Techstrong Gang June 10, 2024

Hey folks, welcome to the latest edition of the Textron Gang. We're gonna be talking about the United States government seems to wanna investigate these deals between Microsoft OpenAI and throwing in Nvidia for good measure. Then we're gonna talk about Kubernetes because, well, it finally turned 10 years old.

And next we're gonna chat about Cisco and what they're up to, because it seems like they got their fingers and everything, and now they're adding a little ai. We'll be back in a minute, and we're back in. We have a little special edition today.

It's just me, Mike Ard and John Willis, who's home somewhere in Alabama. Is that right, John? Let's try Auburn in Alabama.

How are you? Go. All right.

And we're gonna be talking about these issues for the next 30 minutes or so, and I would love to get John's opinion. You've been following this whole AI thing from the very beginning, but we are hearing now that the FTC and some other agencies are starting to look at these deals and relationships between Microsoft and open ai, and, um, all they want to throw Nvidia. This starts to feel a little bit like, you know, Wintel all over again.

But what's your sense of, you know, is there, is there a case to be made here or are they just fishing? I had dance is both right. I mean, you know, it, it's, the stakes are too high not to pay attention, right?

Um, the negative is it's so early, right? This is the, you know, I mean, I, you know, I went back and I looked at some of the, um, you know, what was going on for last year or two on, you know, what did, you know, FTC and DOJ and, and, and they're ever since sort of cloud computing even like, you know, internet, they, their job has been hard, right? Like, you know, I was going back and doing some research on like, you know, what's the responsibility of the FTC versus d oj it's this classic government confusion, right?

Like, and, and again, I I'm not like an anti, the government sucks at everything, right? But, and I, I'll give them their credit, what credit is due it, like the world has, you know, when it was at and t right? It was sort of straightforward, right?

Even early on with, you know, some of the stuff they had to do with the sort of early, uh, you know, sun Microsystems and what was going on and all that was a false start, right? But, but now, you know, when you get into sort of cloud computing and like, and, and then like what, what Facebook is doing advertising, and then, you know, then you Google search, and even that's child's play to what we're talking about now, right? And like, if you look at like, the responsibilities, you know, like over the last at least decade, the lines were kind of blurred between FTC and DOJ.

You can go back and I've read a little bit of history, so you, the prompted of this discussion, like, okay, I'm a geek. Let me go find out what's going on here. And you know, what's supposed to happen is FTC is mostly be more consumer related, healthcare farmer, professional servings, consumer goods, and d OJ is supposed to, it's, it's clearly laid out like lines of responsibility, do o j's, um, telco, transportation, financial services, energy, right?

But then like when these blurred lines started happening, it sort of shifted to, well, DO j's more tech do, OJ is more technical, so we should give them the, you know, the sort of, uh, the, the Google search, or we should give them, you know, these things that are highly technical. And, and, you know, let's face it, most of the government agencies don't have the expertise really to interpret this stuff. But now you get into things like, you know, this started a couple years ago where it's like, you know, um, or at least more than a year ago, Amazon and Anthropic, Google and Anthropic, I mean, and the one in the article that we're looking at, that we looked at, which is, you know, Microsoft and open ai.

And that's an interesting case, right? So all this to say, one, it's so early, it's really hard to, to really put sort of some type of control, you know? Um, and I, I'd like to talk a little about the difference between NIST a government control and a wasp.

And in my opinion, NIST is, and I know this is sort of antitrust stuff, but NIST just my opinion, doing a terrible job. They're, you know, they're comparing, their findings are basically, you read them, they sound like they're talking about Google search or what's wrong with ai, right? Or all the things that can happen.

Like, yeah, well, that Google search, that's Google search, you know, and, and Oasp is doing an amazing job. I think, you know, the, the, uh, LLM top 10 is like, I think it's spot on. You just put out a paper.

But sort of back to the, you know, this, this whole thing, it's really complicated. So one, it's so early, um, you know, uh, you know, two, they don't really have the expertise. And, um, you know, I think that, you know, the, the other thing is we don't know the combinations.

You know, like, we don't know, like what we think right now is a blockade. Like maybe, you know, I mean, on a Wednesday, everybody's convinced, including myself, that Microsoft is gonna own this whole thing. Then Thursday, you know, Google drops something and you're like, oh, well, maybe not.

And then, you know, then you find something from meta or, uh, what is in intro, right? Like, so, you know, I like, and let me just say one thing. I was listening to the Stanford, um, workshop, which really interesting.

Unfortunately, it's a nine hour, uh, video, but it was, uh, a workshop that they ran to really understand all these complexities, right? Like covered agencies. And they had this woman from, uh, from uk, uh, basically a chief data officer for, um, basically what they call the CMA competitive markets and authority, right?

