Techstrong TV July 11, 2025

Hey, is that a phone in your pocket or is your AI glad to see me? You're watching Textron Gang. Hey everyone, it's Alan Shimel.

Happy Friday. It's Friday. Can you believe it's Friday this week?

I blinked. It was Monday. Um, we've got a lot to talk about, including that little phone in your, or maybe it's a big phone in your pocket.

Maybe it's a folding phone in your pocket, but there's a phone in your pocket and it could be your gateway to AI going forward. We're gonna talk about that. We've got more, we've got a great gang to talk about it with.

Um, some of the, our best characters on the gang. Joining us is Kimberly Bates. Kimberly, it's great to see you.

Good to see you too. How are the Yankees doing? Um, well, I, they won, I think they won their last two games against Seattle, but it's almost Allstar break.

We'll, we'll take, we could, we're limping into the break. Um, IRA Weaker is here, of course, our security person extraordinaire, who's never short of an opinion. Nah, glad to be here.

Yeah, always happy to give that opinion. And, uh, also joining us out in Silicon Valley. I'm in, in the, in the Valley, John Swartz, as well as the Dean Mike Ard.

Guys, lady, welcome. Let's jump into it. Mike, Samsung announced the next version of their foldable foldable, I guess, is it still a Galaxy or is it something else now?

Yeah. Was it a razor called These days? Um, that's a full seven Fold.

Seven of course rolls right off the tongue. But more importantly about than the folding is the ai. What, what, what, what do we got here?

Yeah, so Samsung was in Brooklyn this week where all the cool kids are hanging out in New York and had a big event. And, uh, you know, I appreciate what everybody's doing there, but it was basically a two hour plus commercial. And so that was kind of like, you know, a lot like an Apple event.

And I'm hoping somebody might actually change that format one of these days soon, because I swear there was more video showing than people actually talking. But one of the things that was interesting was they were positioning this new phone and all the other phones going forward as your AI companion. And for example, they were talking about how you might be walking down on a subway somewhere and there's an ad for a concert, and you'll take a picture of set concert and then it will automatically drop that into your calendar.

And then not their agent, but some other agent might therefore then go out and buy those tickets for you. And that whole process just becomes something that starts with a click of a photo on your phone. I think, I think we're a ways away from making that kinda everybody's everyday experience.

'cause you gotta upgrade your phones in the first place. But Alan, it does seem to mean that our relationship with our phones is about to change. Yeah, no, I don't think we're as a ways away as you think we are.

I, I think it's really right around the corner. Um, couple things here. First of all, I, I don't want to discount the idea of this foldable phone, right?

Because I, I think it's attractive. Let me just say, uh, from the outset, I'm a, I'm an iPhone user and Apple guy, so I'm not going to use this phone, but I like the idea of the foldable phone. And if you look at the screen specs, it really is a beautiful screen.

I mean, it's a nice screen with great technology behind it, and maybe we're gonna need a bigger screen to, to communicate or interact with our AI agents and alter egos who are out there doing things for us. Beyond the foldable aspect of it though, look, I think Samsung, again, is leading with innovation in, in a hot new area, right? Because quite frankly, what's the difference between iPhone sixteen, fifteen, fourteen, thirteen, twelve?

It's incremental. There's not really nothing. And, and, you know, a lot of innovation does seem to happen on the Android side of the, of the house and with Samsung.

And I think their use of AI here, again, positions them as more of a leading edge. You look at the ho hum apple intelligence about, you know, it was all about what, what's on the comm? Samsung's showing you some real functionality here.

You know, it may not all be available, but it's real functionality. The bigger issue though, is look, eventually the iPhone catches up and they, you know, they're within a release of each other and they have very similar, uh, features. What this means is that all of our phones are gonna be our AI gateways, right?

Think of them as our digital agents, our digital workers as Salesforce calls them, right? Our digital workers out there doing these kinds of tasks for us. And not just tasks, but really kind of thinking for us as well.

And I mean more, you know, just taking it to the next level. Now, I think the issue though is where, you know, where we could see a potential roadblock is how much computing power, how much AI computing power can we put on these phones? If you look at the specs on this fold seven, right?

It, it's on, i I believe it's on a Qualcomm Snapdragon eight or something Chip, isn't it? Or it's chip set, but there's GPU, there's CPUs, there's, what was it, 12? Was it 12 gigabits of Yeah.

You know, was an NPU in there too. And an NPU. I mean, this is all great.

Is that gonna be enough to run the AI in our pocket, or is it still gonna have to No pun phone home to, to get AI done, Kimberly? So I'm not sure. Yeah, I was gonna say your, I'm not sure saw the announcement this week.

I think it might've been even yesterday. Open AI is buying Joni Ibes hardware startup. Joni is the designer of the i the architect.

Nobody knows what they're actually doing. Um, it's well kept under wraps. There's a lot of speculation about whether or not this is eyeglasses or something along those lines.

