Interesting. Yeah. It seems like there's a lot of symmetry. I just haven't thought about it. So—well, interesting.

(Avi Friedman at 01:01:03) Thank you for the deep conversation.

(Joel Beasley at 01:01:05) The most frustrating thing for me, Josh, is when I call support and I can't understand them.

(Intro Narrator at 01:01:10) Yeah, man. I hate that.

(Joel Beasley at 01:01:11) That's why I like US Cloud. Not only is it better, faster support, but all the engineers are US-based engineers, and it's also a lot cheaper. 94% of US Cloud's clients report saving a third or more when switching from Microsoft Unified Support to US Cloud. So now you'll just have to figure out what to do with all that extra money. If it were me, I'm responsible, so I'd reallocate the money to improve my team.

(Joel Beasley at 01:01:34) Josh, what would you do?

(Intro Narrator at 01:01:36) I'd probably just buy more guitars. More guitars.

(Joel Beasley at 01:01:39) Visit uscloud.com to book a call and figure out how much your team can save. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected].

(Joel Beasley at 01:02:02) Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.