I guess sort of like the FTC, right? Um, and it's one of the things she said is the real fear here is given that the strength and the bulk of the big players, it could tip at any moment. So, you know, like, like, like, like we could be feeling really calm and then just like a couple of things by Microsoft and open AI could happen.

And like, okay, now we're, we're sort of locked into them controlling this world, To your point, right? We got Apple floating around here too, and they're reportedly gotta do something with open ai. Or maybe they might think twice about that now and maybe do a deal with multiple LLMs, but I'm, I'm not quite clear why we all feel the need to standardize on one particular provider.

It seems like there's a lot of 'em and there's a lot of choices. Yeah, I mean, it, it's, it's, it's, you know, it really is sort of an ease of use thing, right? Um, I mean, I'll be, you know, like you said, I've been using this technology, you know, I've got now four clients that I'm focused on in this space, and, and I'm, I'm using this technology and anger.

I'm doing a lot of interesting stuff. And you know, I, I, you know, I, you know, you, you sort of navigate to the one that works fastest, easiest is most, you know, reasonable price, but price is not the only indicator. And I just find for me, I can do more with, uh, with, you know, open AI or, you know, I, I actually use open ai, not Microsoft.

If I, if I was running a large institution, I would probably use my, you know, Azure open ai, which they do have guardrails. I don't need those guardrails 'cause I'm usually designing, developing, testing, researching. Uh, but the, the open AI platform, it's just easy.

And, and so why did Amazon win? Because it was easy. You like did all the things you needed when you needed, and we were there.

You know, I, I, you know, I, I always used said, I'm gonna say this, you know, you guys bring me honest to be transparent about my opinion. I'm trying to work out, um, running a, a thing, I, I can, I could build in five minutes on OpenAI. Now granted, I do it through a notebook with a Python code.

Um, but I'm trying to run that on Bedrock from Amazon. And if I didn't have to do it because of some engagement I'm working on, I would not touch that thing. You know, like, like I'd wait a year to come back to look at it.

And so, so I think to answer your question, you know, I think that, you know, people navigate to the, you know, the, the, the things that work the easiest, especially, you know, I can fake a great technologist. I'm not a great technologist. You know, I, I'm sort of a bootstrap technologist.

I can, I can fiddle with almost any technology, but I'm not like, you know, a, a TED X or if there's such a thing in any one technology, right? So, so I, I like my technology to be pretty easy and easy to get started and run, you know, you know, I do every technical stuff. But, and, and so then to the point is, those become the winners and know that could be become the tipping point.

And isn't that just competition, right? Because hey, they do have a head start and they are more advanced in terms of the tools and the integrations, and everybody else is just gonna move down the same path. And it's not like Wintel where there was only one choice, right?

There are multiple choices. They're just not quite as good. And I can't help but wonder, I'm not sure that these organizations, going back to your first point, have actual standing in these agencies to do this because, um, it's the FTC, well, the DOJ, there has to be some sort of crime or some sort of investigation.

And, uh, was there a, who knows that I didn't say. And the FTC to your point, as consumer oriented, do you think we need to go maybe back to some sort of push through Congress, which might be impossible to create some sort of oversight stuff for AI that has the legal authority ascribed to it? I, I'm glad the not so big guy isn't on this show.

When you just asked that question. I can, I can conceal it goes of balance saying, wait a minute, I got an opinion about them. But anyway, um, you know, I, I, you know, I think we have what we have, right?

Like, you know, and, and the whole thing about like the DOJ and ftc, right? They, they, a long time ago sort of created this criteria for who controls antitrust, right? So, you know, from my understanding, not being an expert, it doesn't actually have to be a crime for DRJ to be part of the investigation when it comes to antitrust matters.

So, uh, I don't know how you change that. I think UK is doing a great job, you know, and I don't know as much about like their system, but that, you know, eu U um, you know, the UK just seems to be just, you know, better informed. I don't know if they're coming out with better solutions.

So I don't know if there's some, like, again, watching this Stanford, uh, seminar on, you know, on competition and ai, and it was pretty fascinating, you know, um, you know, just learning, like, I think the idea that they, um, uh, you know, the, the, I don't know if it was the DOJ or, um, or who, who sponsored it, uh, yeah, it was Department of Jeffers who sponsored it, that's, uh, promoting competition AI at Stanford, and they brought in people from the u, you know, eu. So like, they're, they're listening, but I don't know how you change, you know, the, the institutionalized way we deal with these kind of problems, right? I, I don't, I don't know how, I don't see, There are a, a lot of folks who are calling for an open source approach to LLMs 'cause, and it's not clear to me that those LLMs are fully viable compared to the proprietary ones just yet.

But is that just a moment in time? Do you think we will land on open source LLMs as Kind of the defacto standard? I, I think, you know, I mean, the LAMA stuff is really good.

