Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
The Spirit of Curiosity | Randy Bias on the Coming Decade of AI Operations
35 Plays1 month ago
The entire industry is teaching AI to write code. Almost no one is teaching it to run the systems that code lives on. That gap is the biggest open opportunity in infrastructure right now.
In Episode 3 of The Root Cause, Priyank Upadhyay sits down with Randy Bias VP of Technology & Strategy at Mirantis, co-author of "Pets vs Cattle," inaugural board director of the OpenStack Foundation, and founder of Cloud Scaling (acquired by EMC). Randy was early to the internet, early to the cloud, and is now early to what he believes is the next world-altering shift: AI-assisted operations.
We get into why operators carry 80–90% toil while the tooling ignores them, why founders without operational scars struggle to serve them, the difference between a "sandbox" and a "playground" for agents, why operations is fundamentally a team sport, the one job you can never hand to an agent (judgment), and why domain expertise is the only real moat left in an AI-leveled world.
The Root Cause is an original series by RubixKube — Site Reliability Intelligence: see more, plan better, act safely, and learn with every incident.
Platform: https://rubixkube.ai
Community (Slack): https://rubixkube.ai (see footer)
#AIOps #SRE #SiteReliabilityIntelligence #DevOps #Kubernetes #RandyBias #CloudNative #AIagents
Recommended
Transcript
Randy's Background and Role in AI
00:00:00
Speaker
You are the OG SRE, Randy, then. I don't know about that. I know some serious OGs. I'm not sure i would consider myself one, but thank you. I was a deep, deep AI skeptic.
00:00:14
Speaker
um I kind of always have been um to a certain degree. I started building ISPs in 1992, 1993. That's before there was a commercial internet.
00:00:27
Speaker
That's before I was born, Randy, by the way. Yeah. Sorry, I'm going to show my age a lot,
00:00:45
Speaker
All right. Hey, Randy, how are you? I'm good, Priyank. are you? I'm also doing good. So first of all, thanks a lot, Randy, for taking out the time. i know you have a very tight schedule and very busy person.
00:01:00
Speaker
So thanks a lot um for doing this. And the reason I wanted you on the show was because this ah format and the approach that we are taking towards tech technology, towards AI in infra, and infrastructure in general, these are kind of topics where a veteran like you would would bring in a lot of insights and that helps a lot of my audiences.
00:01:35
Speaker
To anybody who is living under the rock, so Randy is with us. He is currently at Mirantis, VP of Technology and Strategy there.
00:01:48
Speaker
Randy co-authored Pets and Cattle. I hope you remember that, Randy. and And he was an inaugural board director of OpenStack Foundation in 2012, founded Cloud Scaling, which was acquired by EMC, brought AT&T and Walmart into OpenStack.
00:02:08
Speaker
And he is the cloud veteran who has given us the vocabulary of what we are using day to day. So welcome again, Randy. Welcome to the root cause.
00:02:20
Speaker
Yeah, it's great to be here. um And the name of your podcast is... Thanks. Thanks a lot.
AI Evolution and Randy's Changing Perspective
00:02:28
Speaker
Awesome. So let's begin, Randy.
00:02:32
Speaker
This AI space in general has exploded in the last two, three years, what you are seeing. And you have been in the industry for quite some time.
00:02:45
Speaker
What do you see and how do you feel about it at the pace that things are moving?
00:02:52
Speaker
ah I feel like that's a two-parter, so I'll take it in two parts. So um so I'm really excited. um I think we're about to see some very serious change, um and I'll explain why in a second.
00:03:06
Speaker
I was a deep, deep AI skeptic. I kind of always have been um to a certain degree, but in early to mid 2024, when I started playing with some of the tools, I started to turn around. I was like, wow, there's something here. And then what's been amazing is how much it's picked up speed and the the pace of change.
00:03:34
Speaker
And I went from this is an anomaly or a blip to like, okay, this is going to be like cloud, the internet or the advent of computers is just going to be like a world altering event that touches the entire IT t stack top to bottom.
00:03:49
Speaker
And, um you know, I want to be in front of that. i I was in front of the internet. You know, I started building ISPs 1992, 1993. That's before was commercial internet.
00:04:02
Speaker
um that's before there was a commercial internet That's before I was born, Randy, by the way. Yeah, sorry. I'm going to show my age a lot, probably. And ah then I was in front of the cloud computing stuff using AWS EC2 in late 2006 when it was in private beta. knew Nobody knew what it was. And there was one instance size, no VPC, no EBS, none of the things people take for granted now. So there's another one i I'm really interested in being in front of.
AI's Impact on Operations and Investment Needs
00:04:33
Speaker
um But kind of more importantly, for me, when I was a kid, when I was a teenager, and I was using the internet before anybody knew what it was, and I was using Unix before Linux existed, and you know just that was like my that was my hobby.
00:04:48
Speaker
When I went home from school, I was hacking on stuff. And i was writing code, and I was doing systems administration, and it was all all computer stuff. It was just fun, you know puzzle solving for me.
00:05:02
Speaker
And you know that led into my first job doing systems administration and everybody was doing everything manually in 1990. And I'm like, I'm not going to do it manually. The computer could do that for me. I'll just write some code that'll like automate the configuration of all the config files for our 300 server fleet at this you know semiconductor startup. And so I did it.
00:05:23
Speaker
And then every single job I had after that, I automated everything I could. And so I look at what I can do and I'm like, wow, man, this is like automation plus, plus, plus, plus. plus We can really empower people.
00:05:36
Speaker
And the thing that I'm most frustrated about that I'm happy happy that I'm on your podcast to talk about is... people are paying so much attention to AI-assisted coding that they're not thinking about AI-assisted operations. There's a few people, there's you, there's the Commodore folks, there are people dabbling in it, but the level of investment, the number of people who are thinking about AI-assisted ops, it's just not enough. There's not enough investment. And the level of toil Yes, developers have to do a lot of toil.
00:06:07
Speaker
But if you look at operations, folks, the level of toil they have is ridiculous. They're constantly in this mode of paying down technical debt and spending 80, 90% of the time youre either paying that technical debt down or dealing with operational problems, and they can't spend time improving the systems. And it doesn't matter whether you adopt ah SRE or DevOps or any other discipline.
00:06:32
Speaker
It's just that like the tooling and the culture and a bunch of the things need to change. And the opportunity is now to make a serious mind shift in how operations works and to use AI to help us. And I think between the underinvestment and then people being gun shy, oh, we can't let agents touch production, you know, is um you know, there's, there's, we don't have enough momentum and I'm hoping we build it soon.
00:06:58
Speaker
Awesome. Absolutely. and And I think you hit the pain points. ah Exactly. I am very young. I have been in the industry for just around a decade.
00:07:11
Speaker
And one thing I felt is it's not a very like pleasing moment when you have to go and fix a production issue.