But, and then, then Wall Street's saying, okay, so this could drive o open AI to a trillion dollar valuation, you know, blah, blah, blah. I think the, the statement here, a you're, you're right. The, the, the consideration is this gonna talk to the cloud?

It has to, in my opinion. We'll see if they can do it differently. But the other piece of that is where is this going?

You know, strategically, how is, how are these folks envisioning us using the devices? I mean, when Tesla says, or, or, or, um, Musk comes out and says, grok is now going to be in Tesla here within the next week. Okay, that changes.

So I think all of this, these pieces of that we look as devices and look at how we interact with things. Um, this AI is not just the phone, it's the car. It's a lot of other things that we're gonna be having, seeing where this AI plays out in.

Yeah. Could I just say, oh yeah, I was just gonna say this all frankly, I'm sitting here, like, everybody's saying, this is revolutionary. This isn't revolutionary.

This is just evolutionary. When you look at it, I mean, a foldable phone, oh my God, does anybody remember the Motorola razors? I mean, I'm like, you know, that's innovation.

We're going back, you know, it's Like, yeah, no, no, but to be fair, But I'm, I'm like, That was a flip phone Ira. These are not flip phones. This is A, I know these are, This is One big screen.

These are like screen technology. Yeah. And I appreciate that.

I hated moving to like a God, I'm a Blackberry and then onto an Apple because they didn't fold and put in my pocket. Well, but with the AI though, it's really just an evolution because you stop and think, like, you know, all of a sudden everything's just evolved as the technology is. And, and when you start to think how seamless can things be, it will be, in my opinion, and it has become more and more, like, for example, QR codes on, you know, on advertisements, allow my phone to pick something up, pull up directions, do whatever else.

Now we're talking about taking a picture, interpreting that, have an agentic AI kind of go out there. You know, I'm more concerned, frankly, a little bit about the security and privacy. Again, because you look at this and yes, it, no matter what, whether or not it goes out to the cloud to process, because when you put something in your phone, it has to do a query.

Now with regard to how fast the chip is, the chip doesn't have all the data set. The chip will just have to process and do basic things, but then still have to go out and pull things. And frankly, if I have your cell phone today, I can pretty much put your life together.

I don't know when, how many people remember, for example, Google Friends, where it's like, okay, broadcast your location to all your friends. I'm like, what part of this seems like a good idea from a privacy perspective. Now we're talking about, like you're saying, here's, you know, let me take a picture and, you know, everything seems wonderful, honestly, that I'm gonna have an AI assistant, like for example, I want to know if it's gonna rain, and I just say, sir Siri, is it gonna rain?

And all of a sudden I get like a weather report saying, okay, where am I? What's the weather for today? Now if I go ahead and say, Siri, I'm hungry, put out my basic, I don't know, Chick-fil-A order, it has to go ahead, process this, understand it, and then interact with Chick-fil-A and make an order for me.

And it knows what are my preferences, what everything is out there about. When you put a query out, it's telling you what you think. There's a difference between a tracking mechanism and people knowing what your Google searches are.

People knowing where your maps are going, people knowing what you're ordering from stores all brought in together. And I'll just leave it there because I think it speaks For itself. You, I wanna, that there's two separate parts of this.

One is the horsepower needed to do these, whether they're revolutionary or evolutionary to, to perform this AI par tricks, you know, and, and they're not just pars to perform this ai, there's a certain amount of horsepower needed, and phoning it out to the cloud or wherever is going to introduce latency as well as security issues, right? It's probably more secure if I kept it on my phone versus sending it out for now. And when you think about it, these little pocket rockets that are in our pockets, right?

There's more computing power on the one you have in your pocket today than there was on the Apollo 11. I think it was 11 that landed on the moon, right? We have more computing power in our pocket than Neil Armstrong had landing on the moon.

But that's not enough for this next generation. And, you know, technology, we used to have Moore's law. You always said, ah, don't worry, you know, we're gonna double computing capacity in 18 months.

Well, we haven't been hitting those numbers in Moore's Law, but certainly here on mobile devices, we're seeing a, a revolutionary change, or not a revolutionary change, but an exponential increase in computing power on board. I read to your point, though, yeah, I read an article just this morning on this, that the fact of the matter is the large hyperscalers, the large collectors of information, the Googles, the Apples, the Metas X, Twitter, whatever you want to call it, they continually are adding to their treasure trove of information they have about each one of us, our habits, our likes, our dislikes, our travels and everything else. And until, I don't know if it's a countrywide thing, a personal thing, but until people say, I'm fed up, God damnit, and I don't, I won't take it anymore.

It's not gonna change. You Know, there, there's interesting premise in Mike's story, which is in our, in ai, the tech strong ai, and it's these, these two major providers, smartphones and tablets are basically locked in this AI arms race. And that's gonna affect millions and millions of people, if not more.