Um, you know, I, I, they're, you know, I think the open source is competitive, you know, and I, I truly, we've had this discussion on previous, uh, you know, text on gang conversations. And I, I think that, um, you know, I think there's a, there's a positive balance that like, as we invest use and, and, and responsibly try to explore what are the best options, and there's actually, there's a little bit of, you know, so yes, I, I think open source, you know, I, you know, it's an interesting race. Anybody who has a prediction is full of it.

Um, because, you know, they're, they're still talking about G PT five and g PT six, right? Like, I can't even imagine what that's gonna look like, right? Um, and or if it's gonna happen or it's gonna be renamed.

But the, but right now, I mean the, you know, the, you know, from what I've experienced, what I've seen, the, the open source models, um, almost all the orchestration tools are open source. A lot of the, the vector databases are open source. So a lot of the scaffolding to get this to work, in fact, almost all the code around how you load data in, how you create semantics around, like, that's all open source.

So even though like the frontier models, you know, like the, the, you know, the open ai, the, um, anthropic, uh, you know, um, the, you know, cohere, uh, are all, uh, I'm, I think cohere, I'm sure that they're all proprietary. I know the open AI and, uh, and the topic are. But, um, but even though they're sort of, um, you know, strong problem, like all the scaffolding to get this stuff to work is all open source.

And, and the models are the sort of meat. But the, the truth is, and then there's one other sort of, I think, you know, sort of defense against sort of like the winner take all is, uh, a lot of companies, and I think rightfully so, are still concerned about using SAS based, you know, lms. Uh, because I, I don't think it's absolutely clear that they can protect your data.

You know, they'll say in license agreement, we don't train on your data. You know, I think there are so many hidden dragons in the way. 'cause you, here's the thing, you know, open ai, I know I know a little bit about more than OpenAI, but, but I do know a little bit about, uh, uh, philanthropic, um, is that if you look at their infrastructure, they're all basically the same.

They're all running Kubernetes. They're running things like Redis, you know, they're running, they're running all these open source stack technologies. They have to run that stuff at scale, like, you know, and so, um, so like, we've already seen some weird, uh, manifestations of like Python libraries that sort of have a vulnerability, or it wasn't even like a bad actor, it was just the code started like leaking and sharing that, you know, buffers got overloaded, and the next thing you know, um, somebody else's data is showing up on an open, you know, on somebody else's chat, GPT, right?

So you have no idea. And so long story short, there's a lot of large corporations that are still not convinced that it's safe to allow, um, other than go finding the best restaurant at your corporate headquarters, chatbots to put, you know, real sensitive high consequence data in these SaaS models. So that pushes the open source as well, right?

That pushes the sort of the, the, you know, the, the, the meta, um, you know, and, and you know, some of these, you know, these different open source tools. Now you gotta run on-prem, right? And you run OnPrem.

And, and I saw something the other day where, uh, somebody was showing the cost model of even with getting your own GPUs and your own sort of, uh, technology on-prem hosted somewhere, you know, in a sort of a service provider, but it was, it o overall, depending on use patterns, it was actually cheaper than running on like Azure open ai. So, so that's gonna also push the gravity towards open source, is that it's still not clear, you know, what you can and what you should and what's your regulatory, I mean, we, we haven't even gotten into discussions. Like we're, they're still trying to figure out antitrust and competition issues.

I mean, I don't think they've even gotten close to like, regulatory control. I mean, there is some discussions about like, early, early discussions about banking, but like, you, you know, as complicated it is about antitrust. It's like an order of magnitude more complicated to figure out what technologies, how fast they're moving, what it looks like today.

How do you put regulatory control I long it took to get regulatory control of just using clouds for banks? Mm-Hmm. And that's just our computer storage.

Well, I, I don't think any of us know how this is all gonna turn out, but I kind of wish they would just cut to the chase and get to that operating agreement that they came up with for Wintel and all that stuff back in the day, and use that as a model because, well, uh, uncertainty is bad for everybody. So let's just, we already kind of know what the outcome's gonna be. So let's just get there sooner than later.

Hey folks, thanks for talking with us about this particular topic, but we're gonna move on to what John just talked about in a minute. Kubernetes, we'll be back in a second. Cloud native now is the web's leading resource for the growing cloud native ecosystem.

com is your destination for news, thought leadership, features and webinars on cloud native architecture, Kubernetes, serverless, cloud native application development, microservices, service mesh, cloud native security, and more. Stay on the cutting edge of modern application development at Cloud Native now. All right, folks.

And we're back. And we're talking about Kubernetes. 'cause well, it's having a moment.

Last week it celebrated its 10th birthday. Who would've thought, it's an interesting thing though, early on, everybody seemed to think Kubernetes was gonna take over the world, and we had full stack developers forcing IT teams to embrace it and manage it, and there was some resistance on that side. And finally they came around to it and said, well, this is gonna be great.