00:07:22
Speaker
The timings are totally off. And every time you do a release on a Friday, that's like that's the meme, right? You do a deployment on Friday and whole the whole weekend goes, you you don't have to even ah worry about it. the The thing is, this pain point, and you're rightly addressing that, the investment and people coming to the space is very low.
00:07:49
Speaker
I have been seeing a bit of that opposite insight and it could be due to the algorithms you know social media has because I'm working in the same domain so I keep seeing stuff but I'm seeing a lot of rise in the AI SRE era or operations era the reason being the natural shift from post-development, what it could be.
00:08:18
Speaker
I think it's naturally tending towards operations, right? So a lot of companies who are thinking a bit of tomorrow would be you know typically picking up the space and working something on it.
00:08:30
Speaker
What do you say? Yeah, so I definitely think a lot of people have identified that AI-assisted operations, is an opportunity that's underserved.
00:08:44
Speaker
But at the same time, the the cynic in me looks at some of these, and I won't name any names, and I see companies with founders who have zero operational experience.
00:08:57
Speaker
And i am like, you know, you're going to have a very hard time figuring out how to serve operators if you haven't had their real world experiences.
00:09:10
Speaker
And i think people think they can get around that by kind of using product management techniques to go in and do customer interviews and try to understand their pain points and problems and all the typical stuff. But the problem is, is that as somebody who's delivered a bunch of products to market, um you know, when you go in through that product management process, a lot of what you're trying to do is suss out what the customers are not telling you to read between the lines.
00:09:36
Speaker
And you cannot do that with operations staff and operations executives unless you have operations experience.
Building Trust and Adoption in AI
00:09:45
Speaker
And so you may just be serving what they you know tell you are the symptoms as opposed to trying to figure out what the underlying causes are. Because as I look at this, I say to myself, OK, if ai agents can start ah taking ah operators and making them 10x operators, if it can speed everything up in operations, like it's speeding it up in software development,
00:10:08
Speaker
then a bunch of the things that we've done before, trying to patch operations with heavyweight procedures and things like ITIL and ITSM and all that, that will break. All of that will break.
00:10:23
Speaker
Right. You even just look at what's happening with vulnerabilities right now. And the speed of vulnerabilities is now like ratcheted way up. Yeah. And so the pressure is going to increase on the security teams and the operations teams to respond to that.
00:10:37
Speaker
And if they try to respond with the existing processes and procedures, they will absolutely break under the load. Yeah. they had to figure out how to use AI assisted operations to allow them to take the fire hose and manage it.
00:10:52
Speaker
And I just don't think that, um you know, we have enough of people with real operational background actually trying to work and solve these problems. I see some opportunists. I see a lot more opportunists who are trying to jump into the area because it looks underserved than I'm seeing, you know, real domain experts.
00:11:10
Speaker
I agree to that. And By the why do you feel people are not adopting much AI who are into operations?
00:11:21
Speaker
I see a lot of reluctance on SREs, platform engineers that I talk to, are very biased, I would say, not to anti-AI in most cases. Why do you think that is happening?
00:11:36
Speaker
Security guardrails? I think it's a bunch of different things. and you know I was just helping my my brother out who has been doing security operations like his whole career, network security, and um he just hadn't started playing with the tools yet.
00:11:51
Speaker
He didn't know what was possible. And if you look at his day-to-day job, he spends a massive amount of time on stuff that has very little to do directly with a lot of hands-on security operations like documentation and things like this.
00:12:07
Speaker
Now to migrate from one VPN platform to another, all those things that he could use those tools for. So I'm starting to try to help him level up. So I think, you know, some of it's the classic problem of like operators are so snowed in. They're so overwhelmed by, you know, the 80, 90% toil that they have that thinking about adopting new tools is just really low on the priority list. So they need to be encouraged. They need to be empowered. You know, people need to ah bake the tooling on top of the generic agents so that there's more operations centric tooling. and And then I think there's the big, you know, thing that we sort of just discussed, which is the risk aversion.
00:12:49
Speaker
Almost everybody in operations is risk averse. The security operations folks, network, everybody. you There has to be some way to prove to folks that they can start leaning on AI agents to remove toil. And I think that that begins...
00:13:08
Speaker
in in order to build trust and gender trust, the best way to begin is to apply agents in the areas where it's read only and they're not taking any potentially destructive actions on the systems. And then I think if you can kind of but make those two things happen, we'll start to see that groundswell.
00:13:24
Speaker
It's not a question of if it's a question of when, there will be no choice. Like the the the environment will force operations people to adopt these tools. So one thing you also brought in our introductions was the the velocity and the pace as at which AI is booming. There is a bubble kind of a scenario people are often referring to as an AI bubble.
00:13:50
Speaker
You have seen a lot of bubbles in past, right? Do you feel that is also a factor people are not able to adapt to the pace that which things are developing? ah because So let's say you you are on a blank canvas and something comes in, you are fine. you Okay, you write it down. But if lot of things come in, you get overwhelmed by the things that are available to you and the directions that are possible, verticals that are possible, that you are too scared to even start. Is that a problem or challenge?
00:14:25
Speaker
Yeah, I think it's part of sort of this just broader situation, which is like, I've got things under control. Okay, maybe I'm spending 80 or 90% on toil, but I've got a set of vendors that trust. I've got my tools in place. You know, we are keeping the system up.
00:14:42
Speaker
Like, I don't I don't need to make a change. And if I don't need to make a change, I'm not going to. And that's that typical risk aversion that most operations teams have. And and and rightfully so, that like there's no criticism.
00:14:56
Speaker
There's also the classic, especially with enterprise IT, t enterprise operations team, there's a classic issue of like, they've got the vendors that they've got, you know, they want to hear from their vendors what they're going to do. And there's an over-dependency on classic legacy enterprise vendors trying to solve modern problems where then there's an IT revolution going on, which didn't work well in the cloud days. So of course it won't work well in the AI days, but the the risk aversion
00:15:30
Speaker
While I understand it, it's like I said, and I'll probably just keep harping on it, teams will be forced to. They will be forced to adopt these tools because it won't just be CVEs and the increase of vulnerability. There will be a lot of other competing pressures.
00:15:45
Speaker
As enterprise development teams start to adopt these tools and ramp up their coding velocity, the the level, size, frequency of updates to production will also increase. i i was I forget which podcast I was watching or or blog I was reading or whatever, but they were talking about GitHub's increase in commits and since the beginning of the year.
00:16:10
Speaker
And it's an exponential increase. Exponential. Right? I mean, and so that's that's… GitHub is the canary, right? It's telling us about what's happening everywhere, which that velocity is increasing. You just look at my GitHub and you can see going back all the years, how much I've committed. And it's a smattering here there in the last nine months, it's like a ah metric ton of code that I've been generating using coding agents. to do our exploration in this operations area. So, you know, it's just a matter of when, not if. And so I i feel for the operations teams who drag their their feet on this because they are going to be in a world of pain as the pressure increases. This is one of those times where...