We're all gonna be platform engineers, and now they're trying to get the rest of the development community they assign on board for this thing. And we're encountering additional resistance because, well, a lot of the developers who are not full stack developers are going, Hey, this thing's hard, and I gotta know more about infrastructure than I ever wanted to know. So we're kind of like, it's 10 years in the making, and I guess we could call it a success, but it sure feels like it's taken a long time and there's much work to be done still.

John, what's your take on Kubernetes in general? Do you think it's becoming a defacto standard? Yeah, I think it is.

But like, I, I kind of think what you're talking about, like that where were you in the summer of 2014? Yeah. Um, yeah.

You know, and yeah, you know, I, I I, the the other point you made, right? Like, just like this, like sort of, uh, two views of this one is, it's one, it definitely won, right? Like, if you look at like, you know, the, the history here of the competition, and like, it's just been eradicated, right?

Uh, and, and there's a lot, and, and one of the reasons it it won is not just that, it's just Kubernetes itself. It's, you know, it's E-K-S-A-K-S-G-K-S. It's, uh, it's the variance like rancher, you know, it's, it, it's so what you all, if you quote unquote say, are you running Kubernetes?

They would all say yes, right? Um, but you know, the swarms, the nomads nomad, I thought nomad was like, like, come on, Noad, you can win this race. You can do it.

You know, and like, even they pun it, um, you know, fargate, like, it's getting like, you know, um, gee, thanks. Um, so, so, yeah, I, I do think it, but the point you made, the other point I think is I, you know, I, I'd forgotten that I, I stopped, I'm, I'm so AI in the brain these days. I'm like, oh, Kubernetes, do I really want to talk about that?

But, but one of the things I, I think what's interesting is if you think about what Docker did, right? Docker basically engaged the developer to, to infrastructure in a way that had never been engaged before. The reason the world fell in love with Docker, because now a developer could literally emulate almost nine, like a high percentage of what the operating environment was gonna be like when they code rent, they could run in their desktop, they could simulate different containers, how they interact with services.

I mean, it wasn't a hundred percent complete, but this idea that you could create this, um, you know, uh, you know, not, not just convergent infrastructure congruent, we called it like the chefs, the puppets infrastructure code were all convergent, like every 15 or 30 minutes, make it look like this. But now you could actually create congruent. Whereas like, for the most part, what you ran in environment A, which could have been in your laptop, ran environment B, which was some test environment and ran in production, was not a hundred percent congruent, but pretty darn close.

And now developers had sort of authority to operate, you know, in, in, in a sense that they could, whether they could be controlled by risk and, and, and, and controlling bodies, but they could run their code pretty much the way they ran in their test environment. And that's what they always wanted. They didn't wanna have to go to the ops people and say, please, can I have a little bit more storage?

No, go back. You know, you know, the Oliver, I go the Oliver story. Um, and then that was great.

Like, that's why Docker went nuts, right? And then we say, okay, well, there is a problem with Docker, like, how do you orchestrate it? And Kubernetes comes along, and Kubernetes is not easy for the developer.

Developer can't do the full, you know, string. Like they, they either have to like, you know, sort of give up like part of their time to become an expert in all the things. They shouldn't have to be an expert network, compute, you know, storage, you know, all the complexities of, you know, then you add the service me.

Anyway, so I, I, you know, that still is the problem with Kubernetes is I think who owns it, right? Um, you know, it's supposed to be seamless from, from both sides, the developer, any operations, and it's not, and it's never been absolutely crystal clear from a lot of the organizations I work with is who now owns that. Like, it was clear that develop could own everything from building, you know, for creating a code, a docker file, creating a docker image, deploying the docker image, but then when you have to run it in these clusters and ponds and all like, and all the networking and confusion and, and the Kubernetes, you know, I, uh, the one last thing I'll say about this is I, you know, I think at some point I started saying the inmates are running the asylum.

You know, the people who are literally building this code at lightning speed are not actually using it. You know? And, and yeah.

So there's, There's two ironies here, right? One is, if you talk to AWS and Google, they'll tell you that there's still more code running in container platforms that don't have Kubernetes than there are ones that do have Kubernetes. And a lot of developers are probably voting with their feet and saying, I don't think I need orchestration, or I'll get by without it.

Um, and then secondarily, to your point about who's running this stuff. So we're hearing everybody talk about platform engineering teams, and it kind of raises this whole other issue, right? So theoretically, they're out going to go build, uh, an internal developer portal, and they're gonna provide self-service capabilities to all these developers so they can go create applications and not worry about Kubernetes.