00:16:55
Speaker
I really strongly believe that you've got to get out in front of it right now. And the thing is, is like it's like with cloud, right? yeah Nobody can walk in with a playbook. your Your team has to go figure out how these AI tools work and what they mean for you. and What I'm seeing is everybody's kind of doing it slightly differently and they allow for that. There's not one way.
00:17:15
Speaker
And so you have to go exercise the muscle. There's there's zero shortcuts and you cannot depend on your and your legacy vendors to provide you any kind of solutions anytime soon. Yeah.
00:17:27
Speaker
We are on the like the section of the podcast. I'm seeing this going in a naturally direction of pattern recognition. You're identifying a lot of patterns from your past history plus what you are seeing in the industry. Do you feel there is a like similar kind of thing that you saw in your your journey, in your career?
00:17:53
Speaker
You talked about pets and cattles. I was reading about it and you said that to manage infra, you have to treat it like cattle. You and don't treat it like your pet. so Yeah, the cloud native way. Yeah. yeah So do you think similar kind of patterns you are seeing repeated over your career journey?
00:18:16
Speaker
Yeah, I mean, i'm I'm still trying to form some of those opinions. If you kind of look at the timing on the pets versus cattle thing, um that came out in about 2011, which was about five years after I started using AWS, four and a half maybe.
00:18:35
Speaker
And um i for me, I'm about two and a half, two years into this journey. So I still think it's really early.
00:18:46
Speaker
the The stuff that i i the the hints that I'm getting are that there may be such a profound shift that it's kind of hard to get our heads around. Say, for example, if we look at what's happening with AI-assisted development, they're literally, people are literally writing code and learning how to get the agents to write code in such a way that nobody's looking at the code.
00:19:14
Speaker
Yeah, yeah. And I know it sounds crazy and I understand that people feel like that may not be sustainable, but you know, the one thing I've seen when these kind of seismic events happen, when there's a massive transition like that, is that you look at that you look at the momentum and there's just a certain amount of momentum. And when it's just like at this like 45 degree angle,
00:19:42
Speaker
You know, it just won't fall off a clock fall off a cliff. It never falls off a cliff. I remember people in the early days going like, oh, Amazon's going to hit a cliff and, you know, there's not going to be enterprise adoption. Never happened. Just straight line up.
00:19:55
Speaker
Same with Google, same with Azure on their cloud offerings. I mean, yeah there's i just don't see, you know, There might be an AI bubble, we might see an economic implosion, but I think adoption of the tools is very early. It's just going to continue. And the things that we see already are that you know the model increase in terms of its capabilities is just relatively linear, but at a very fast pace.
00:20:23
Speaker
that people are figuring out how to build the frameworks around the models that make them super powered, to borrow from that name of that agent skill. And you go and you look at, say, Cloud Code, if you could go back a year ago and how good it was, it's nothing like it is now.
00:20:44
Speaker
And then while the model has gotten better, the harness around it has gotten much, much better. And so I think that our ability to figure out how to leverage these tools and use them is is very early on.
00:21:00
Speaker
And you know we are going to increase in in our learning and how we leverage them like we did with cloud and cloud native, and the models are going to increase in their capabilities. like We haven't seen what I'm thinking of as system harnesses or operations harnesses yet. They've all been focused around coding.
00:21:17
Speaker
you know You go take the cloud code system prompt apart, and it's it's almost all about software development. right Nobody's done that for operations yet. And so, you know, going back to kind of your question that that kicked this off was like, you know, what are the patterns? Well, I think that there's a high probability that despite what people think that that it's dangerous to let agents operate in production, that it may wind up being that agents do operate in production with with very little human oversight over the very long haul.
Future of AI in Operations: Autonomy and Resilience
00:21:51
Speaker
And you might ask yourself, well, you know, how can, you know, that be? And so my kind of early thinking is, well, in the cloud days, in the early cloud days, we sort of posited, you know, the pets versus cattle thing and stuff that, that, you know, hey, right. You don't need staging anymore. You've got production A. And if you're going to do a rollout, you spin up a whole new production B that's exactly the same as production A.
00:22:17
Speaker
right You bring it all up right for the period of time that it takes you to cut over. You synchronize the data. You run your A, B tests. You do some load balancing of some of the capacity over to the new production B. You make sure you're good until the point where you you know are running production B and you shut off production A. So things like staging ah environments become less important.
00:22:38
Speaker
And you know we're going to run multi-cloud, blah, blah, blah. And a lot of that didn't actually happen. And part of it was because it was complex. It took a lot of, you know, work and attention to detail and toil to make it all happen. Fully automating it with tools to bring up production B was one thing, but doing, you know, kind of a full regression and test harness against it was very, very difficult.
00:23:03
Speaker
And maybe some folks could do it like the Netflix's of this world, but the average enterprise couldn't. But if you've got AI agents to help you out, a whole bunch of that stuff may suddenly be very, very easy. So now you're doing things like letting agents drive everything, but they're using a lot of the ideas that the humans have had, the patterns that humans have had to basically give you ah resiliency and checks and balances in your systems so that even if they did something catastrophic, they're
00:23:34
Speaker
There's a clear way for the agents to recover from that failure. It reminds me of some of the times I've seen agents do stupid stuff in my environment where you know they delete a file and I'm like, yeah, I didn't want you to delete that file. But I created a new a rule now. ah This is just, for example, where they know that when they delete a file, when i ask them to delete a file, they stick it in the archive directory.
00:23:56
Speaker
So now if they ever screw up, which doesn't happen as much anymore, they just go back to the archive directory, pull it out and they stick it back. So there's there's a way forward here that I think people can't really see that I do believe involves agents eventually largely running our systems without much human intervention.
00:24:14
Speaker
Yeah, so so how do you sandbox the agents? you know I'm not sure that there's like a ah final solution here, but um you know one of the things that I noticed in the case of, um for example, OpenClaw is that the more that OpenClaw was hardened, the less useful it was.
00:24:38
Speaker
And I'm not suggesting that agents should run free, but I'm wondering, you know, where the trade off line is. And I think it might be use case dependent.
00:24:51
Speaker
If I've got some kind of agent that is going to be touching very, very secure information, you know, very secure APIs, information silos, whatever, you know, maybe I want to harden the absolute crap out of it. Whereas if it's like a personal assistant, you know, setting it free on a Mac mini, maybe that's okay.
00:25:10
Speaker
So what does that mean for operations? Well, I'm not a hundred percent sure yet, but I, I, one of the things that I think is interesting is if the AI agents in operations, know,
00:25:26
Speaker
are constrained so much that they can't think out of the box and take certain kinds of actions like search the web to find unique solutions or maybe new latest information about how to solve a problem, then I think, you know, how are they better than traditional automation systems?
00:25:47
Speaker
that there has to be some ability to give them to you know do their thinking, right? we We just identified that the harnesses were a big part of you know the increase in quality of of what we get from them. Well, you know a systems harness or an operations harness has to provide a certain amount of capabilities to the agents to basically help in in operations. and Maybe you can kind of confine them by role so that they can only access certain information by role. So it's not a bunch of agents who can access all things, but you've got to give them the capability to do more than like do basic web search and write some code and things like that. And so I'm sort of like drifting away from kind of put them in a straight jacket and i'm sort of more thinking about what I'll call playgrounds.
00:26:37
Speaker
which in my mind is not a sandbox. A sandbox is very, very constrained. It's like, how can we make sure the agent doesn't do anything that we didn't pre-approve? And I'm like, well,
00:26:48
Speaker
Is it really about pre-approving or is it more about putting them in a space, possibly with other agents, a playground, not a sandbox, where they can do different different stuff? They can freestyle in that playground, not on production, but in that playground.
00:27:04
Speaker
They can do certain things. They can write code. They can maybe execute more tool calls, but you've got auditability. You've got traceability. Maybe you've got AI agents who stand aside from that playground whose job is to provide oversight, who or the adults in the room who make sure that kids on the playground don't make a mess. you know I think there's a variety of ways to approach this that aren't put the age in a straitjacket. The other thing to kind of just drift off just a little bit is I've come to the conclusion that right now everybody's focused on that kind of open claw, you know aha moment, which is like, hey, I've got this virtual assistant.
00:27:43
Speaker
Is it really going to be a virtual assistant like OpenClaw for operations people? I don't think so. Is it going to be ChatGPT or, you know, Claude? I don't think so. See, the thing is, is the difference between the OpenClaw use case and the software developer use case is that operations is inherently a team sport.
00:28:04
Speaker
Yes, developers have teams, but a single developer can do what they need to do to get things done in isolation. They don't need the team. Operations people, hopefully I don't sneeze, sorry. um Operations people must have a team. If you're running something in production 24-7, you're not staying away 24-7. Not only that, but when something happens, if there's an in incident event, you need multiple eyes on it from multiple angles.
00:28:34
Speaker
Right. So you it it operations is inherently a team sport in a way that software development and entrepreneurship and many other things just isn't. And so in a team sport, you've got to have these agents be able to talk to each other. You've got to have them sort of have the same situational awareness. You've got to have them be able to come at the problem from different angles and help each other out to resolve it.
00:29:00
Speaker
You've got to work together to do new releases. So I just think that... If you're going to have a team, you know, trying to put a straight jacket on every agent is going to make that even harder. So I think there's an in-between concept that I'm calling a playground.
00:29:15
Speaker
um And I just don't think that people are thinking about that right now. They're very focused on like, let's make sure the agent doesn't do anything that we say it can't. And it's like, well... You know, rather other than a few LLM um calls in that case, you know, why didn wouldn't you just you know use traditional deterministic software and make Is it because of just the trust rate? I mean, in my opinion, trust is earned as as you rightly mentioned, start with a know visibility approach, then go for you know certain actions and stuff.
00:29:50
Speaker
Trust is always earned and it takes time and nobody knows what the you right process looks like or AI in operations looks like. They are very like reluctant to trust it and that's the that's the exact problem why people are forcing AI to be a guard railed or a sandbox thing.
00:30:12
Speaker
Yeah, i mean it's it's right not to trust it. But, you know, sure... i'm not sure I'm not sure that it doesn't even I'm not sure that doesn't translate to human human interactions as well. Right. Trust, but verifies one of those old things. And I, you know, and I do think that um it's right to not trust the agents. But if we're going to get leverage from them, if we're going to get 10x operations, if we're going to dramatically reduce toil,
00:30:43
Speaker
We need to figure out the areas to to trust them or at least empower them. And, you know, there is a lot of stuff that operations teams do that is not on production systems.
00:30:56
Speaker
A lot. yeah A ton. You know, inventory management. What are all the things that I'm managing? Where are they at? What hardware is it? What cloud resources is it? you know What software is running on them? Can you get me an S-bomb from you know my fleet of Kubernetes kubernetes clusters running in the US East 1, Amazon, so that I can look at the latest CVE and see how it impacts us? like All of that could be done in a rapid fashion in the same way that we do code reviews in a rapid fashion by AI agents. So there there's there's massive amounts of operational toil.
00:31:39
Speaker
if you just If you take aside doing any you know production interactions other than read-only, there's like probably 75% of what operators do is toil that could be taken over by AI agents.
00:31:54
Speaker
Right. i and And not only that, but like, I still think that people are like not using their imagination. If the if the velocity of CVEs is going to increase and your time to address them is going to shorten,
00:32:08
Speaker
then having some AI agents that like monitor CVEs, monitor news, and automatically you know do an analysis against your current systems to determine your risk and then deliver that to a human for evaluation, that could be cut to like five minutes, which gives you longer time to respond.
00:32:30
Speaker
right If you take like a full day to figure out there's a CVE and determine that it impacts you, you'll last a full day. Whereas if an agent does it, it can be done in five minutes or less. so So there's just places here where people have to use their imagination. And I don't think it's very hard. You just have to be willing to understand that like all of the old ways of doing operations, pretty much probably all of those are going to be thrown out, you know, over the coming five to 10 years.
00:32:58
Speaker
I also personally feel if you restrict AI on certain things, if you... Don't try to change or be very rigid on certain traditional approaches. Things are not going to cut out because things have shifted. And rightly mentioned, it this is a canon event.
00:33:17
Speaker
If you will not adapt, somebody else is going to and that is going to impact you in certain ways. Developers have already adopted AI and now their output is going to be your input for operations.
00:33:30
Speaker
And if you do not change, then that's going to... be a very massive shift or pressure that you'll start feeling soon. By the way, what do you think is the timeline that you see autonomous operations will start to happen very or a common thing in the industry?
00:33:52
Speaker
Is it two years, 10 years? i don' I don't think I can call it a timeline. I think it'll be very slow and then suddenly fast.
00:34:02
Speaker
Just like clock, what was that? OpenClaw. Yeah, I mean, it kind of goes to um what I've been you know kind of hitting that over and over is more of an idea of a systems harness or an operations harness, right?
00:34:16
Speaker
Like when I'm using, you know, Cloud Code and, you know, it's to the point where you can do these things like you can set a goal and, you know, so you can tee everything up. You can build a plan, a specification for what you're going to build. You can create sort of like the test plan. you can create end to end live scenario testing so it can validate that everything works.
00:34:39
Speaker
You can set up adversarial reviews. um so that you have other, use other models, like I use Codex and Gemini to do an adversarial review, and give you feedback on the code base.
00:34:51
Speaker
You can set up a architecture reviews with agents and and you can put all that together and then you can fire cloud at it and say, you know your goal is to make sure all the end-to-end test pass and all the feedback from adversarial reviews and architects is integrated.
00:35:09
Speaker
And it'll run that thing as long as it can overnight for hours and hours and hours until it comes to a conclusion for you to look at. And and and that's all the harness.