But that's what work in progress at best. And then it brings up this whole question of, well, what is the future of DevOps in that context? 'cause a, a lot of folks would say, we embrace DevOps to get out from underneath centralized it, and yet, with the rise of Kubernetes and platform engineering, maybe we're headed back to centralized it.

So is there some way to strike a balance here in your mind? Well, you know, the, the, I, I, well, I think there is a, a freight train coming down the pike, and I'll, I'll, I'll table that for a second. But, you know, the other thing about, um, you know, when, when, when we first started playing with Docker, you know, I was like the eighth person and I sold the company to Docker, right?

Like, I was like eight, you know? No, no, I was an eighth. That was chef.

I'm sorry. That was, uh, I was like, I don't know, maybe 60th person. I sold the company to Docker.

I was in into Docker real early, right? And, and I remember like, the idea was we could start creating ratios of 10 containers per vm, right? Or 20 containers per vm.

Like, so there was gonna be what seemed like a real positive scale opportunity. And, and that sort of like model or operating sort of model or theory really demanded orchestration. But a lot of people, I'd run into 'em early on and they'd say, oh no, we run one vm, one container.

Like, what do you outta mind? Don't you understand Docker? You know what, what, you know what's wrong with you?

You know, that's not what, that was not division. You know, you broke the, you, you, you, you broke the mold, right? And, and, uh, the truth is, maybe they were right.

And maybe the people who are responding that say they just run containers, are just stuck with that. And at the end of the day, it's all virtual anyway. So like, and, and if you're running on the cloud, you don't consume that overhead anyway of how many VMs per physical server, right?

That was the other problem. We had this, this containment methodology, right? Like, okay, we went to virtualization to be able to run more, um, sort of like operating systems per physical machine.

Then we went to containers to run more sort of these, you know, container operating systems per vm, right? And, you know, we were supposed, but at the end of the day, if it's already virtualized and a pro and a provider's already assuming the cost of the overhead for virtualizing on physical machines, which they, they do really well. The Google, Microsoft, and yeah.

And, uh, but onto like, where are we gonna go with DevOps? And, you know, it is a little bit of diversion here, and we'll have to keep it quick, but I think we should have a whole topic on this. It's this idea of what they call autonomous dev tools is happening.

And it is just, uh, I mean, I don't, I really don't think pretty soon it's gonna matter to anybody, whether you're running, it's, it's sort of like when you use, when you use, uh, chat GPT, you're using under the covers a whole bunch of technologies. You care, you, you know, you, you're using, you, you, you know, you're using Kubernetes, you're not, you're not, what you're doing is driving somewhere. So these autonomous dev tools are gonna be, Hey, fix my problem, do this, install that, do this, and maybe Kubernetes is gonna get installed as part of an instructor.

And you said, Hey, I wanna run a cluster of these services and here's the code. And they're like, oh, yeah, well, what, what that person needs is a Kubernetes cluster with three pods and right. And you're not, you know, my sort of joke is, oh, are we really gonna need a conference where 5,000 people are gonna be running around some larger city trying to figure out where to put semicolons into YAMA piles?

And I can tell you emphatically the answer is no, based on the stuff I'm doing with some of these autonomous dev tool, open Devon, uh, there's a, uh, the rancher guys created something called, uh, GPT script, um, there, you know, and, uh, like these tools are just, um, the stuff you can do with them by just having a natural language conversation. Um, in fact, the one last one is that I just got early access to co-pilot workspace from, uh, from Microsoft. And it's literally a development.

You, you point at an incident, it analyze the incident, it gives you a task list for correcting the, uh, the issue, sorry, not it's an issue. Uh, can have issue. It gives you a task list.

You can mutate it, you can then execute it, and it'll fix the problem. You know, and it's funny how related all these things are, right? 'cause everybody's talking about this brocade bought VMware thing, and they change the licensing terms, and suddenly there's a lot more interest in taking that VM and just encapsulating it into a container and running it on Kubernetes, and that gets you out from underneath VMware.

So it's funny to me though, none of these technologies seem to exist in isolation, and they all have relationships with each other. And, but I'm not quite clear that we ever think about it holistically enough. I, I don't think, uh, I think at the end of the day, like I said earlier, I think it's like we're not really gonna care except for the fact that we need to be diligent about observing the potential opportunities of breaches and, um, you know, adversaries taking exam, you know, of like this sort of, the fact that what I don't think is going away.

Like, I mean, I think there's a world very soon where most software engineers are literally just using NLP, you know, basically chatt PT or some variant of a generative AI to do their work. Mm-Hmm. I mean, I, like, I I think we're already beyond copilot.

Like copilot is like, Hey, you know, write me code that does this, this, and this, you know, with things like copilot workspace or open Devon or these tools, these are more like, Hey, I want an application that is a web service. If you find it desirable, use Gin X. You know, like, you know, that kind of stuff, right?