00:35:21
Speaker
That's all the harness. Where's the harness for operations and systems where, you know, you go to make a change in production and you're following the path and the AI agent pops up that whose job is change management and says, did you get sign off?
00:35:39
Speaker
Did you talk to whomever? Did you follow the process? And it walks you through it all and maybe notifies in real time and helps you get sign off if you need to make the change. Where is the AI agent that watches what you do to make sure that there's a full audit trail so that if a mistake is made, because operator mistakes are and a big cause for production incidents, that you know there's ah there's a way to go back and do the blameless postmortem and understand what happens so we can make sure it doesn't happen the next time. There's all of these sort of like facets to how operations is done, you know incident response, running war rooms when there's a sub one, you know all of these things that you know aren't in any kind of harness anywhere. They're in institutional knowledge, tribal knowledge. They're in ah papers and presentations, but they're not in the harnesses.
00:36:37
Speaker
They're nowhere in the harnesses. I've gone over Claude Code system prompt with a fine tooth comb. you know And then more importantly, like if we look at like a lot of the traditional operational tribal knowledge, institutional knowledge, documented ways of handling things, a lot of it's old in the tooth at this point. It doesn't take into account what you can do you know in you know sort of an AI era, right?
00:37:02
Speaker
And so you know there's a lot here to change for operations. And I think it goes from how we build and deploy systems to how we operate them in real time, to how we provide visibility and transparency to the rest of the business. like I think all of that gets impacted and changed.
00:37:26
Speaker
I mean, there's so much to say there. you know i it would take us a bunch of podcasts to get through it all, but as one poor example, Immutable operating systems and trusted computing platforms have been around forever now, and they are not broadly adopted because the level of effort and pain to sort of deal with them is very, very high.
00:37:48
Speaker
But if you can train a bunch of agents to basically handle a lot of that for you, to know how to troubleshoot it, to deal with the common problems that occur, to resolve them automatically, Maybe those can be adopted.
00:38:00
Speaker
And then that takes away a whole class of vulnerabilities. Hey, there's a CVE that affects X, Y, Z. They got to be able to write to the file system. That CVE is gone.
00:38:10
Speaker
Move on. We'll patch it when it makes sense. So like there's a, when I say rethink the whole stack, the whole way of operations, I mean the whole thing. soup to nuts, like all of the assumptions that operations people have today about operations, I think they're all largely invalid in 10 years.
00:38:30
Speaker
Okay. Makes sense. I haven't seen anybody with your experience, ah Randy, doing coding anymore.
00:38:42
Speaker
But you showed me a full application to manage a rack you built overnight. And how do you see there is a shift of this? You were mentioning you at your... know code commits are increasing this year a lot.
00:39:00
Speaker
And do you see a shift with AI that has given you more opportunities? Let's say your personal interns just working for you, you're thinking and they're doing... no It's the other way around. It's all the experienced guys. ah Sean O'Meara, our CTO at Marantis, built this incredible bomb calculator for building AI factories that allows you to basically select your GPU clusters and systems, your network topologies. It helps you build like the bombs. like it's It's pretty sophisticated. Jerry Ibrahim, who runs our Cordon AI engineering effort, he's using the AI tools. Our CEO,
00:39:41
Speaker
Alex Friedland, who does not have a technical background, has used Cloud Co to build a Minesweeper game. It's the more experienced folks who have an open mind um who actually see the value of the tools. They have less time, but they've got domain expertise and knowledge in areas that that you know, or ah that just the youngest people don't have. I actually see younger people struggling more because they don't have the domain expertise.
00:40:09
Speaker
I was thinking about this the other day, like if you are talking about AI assisted coding, there's a lot of discussion around humans have something that the AI can't do. They have taste.
00:40:21
Speaker
And when I think about operations, I think about, you know, the main thing that humans have that, you know, agents don't have is judgment. Right. Something's going on in production. Seven, one you're down, you're losing millions of dollars an hour. You know, you have to make a decision about what you do. Roll forward with the fix that's unproven, you know, or try to reproduce it somewhere else.
00:40:44
Speaker
Well, an AI agent, you know, can't make a good judgment about that. It doesn't have the business context. It doesn't know about the risk rewards. Even if you were defeated into it real time, you couldn't manage it well enough so that it can make a decision. And it can't take responsibility for that decision, which is most important, responsibility and accountability.
00:41:03
Speaker
That's like a team, a human team effort that's got to happen because the team's responsible for the business. And so, you know, I don't think that judgment can be pulled out of operations and handed to agents. So there's no way for humans to be pulled out of operations. So the fundamental thing here is that, again, you know, I don't think that we have even gotten to the point where we understand exactly how it's going to impact operations. The
AI's Potential and Knowledge Management
00:41:29
Speaker
only thing I'm certain of is that it is going to transform it completely top to bottom. People often talk about, i mean, the recent trend is about context.
00:41:40
Speaker
Any AI agent would perform more accurate, more better if you give it correct context at the current correct time. Do you think building this or converting this tribal knowledge and institutional knowledge into something more consumable by agents is going to help?
00:41:58
Speaker
Yeah, I mean, that's what I was talking about kind of with my my pet project, right? Like the architectures manager's job is to create kind of the the the overall piece of that, which is that there has to be- A virtual view of everything that's running, yeah. Yeah. There has to be a way for the shared memory of humans in operations teams that we've had in the past that is a combination of the tribal knowledge and the written knowledge to be into some kind of living set of documents
00:42:29
Speaker
in whatever format you think makes sense, markdown, rag, you know, whatever works for you that, you know, is shared between the humans and agents. And as it updates, whether it's an agent or human, you know, as the system updates, you know, the the enshrined memory and knowledge of the system updates. I mean, that I don't think there's any way forward except that way.
00:42:53
Speaker
ah Do you think only technical information is required or business context would also be beneficial for building such tools? Yeah, i absolutely think that some business context makes sense. I mean, you take the example of having sort of a SOC engineer agent that, you know, sees vulnerabilities and, you know, makes some kind of initial assessment on them. It it may be wrong. it It's fine if it's wrong. What it's doing is it's saving a bunch of time on the initial assessment as if it was a junior SOC engineer. But giving that SOC engineer, you know, some kind of context about ah the
00:43:29
Speaker
the business, especially at the intersection of the system architecture is important. And so I'll give you a concrete example of that. If we were to look at, say, eBay systems, not all API endpoints are the same.
00:43:44
Speaker
Some of those are going to have a much more dramatic impact on you know revenue than others, right? Like an API endpoint that say lists inventory for a small a shop managing their inventory on eBay. That's that's you know an impact to that set of customers and it's an inconvenience, but it's not directly impacting revenue.
00:44:06
Speaker
So being able to describe not only system, but you know how the system interacts with business and you know what parts of the system are more vital than others is absolutely something that needs to be enshrined so that some decision making can be informed by it.
00:44:23
Speaker
And also one thing, Randy, is what I feel is code is becoming fungible now. You don't have to pretend to have you know one architecture and a a quarter of product shipment.
00:44:39
Speaker
That waterfall model or even the sprint model is no longer valid anymore. you You can scrape it out altogether and build a new one overnight. Yeah, when when code is cheap, what happens? What I used to say back in the day on the around the cloud stuff, and I think you'll you'll like this, is i you know, was a little bit before the pets and cattle thing, but I still use it for a long time. As I said, you know, you have to ask your developers, what can they do if they can have 10,000 servers for an hour for a hundred bucks?
00:45:09
Speaker
And it turns out there's a whole bunch of things that unlocks because they couldn't do that before. And now there's a similar question, which is, what does it mean if you can generate 100,000 lines of code in less than a day that does a specific thing, you know what does that unlock for you? And and you see people doing it all the time now, like like ah you know proof of concepts,
00:45:33
Speaker
cost nothing, absolutely nothing. And you can prove it out in less than a day a lot of the time. So, you know, it's just there's just it's just going to have a major change in the way that we think about how we do systems engineering, systems thinking and and build even businesses.
00:45:52
Speaker
Awesome. I like the line. Absolutely. This is what we also, I was learning in one of the mentoring sessions for, um as a founder, right?
00:46:05
Speaker
If you have all the million bucks that you need and what are you going to do? Think think in those terms. of this This totally like opens up a lot of possibility and even i personally don't have an answer what would I do if everything is you know at my fingertips and it's just that sky is the limit, whatever I can imagine, i can deploy, develop at an instant.
00:46:33
Speaker
So it's very interesting and very interesting times to be in at the phase, right? At Mirantis, what is happening, Randy?
00:46:46
Speaker
What do you mean? In terms of what? What are you doing? Anything in particular? What am I doing at Mirantis? I'm doing a lot of stuff. You know, we've made this pretty hard pivot over towards really working on building AI factories initially for neoclouds who are adopting those, eventually for enterprises as well.
00:47:07
Speaker
um You know, we are... You know, it it it's great because the executive team, especially Alex Friedland, our CEO, Boris, who is ah one of the original the original co-founders along with Alex, and now is kind of advisor, and Sean, you know, our CTO and my boss.
00:47:31
Speaker
you know, all kind of came to the conclusion, you know, after sort of a weight of evidence that we needed to be kind of a AI forward. I'm going to avoid AI native, but an AI forward company.
00:47:41
Speaker
And, you know, when they signaled and unlocked the gates to the rest of the business, people really leaned in hard on that. and you can see everybody at Rant is really trying to use the tools, even people who I thought wouldn't.
00:47:54
Speaker
um And, I think we're still trying to figure that out, but it's it's it's changed the tenor and and tone of kind of what we're doing and how fast we're moving. I'd say we're kind of a little discombobulated because it's early stages of broad adoption. Some of us have been at it for a while, but now like the whole company seems to be trying to use it.
00:48:18
Speaker
And how that's result what that's resulting in is a business that is trying to reinvent itself kind of from top to bottom. And I think largely getting there, but you know, only time will tell.
00:48:34
Speaker
And so
Mentoring and Generational Impact of AI in Tech
00:48:35
Speaker
for me, it means that I'm, you know, because the AI factor changes, I'm getting to really work directly with some of the data center grade GPUs. And it is really fun to have GPUs at your disposal and run DeepSeq V4 Pro you know at scale um to run Quin3, Quin36. Those are all really impressive new open weight models.
00:49:00
Speaker
It's, you know, the second thing it means is that, you know, as one of the early adopters and one of the people who's gone further and further further and farther with AI agents and agentic coding than others, like I'm getting to mentor and and help others with the adoption process, which is great.
00:49:17
Speaker
And then the third, which is the thing that always gets me pumped up, which is why I'm here on this ah podcast with you, is i just know that AI ops or agentic ops, whatever you want to call it, is going to be a big deal.
00:49:30
Speaker
And, you know, for anybody out there who's listening, I'm usually about three years early to like ah with my ideas. So what I do is wait about three years and they and do an agentic ops startup.
00:49:43
Speaker
Sorry, pretty young. um because that'll probably be about the right timing. I don't know why I'm always three years too early, but, you know, I always follow my nose and, and, you know, my nose right now says the AIOps is going to be really interesting, but, you know, I'm, you know, and Priyank are pretty far ahead of the curve, I think on this.
00:50:02
Speaker
um But that area is just keen to me. i I goes back to that story I told at the beginning of the podcast of like walking in and finding everybody doing everything manually and I automated everything. Like if computers can do something for me, i want them to do it for me. I don't want to do the boring stuff.
00:50:19
Speaker
I have zero interest in this. Like I just really have zero interest in this. So, you know, I think lazy operations guys, just like lazy developers are the best.
00:50:31
Speaker
It means that they're going to put the effort in to make their lives easier. And, you know, we should unlock those kinds of personalities and let them go at it. And, you know, to a certain degree, that's somewhat my role in the business. I've got a lot of other stuff to do, but, you know, Sean has asked me to really focus on you know, AI ops, especially for AI factories and what it means.
00:50:53
Speaker
And, you know, I think the first kind of like, you know, dark factories that we'll see, you know, that are mostly run by AI agents on the operations side will be the neoclouds, almost certainly.
00:51:06
Speaker
They've got to get to a certain amount of scale, you know, and they've got to be sort of the power plants that feed all everything else and um operational scale. you know the lesson learned you know from the hyperscalers in the cloud days is that operational scale comes from ah intense amounts of automation, building ah ah homogeneous power plants. right Google famously was managing 10,000 servers with one admin and trying to get to 100,000 servers with one admin.
00:51:38
Speaker
NeoCloud similarly have to have kind of that same kind of level of scale in order to succeed and to compete against the hyperscalers who already have that as DNA. So you've got to bring AIOps in to basically enable these folks. That's why this acquisition by IREN for Mirantis is so important. We can bring in a bunch of that DNA And I'm hoping you know that part of what I'm going to contribute to our future together is being really on that AI ops side and the guy who knows how to get 10x on our operations folks, even possibly 100x.
00:52:12
Speaker
Awesome. This is awesome. Talking about the general generational impact that you have made, Randy, what is the handoff going to look like for you?
00:52:25
Speaker
Or is there going to be a handoff, first of all? ah Yeah, um I don't think I've really seen handoffs. um You know, it seems to be fairly smooth transitions because and between any generation of um operators or tech people you see,
00:52:48
Speaker
as many people who lean forward as lean back. I've got a good buddy of mine, Jeff White, shout out Jeff, um who is a bit older than me.