Like, we're, I, I think we're gonna be an abstraction removed away from like how we build software, you know, or how software engineers develop structure. Um, so I, I like that's happening that, you know, whether you like it or not, it is happening as we speak. If you start tracking, uh, there's something called this SWE software Engineer Leaderboard or Workbench, and you watch every week, there's somebody on it who's given a higher percentage of what these autonomous dev tools can do.

And you know, it, it, when, you know, when open Devon, when devvin was, Devon is sort of a joke, sorry, I'll say it, but open Devon is not a joke. Um, and like it started out like, what, 9% on, on this leaderboard? Like, they're up like the high, the low twenties now percentage of what they can accomplish based on this, uh, um, workbench, um, uh, the, what's called the SW dash workbench.

Anyway, so that's happening. But the thing I worry about is being somebody who's always been on the sort of protector side is, you know, how are we gonna keep up with the infrastructure that's running all these things for us? Which is a good thing.

You know, I don't wanna have a thousand people that know how to bill ya files and a hundred thousand per person bank. Like, like, I don't, like, that's not needed today, right? Um, like if your full-time job is creating configuration or YAML files, or I'm sorry to my good friends who I still love it at Chef, you know, like, it, like, no, no, no, there's no need for that.

Now, the Terraform, you know, like, that stuff will happen automatically. Uh, you know, here's my descriptor. Make it so, and we're there.

But the thing I worry about a lot is all the things that, like, who, who's gonna watch what's, 'cause they're all those technologies that will do that for you are all running common cloud native stacks. All right, folks, well, think about it this way. Maybe by Kubernetes next birthday, it's 11th, we'll be talking about how we applied AI to the management of Kubernetes and just rendered the whole thing.

We All right? We'll be back in a minute, talk about our next topic, but stay tuned, Tuned. Discover Textron Group, the epicenter of tech innovation.

We are your go-to for reaching IT leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us. You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more.

Join our satisfied clients. Let's revolutionize your tech journey. Contact us today and tell your story to the world in the most powerful way with Techron Group.

All right, folks, and we're back. And we're talking about Cisco, and they had their big Cisco live conference last week. And if you haven't noticed, it is conference season.

There's about three of these every week. But Cisco is interesting because they have AppDynamics and you're talking about first applying gen AI to Dynamics, but then they're gonna extend that out to integrations with Splunk that they recently acquired. And they have thousand eyes for network monitoring.

And of course they have all these other tools that they use for the management of networking and security, not to mention all their server infrastructure. John, do you think that we're on the cusp of maybe bringing all these different frameworks and tools together in some sort of cohesive way, that it makes it easier to manage things end to end? Cisco is making a bet that that's exactly what Will happen, or Are our silos so entrenched that we'll never get there?

Yeah, I, I, I think that, I think we could be on the path to having these, like, these incredible integrations, but I think a lot of the vendors I'm talking to don't get how to operate in this new domain. And, and, uh, you know, what I mean by that is I think this idea, like I've had a lot of startups come to me and say, Hey, you know, we're thinking about, you know, integrate Open High. And the first thing I say is, don't build your own LLM or co-pilot.

That's a bad idea. Because what, what the industry will not tolerate is 'cause like, think of this world where now a large bank's gonna have Cisco's, LM Salesforce, LM, um, or the Einstein. It's gonna like name every product that you got.

And that's not even including things like Workbench and, and other HR tools, right? Like, and that is gonna be technical debt. That is going to be, you know, sort of orders of magnitude complex than anything you've seen before.

I call the, the, the upcoming technical debt tsunami, right? And so the, you know, like there's gonna be a blockade on some point. Like, all right, no more co-pilots, please.

Right? So if you're a vendor right now, and then like, like Cisco would Nash for my opinion, and like, you know, um, then, and, but like the small guys come to me all the time, and I'm like, the new world is to basically, so what happens now is expect that all of the copilot and, and sort of, you know, LLM magic is going to happen outside of your, so what you need to do is expose all your stuff as APIs. And so the new environment is, I sit in an environment where I have access to, um, to your information via command line preferable, preferably APIs or an API gateway or so not, like, not in the sort of classic gateway, but like a, a new router model where if I'm Cisco, what I'm doing is I'm not trying to provide my customers, you know, uh, uh, gooey by the way, I don't think people are even gonna use your gooey anymore.

The new Gooey is not going to, um, go into your screen to see what histogram you show the new gooey is having a, a, you know, a question and answer with your product, Hey, what, which, you know, which, um, you know, switches are behaving abnormally, for example, greater than this, um, many milliseconds later. You know, I'm making s**t up. But, but the point being that, like the new domain is assume that everybody's gonna use you from a common architecture and don't, like, I mean, I guess you're still today, you know, um, whatever today's date is, June 7th, 20, you know, 2024, you still probably have to have a gooey, because as people are still like, Hey, that AI stuff, but like, I would put all my oxygen into thinking about the new architectures of like these, you know, sort of routers that work with APIs and data consumers that then feed into LLMs that now I can basically integrate at my conversations with whatever I need from you and everybody else.