00:52:59
Speaker
And he's like all in doing all kinds of crazy stuff with agents, um building teams, building communication systems between them. um and um you know, i think that,
00:53:13
Speaker
In the same way, this is going to sound like a really crazy ah analog, but in the same way that kind of smartphones kind of empowered older people who decided to ah use them because it helped them do you know get through life a little bit better in lots of ways. when And we'll see even more of that now with ChatGPT. I mean, And I think that you know older folks who, you know even older and me in their 60s, 70s, who um embrace and use new technologies like some smartphones find out that you know even though they're scary at first, and they're actually way easier to use as computers than computers have been. and and you know, LLMs and AI democratizes it even further.
00:53:55
Speaker
I think in that same way in the tech industry, you see some, you know, older generations than me um who are more empowered because ah maybe they want to do a certain kind of project and, you know, it's just easier to do now that they've got the tooling, and but they've all got all that domain knowledge. Like I said, kind of judgment, expertise, domain knowledge, wisdom. Those are things that AI doesn't have that,
00:54:21
Speaker
that older folks bring to the party. i actually think the challenge is harder on the younger generation. I really, really worry about um some of the technologists who come in. i had a real young gentleman working for me, a couple in the Philippines, and they had a real hard time really using the AI tools in like ah compelling ways. They wanted to abdicate decision-making to the ai It's like if you're going to be a scientist or not a scientist, a mathematician, you know, abdicating all your calculations to a calculator, as opposed to learning more about the domain of math and getting to the point where you can use something like R to do very complex things with mathematics based off of your understanding.
00:55:10
Speaker
And um I don't think, and this is my personal bias, um that schools do a very good job of learning Teaching most schools, not all, but most don't do a very good job of teaching important things um like critical thinking, um domain expertise. They focus more on sort of rote memorization and regurgitation, especially at sort of like that high school level. When you get into universities, more elite universities are better, but they're still an over-index on research.
00:55:43
Speaker
ah knowledge as as opposed to wisdom, um regurgitation of information as opposed to systems understanding, systems thinking, critical analysis, you know, those things that would actually empower somebody to be more effective using the AI tools. You know, one of the the problems I saw with a lot of young adopters of AI tools, again, like I said, is they abdicate their thinking, do this thing for me.
00:56:10
Speaker
As opposed to explain to me how this system works. Explain to me how XYZ language works. You know, you can you can have them.
00:56:20
Speaker
You could literally sit there for hours and hours and hours and have... a AI tool walk you through, say, how the Go language works, how the internals work, how the compiler works, you know, the difference between functions and interfaces, how memories manage. Like you could have it basically train you on how to use Go better, which would make you better using the agent to write Go code.
00:56:47
Speaker
But I don't see ah younger folks thinking in those terms. It's like, oh, I'm just going to have a go write this function for me. And then they can't really understand what the function does. um you know I am regularly regularly using the tools to to explain to me how systems work that I don't know. is ah For example, I was never much of an expert at Ceph.
00:57:09
Speaker
ah the distributed storage as software SDS. um And, you know, I needed to bring stuff up for some of the AI factory labs that I've been tasked with ah building for Mirantis. And, you know, i wanted it, I could have it go build the stuff clusters for me, but I started with,
00:57:27
Speaker
you know kind of explain to me how stuff works and the difference between its pools and you know its clusters, which is different than a lot of other storage systems. And really walking through and having to compare against storage systems that I knew really well about its idioms and, and you know, ah the way semantics and the way it does things and the way others do things so that I had a better architectural understanding of the way stuff works so that I could make better design decisions about how stuff would be laid out on the systems I was building. and And that's just normal now. That big part of how I'm using the tools is to have them educate me on how to do things. And I think younger folks don't understand that that's where they need to start.
00:58:11
Speaker
do Do you think this is because so one of the. How do I phrase it? um Approaches that we have learned is by doing stuff, by experiencing the problems and doing stuff. Younger folks did did not even have that experience. Now everything is, you know,
00:58:32
Speaker
very garnished and polished and given to them in a very easy way that they never learned it the hard way Is it going to be a bit difficult for newer folks to adopt things?
00:58:46
Speaker
i I think there's so there's this notion of a J curve, right which is that you start to adopt these tools and and and you do really poorly at first. And then you kind of as you go down, you kind of you know learn enough that you come back up and then you sort of get productive eventually.
00:59:02
Speaker
I think for older folks, the J curve's got a very shallow dip. And for younger folks, it's got a very, very deep dip where they are ah not very productive and they are making mistakes they can't see. um And it it'll take a while to bottom out.
00:59:19
Speaker
And at the same time, they probably you know are so AI-pilled that they think that they're like, you know, doing amazing work just because they're valuing the quantity over the quality.
00:59:33
Speaker
um In my personal experience, the only way to learn anything is to break stuff. you You must break things. um I think the danger in AI land is that you can't see what you're breaking if you're not the one breaking it So if I had any piece of advice, it would be to really try to deconstruct things and to get into the guts, especially as you're learning about what you're trying to build, um about like how it works and to actively break things, to do experiments. like If you're just running to kind of the end goal constantly, you can't see the messes you're making. They're hidden.
01:00:14
Speaker
Whereas if you're doing kind of like you know separate spikes and experiments at different times you know on on sub-components so that you understand them, um you know you you will learn more. As a, for example, say you're building a component of a piece of software that um is an API that um you know is used to control some subsystem. i Take some time and run an experiment to see if you can break the API.
01:00:44
Speaker
Learn about fuzz testing, learn about and you know ah security and how you might attack the API. Spend the actual time to use the AI tools to help you try to break your own system. Have that hacker mentality because if you can have that hacker mentality, you could try to break your own stuff, then you're going to you know learn a bunch through that process.
01:01:07
Speaker
um So I think it's really easy to get on that dopamine treadmill and like, I'm making lots of progress and it all works. And not to realize that, um you know, under the covers, you know, you're maybe making sort of a house of cards that could blow up. If you come at it with the assumption of that you are making a house of cards and you start to say, how can I blow up the the stuff that i'm'm I'm working on?
01:01:31
Speaker
um Then you're more likely to ah get the learning that's necessary to level yourself up faster and shallow out that J curve on adoption and get to a point of being extremely productive.
01:01:43
Speaker
I
Career Reflections and Domain Expertise in Tech
01:01:44
Speaker
think you answered it in the in the previous question itself, but still wanted to hear. Let's say you're starting your career today as a 25-year-old.
01:01:58
Speaker
And what would Randy do? or How would he approach the industry? i am I'm not really sure how to answer that question, honestly. It's a little tough because I think...
01:02:13
Speaker
You know, I kind of grew up at a time where I was really steeped in computers. I got access to them at a young age. And it was at that time when you had to kind of get in and understand the guts of the computer more than you do now. Right. I mean.
01:02:27
Speaker
these These phones you know are big, huge, monstrous computers compared to what I grew up with and super capable, and you need no knowledge to use them. So i don't i I'm not sure I can put my myself in the shoes of a 25-year-old. All I can do is observe some of the you folks like the Filipino guys who worked for me who were at that age. and And the thing that I observed is that their their computer fundamentals were not great.