You know, I'm working with a prototype right now. I mentioned, uh, these, uh, GPT script guys, right? They, like, I can set up an environment right now where I have access to my machine that has all my Amazon commands, it has all my Kubernetes commands, it has GitHub, and I sit there now and, and they integrate the, uh, the LLM, like A-G-G-P-T four oh.

And what happens is the things I run in my environment actually get put into the context window via functions is, you know, a little technology here. But the magic is I can just ask questions like te uh, tell me all my EC2 instances that are running with, um, with, uh, tag, uh, uh, dev local one, okay? Of all those instances, which ones are running Kubernetes and have pods that have this, um, like I'm doing this right now, right?

And, and, and can you tell me any issues related to these, you know, for Kubernetes pods? Oh, can you go ahead and explain that? And then when you start like adding in things like, you know, copilot, wis, uh, uh, workspace, now you're gonna say, oh, there's an issue.

Can you go fix it? That's the world. I, if I'm, if I'm Cisco or many of these vendors, I'm assuming very soon, the architecture is not to have an encapsulated, I'm not ship my own LM with my own vector database and all my orchestration, think about all the technical debt for them when the customer's gonna have to build their own orchestration engines like Lang Chain or LAMA Index or, right.

And they, they're gonna have to have their own vector database sources. They're gonna have their own architecture. So the smart play is to integrate, you know, so pushing your delivery, you know, because Dynatrace and Splunk and all those tools and thousand eyes, uh, you know, maybe less thousand eyes, but, but, uh, are all gonna be around, you know, that technology is not going away.

We need that information about our infrastructure or the infrastructures that we're running. But I just don't see a world where, um, where not only do I have four gooey, like think about Cisco right now, I gotta have a gooey for, uh, you know, for, um, for AppDynamics or, um, and I gotta have, uh, a go for, uh, you know, sort of Splunk, I have to GUI thousand eyes and now I'm gonna have additional GUIs that are these co-pilots or some integrated version of those. Plus I'll probably have a, a documentation, um, rag built.

So now for one company for like three products, I got 10 different gooeys. When right now the assumption is I don't talk to Gooeys, I talk to LLMs. And so, yeah.

So I, I think, um, So, so play that out a little bit for me. If you would, if we're gonna talk to the LLM, do we need the gooey in the first place? 'cause it's just gonna be maybe a verbal in interface or speech.

And then secondarily, am I gonna have, you know, like a, a master AI agent that will talk to all the, uh, other AI agents on our, my behalf to go execute something asynchronously? Or do I have to talk to each one of these individually? I think, you know, these, this will all evolve, but, um, I, I think there, the, the, the, you will not really need, uh, you know, a hundred GUIs, right?

Um, there are some models where, um, when I say models, I mean some sort of implementations where you might want to have like systema record data available. So then now you consider your natural language question, maybe it's too much, or the answers have to be absolutely correct. So let me use the LLM data to get me to sort of a system or record answer of what to do, right?

And so in that case, we still will have GUIs, but I think this idea of of going to some navigation of screens to find out what the answer is of the performance of this service. You know, like, okay, go here, I log into here, am I logged on? What about credentials?

This, this, this, this, this, this, this. And then I go all the way down the stream and I say, oh, yeah, look, it looks like this is, but yeah, you know, I need to go back to that other screen now, right? The, like, the, the future is more like, and, and it's still nascent, but it's, this is going to happen.

It's gonna be, can you tell me any of the performance? It's like the thing I told you that I'm working on with just Amazon, like gimme all my instances of running, gimme all my instances that have tagged, uh, you know, that have a tag with dev local one, right? Um, like I don't need to scroll do, and you know, like it just answers the question.

And I can say, you know, of those, which ones are running Kubernetes, people are gonna expect the dialogue like that. Um, now as far as the architecture, I think the simple architectures that, you know, some of us are sort of designing and working on, um, there's a group of like sort of, um, you know, DevOps pioneers who we've sort of been playing around with different ideas, you know, helping some companies build this, but you'll have like, you know, all your sort of API, um, you know, like you might have an S-Q-L-A-P-I, you might have a database, you know, you might have, uh, you know, a performance tool and an orchestration tool, API, you might have, you know, just go down the list of all different APIs you might have, and then you have some router technology, which is sort of part LLM, or it's, you know, like some of the, uh, orchestration tools like, uh, like Lang Chain and LAMA Index have these capabilities to do this type sort of like, um, semantic or intelligent, like, oh, I know they, I don't even have to set up a config like it, it says, oh, I know they want to talk to the SQ l agent. So, and then that will all load into, you know, initially architectures like one LLM, but the truth of Army of LLMs, that purposes.