01:02:57
Speaker
didn't really understand how CPUs work or RAM, how CPUs and rams interconnect how cp and RAM how storage systems work. um I don't know if that's the Philippines or more broadly. I'm i'm not sure um what it looks like because I haven't interacted with a lot of 25 year olds um so there's a there's a there's a strength and a weakness to being 25 year old coming into the industry so so the strength which to a certain degree is also the weakness is that they don't have preconceptions and so they're not they're going to adopt the new tools right out of the gate and and and they're going to be ai pulled like right out of the gate and i i that's that's a big strength
01:03:43
Speaker
A lot of older folks aren't going to do that and they're going to hold themselves back. But it's also a weakness because not understanding what's happening underneath is a problem. And like I said, like, um you know, I think that spirit of curiosity is the biggest thing. When i when i started, when I was in the industry in the early days,
01:04:07
Speaker
I just had this crazy love of computers and everything about them. Like I, it had been my hobby as a teenager. So, you know, I just, I wanted to know everything. I remember in 1992, I'd been using the internet since about So I'd been using it for six years as ah 21 year old and, you know, which was really abnormal.
01:04:32
Speaker
um And um I saw, ah you know, ah the internet getting talked about on TV and and nobody knew what the internet was. And I was like, holy cow, man, this thing that has been amazing for me.
01:04:45
Speaker
is now going to be a big deal. And I was like, I don't know anything about networking. I want to go learn about networking. I deliberately went and found a job to at an ISP so I could learn networking like really deeply. I had been doing some networking as a sysadmin, but you know I didn't really know it. So I i i was like, I want to do networking.
01:05:04
Speaker
So I did networking. And then at some point I was like, I want to do security. I've done a little bit, double and there, but i really want to do it. So I went and I did security for a while. And that's the way it always was. Like I found some area that I was like super interested in. I'm like, I'm going to go figure this out I'm going to go learn it.
01:05:21
Speaker
And I'd say that's the thing. If you can come at this, more like I want to learn and understand this domain rather than I want to deliver this app and make X many dollars, you know, by putting it on the app store.
01:05:35
Speaker
Then you're one of like, you know, a million other 25 year olds who think they've got the greatest idea in the world for a new app that's on the app store. And because you're AI powered, you think you can go like crank this thing out. And if instead you're really focusing on the problem domains and learning the problem domains, I think outcomes are likely to better because by understanding problem domain, when you actually tools you need, you're more likely to succeed.
01:06:15
Speaker
for the podcast was placeholder, Randy. I'm choosing spirit of curiosity as the title now. All right. I like it. Let's do it. Awesome. of All right. I think we have spoken a lot about the practical use cases of AI, spoken about the patterns that you were seeing from your career journey versus what you're seeing now, how do you think the future is going to be driven?
01:06:51
Speaker
An autonomous or AIOps approach, you're saying three years, you're early for three years, right? So and within three years, we are going to see a lot of automation and AIOps based systems, right?
01:07:06
Speaker
ah I think we covered a lot and I really love to you know talk a lot more with you, ah but I think we should confine it here.
01:07:20
Speaker
We could do our episode two. Yeah, we're going to do another episode. People are going to get bored. not Not from you just seeing my face. um Final messages for my audiences who are both young, new to the industry. I think you covered a lot in the last ah section, but coming new to the industry, what they should be aiming towards and some serious folks who have been reluctant to adopt to AI, why they should see things a bit differently now.
01:08:00
Speaker
I mean, I i guess I'm, I guess I feel like everybody kind of finds their own path to a certain degree. um And in a way, AI tools empower that.
01:08:11
Speaker
So maybe your path is like jumping on the bandwagon and trying to deliver, you know, the first app on the app store that allows people to schedule their dogs for dog wash because you're going to make a billion dollar business out of it. And you've got to go down that path before you figure out that path doesn't work. Or maybe your path is to, you know, kind of, um, uh, attach yourself to a mentor and, you know, level up slowly through, you know, kind of an internship kind of, um, apprentice master type, you know, thing. I can't really say, i think everybody's a little bit different. I personally, when I was young had to, and still today mostly i have to just teach myself everything. I'm not good at taking like,
01:08:58
Speaker
courses and things like that. But everybody's a little bit different. What I would say is kind of to your point um about the overall theme here is that, you you know, you do have to keep your curiosity engaged. You do have to go understand how computers work and computing systems and systems theory and critical thinking.
01:09:19
Speaker
um You do have to go kind of establish a bunch of fundamentals. It's, it's like, it's yeah I'm going to give like a a terrible, terrible analogy, but imagine that, you know, suddenly everybody could have an exoskeleton that made them really strong and very athletic and fast, and they could jump really high so they could all be in theory, NBA basketball players. But if they didn't have foundational basketball skills, know,
01:09:53
Speaker
They're going to be shit, NBA basketball players, because the leveling of the playing field empowers everybody equally. So while you may suddenly feel superpowered, the people with the domain expertise are just as superpowered.
01:10:10
Speaker
So the only way that you can actually get to a point where you can be as successful as you want to be is to go get the domain expertise. And part of what I'm saying is that if you're the real smart AI peeled person that you think you are, you'll use the AI to help you go get that domain expertise, which it is more than capable of.
01:10:34
Speaker
It can teach you how to understand these systems. But if you just use it to do your thinking for you, if you abdicate responsibility for the domain expertise to the AI, then you'll only get what it's got. And what it's doing is it's regurgitating the existing sum of human knowledge.
01:10:55
Speaker
right? If you want to be able to go past the edge of human knowledge, if you want to break new ground, you personally have to have the domain expertise. So you can build it with the AI tools, but you can't abdicate it to the AI tools.
01:11:10
Speaker
And so that's the main thing I'd say. And then to the old farts, the graybeards, you know, whatever you want to call them, um to the folks who've been around the block,
01:11:21
Speaker
and um That domain expertise you have is the thing that's going to you know make you super powered by using the tools and you've got to go leverage it. you know i spent a long time with my brother. I'm going spend some more helping him kind of understand what's going on. um But you know he really knows security operations. And so when he gets up to speed, he is going to be like a super badass using the tools.
01:11:46
Speaker
And excuse me And so I think that it's important for everybody to start you know leveraging the tools because you have to build the muscles, you have to learn.
01:11:56
Speaker
um And if you've got the domain expertise, you're going to be a badass pretty quickly. If you don't, you use it to level yourself up and then you'll be a badass, maybe not quickly, but in relatively short order.
01:12:10
Speaker
And just for the spirit of curiosity, thanks a lot, Randy, for taking out time for the podcast. I'm pretty sure people are going to love it and they will have a lot of more questions.
01:12:27
Speaker
Be ready for a episode two of of the same. And whatever questions do come, I'll send it over so that you can answer them on social medias.
01:12:38
Speaker
right. Awesome. Appreciate you, Priyank. It was great. Thanks. Thanks a lot, Randy. Thanks for your time. All right. Take care, man.