And then there's a front end on top of that, you know, there's a rag, there's a, you know, database, there's, there's a graph component of the data, you know, so the, at, at some point, the architecture of these things will sort of be, um, and you know, this idea of sort of, um, some people call it army of bots or the army of LLMs or mixture of experts. And then, and then on top of that, you'll have all this data, and then you'll have sort agents, like agents who will now sort of have specific things to do. And this is all happening right now.

Like, I mean, people are building these incredibly sophist, Patrick Devar is doing some incredible stuff with like, actually platforms and agents. And so like the, all that stuff I'm talking about, like, like this gets real scary. You'll have just agents that run once you have that architect set up, they're just gonna basically look for those problems all day long.

Mm-Hmm. Well, and that's my next question to you. So right now it requires a fair amount of intelligence on our side to go interrogate the systems that figure out what's going on.

Are we gonna get to the point where, you know, we'll just come in, in the morning and the, the agents will tell us, here are the three things that are going wrong, and what, what do you want me to do about it? Yeah, I mean, like, get the evolution of being a CI admin, right? If you go back before we had a lot of these tools that aggregated all how that stuff, that's all we did.

You know, you came in, you know, in the morning you ran this command, that command, that command, that command, oh, this command told you to run these three commands. You're on that command, that command this command, you found something there, right? And in a lot of ways, you know, these observability tools and stuff like that have built an abstraction, or now we sort of come in and we, we look at, you know, the red, you know, whatever's pops up red, right?

Or, you know, for, if it's 24 by seven, it's a right, it turns into a PagerDuty alert, right? Um, this is just another evolution of that. Yeah, I think that, I think we're gonna find that there are narrow scenarios where agents can do a lot of that.

Now, there's always gonna be the human in the middle, right? Like there's a sort of, like, I, I'm a big believer maybe just optimistically that, you know, the human is going to have to like mitigate because these, I think they, as much as you can control 'em from hallucinations and, you know, sort of bias and like, they still will always have, and they will all drift like a little bit better, a little bit worse, a little bit better, a little bit worse. We're actually seeing this in the industry with some of the Google stuff right now.

Like, so I think that there's always going to have to be a human, it's just what the human has to do to make that arbitration. It's just a lot of that's gonna go away. Mm-Hmm.

At the end of the day then, are we gonna be able, people are saying for instance, that we're gonna have more applications built and deploy in the next two years than we had in the previous two decades. Um, You know, Are we on the cusp for that? And, you know, can we manage at that scale?

Well, that's another question. And I, I worry about the technical debt issue of all the hidden debt that comes from having all this. Like, it's always you is the, the sort of idea of this, there's oxygen in the balloon, you squeeze the oxygen, it goes somewhere else, right?

Uh, the, it still stays in the balloon. Um, I, I, I do think like the, this idea of like this, I, you know, Jevons paradox is something, a topic point we've had in DevOps forever, right? Which is this idea of, of like, the more you easier you make, more you make abstractions, the more you get, right?

And I always use the, you know, everybody thought the, uh, video stores were gonna put the theaters at 'em, you know, 10 years later we went from like two theaters to 36 plexes, right? Like, so that, that's sort of an example of sort of Kevin's product. I think all these abstractions are like gonna create tighter feedback loops, which are actually gonna produce more results.

We're gonna get more sort of feedback from our customers. I mean, I think, and the fact that we can code faster, we can build and implement, we can fix faster. Um, I think we're gonna, I think we're gonna see, you know, business commerce like, um, you know, and again, like if you're worried about losing your job, you know, change your job, but like, because, um, you know, if you're a Java coder for a bank and you've been basically copying and pasting code for 25 years, you're not gonna have a job.

But if you've been in a bank and you've been innovating through your code on how to create work with the business and innovate new ideas and, you know, help the business and the PMs understand how to turn technology into business opportunities, then you're gonna do better than you ever have. Drew that. Hey, John, as always enjoyed the chat immensely.

We give 'em everybody a glimpse of the future. Of course. There the question is our Future.

Yeah. I don't know. Yeah.

Like, yeah, like I was say, if I'm in a line for in the shopping, you know, in the grocery store, do not get in that line. It will by default be the longest taking line. So just buyer beware.

All right, well, they say cowboy wisdom, never confuse a, uh, good view for a short distance. But we'll see what happens. And hopefully we'll all be here to talk about it when it happens.

Hey everybody, thanks for watching the latest edition of the Textron Gang. Stay tuned for some of the awesome episodes we have on the rest of the show. And till then, we'll see you next episode of the Textron Gang, which should be tomorrow.