And it's interesting to me, we talk about evolution. I think we sh also mentioned transitional, because Apple, in a sense, I think, and what I've been hearing through the company and people who've left the company, is that it, they're in a sense, kind of in a transition from the iPhone era to the AI device era. I'm not saying the iPhone's gonna go away, but there might be a new device that, But, but that's the whole to Kimberly's point, right?

John, that's the whole thing behind the Johnny I smart. Uh, That's where I'm going to next because that deal just closed. And the, from all indications are they are, OpenAI is gonna have some sort of device within a year, maybe six months, which also will have an influence on Apple because Apple is looking at what, at a ways to build out their AI presence.

And they're looking at companies like perplexity, and in a sense, even teaming with a company or buying it outright to develop an a AI search engine that would twin with Siri. There's like a series of things that are gonna happen and, and Meta's gonna do the same thing. So we've got all these, this kind of transition as, as we go back to 2007, right?

Steve Jobs, we have the iPhone, which is now our computer in our pocket. And now they're, we're probably in this era where we moved to an AI device, which is the next iteration of the, of a phone. Well, I just wanna, sorry, if I could just reiterate a critical point.

When people talk about ai, and I hate the term 'cause it's too broad, it's too nebulous, but AI requires two things, and this is where I have to keep harping on it. AI requires processing capability. It requires, well, actually three things.

Algorithms processing capability and data algorithms have been around, we could tweak the algorithms or whatever, but they've been around for decades. Processing power allows the algorithms to function more efficiently. However, the key thing is still the data, the data you need is not gonna be on your phone.

The data you need is gonna be out in the ether, the internet, whatever networks develop. And so while it's critical that we are talking, how much processing can be done on the cell phone itself, which means maybe it could be more efficient, you're still gonna have all the concerns about going out and seeking data. Because until you're at the point where you can put all the data of the internet and all the functionality on the phone as well, which will never happen, you're still gonna have to go out to the rest of the world.

And we're talking, I mean, frankly, this is an engineering issue that was released that we have an AI phone. No, you have chips that can process AI algorithms more effectively, but you are still gonna have to go out and get the data. And whether the queries are done on the phone itself or the queries are done, it still has to do the queries itself.

It has to still search, bring all the data back it needs, and then do more things. Anyway, I'll leave it there for that. I'll tell you what I am concerned about that goes beyond the IT and the security and all those other things is, you know, are we gonna live in a world soon where my AI agents are gonna call your AI agents and I may never actually interact with you or have a conversation with you and it's just gonna be kinda AI agents do this back.

We talking about, it's, it's Wally, the movie Wally, you'll be You. Actually, that is what's so important, I think, of what's happening with the open AI acquisition is that, and IRI agree with you, there's three elements there. The fourth element is our interface.

And that what potentially is happening here is, um, we're having a rethinking of what those interfaces are going to be Like. Yeah, yeah. The human, human computer inter interface.

So, so what is that major interface? And that's going to be a race to what that is, and most likely is going to be some level of a, you know, small device that we can hold in our hands because we wanna carry it around and look at it. But I think that's what, where these guys are going is gonna, well, There, there was that one company that had the pin, if you remember that didn't go over so good.

No. Anyway, hey, we're over time. We gotta stop on this.

It's, look, we're not stopping, we're not, we're gonna be talking about this. I have a feeling for weeks, months, years. But let's take a break here on tech strung gang.

We're gonna come back and, you know, AI can't make up legal citations and courts can't rely on them or can they? You're watching Textron Gang, 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 Textron Group. Group.

Hey folks, we're back. And it's not just AI and law, it's AI in just about anything we've seen this week where the, uh, rock AI was spooning out all kinds of Mecca Hitler nonsense and lawyers are citing AI case or cases that never existed in the first place. And it feels like maybe we're getting a little too dependent on AI already.

But John, you've been covering this whole space, it seems like now every other day there's a new it. Yeah. And so it's, I mean, in a sense it's like AI gone wild, right?

It's, it's, it's, uh, it's unpredictable. Um, and it's, it's embarrassing companies, as you mentioned, there was the incident at X, which the timing couldn't have been worse. The CEO just quit.

The CEO quit basically the day after an update to Rock was spreading anti-Semitic and pro Hitler messages. Uh, her timing couldn't have been worse. Uh, so she left, uh, Musk threw his hands up and said, you know, this is the world we live in.

That's just a nev never a dull moment. So, ironically, Musk, the guy who warned us about what could happen with AI was a victim of it. Um, we mentioned it, the Georgia Court of Appeals decision toss out a, a ruling that relied on fake cases generated by ai.

Uh, that's not the first time this has happened, but this is the first incident that I think has been where a court case, uh, ruling has been reversed. There have been incidents over the last couple of years where AI has been used to site cases that didn't exist or to mislead, um, a judge. And the judges have been rebuking some of the attorneys and finding them.

There was a case in New York, there are multiple cases. And then there was a federal judge in Colorado who rebuked, uh, the attorneys representing the MyPillow guy in a defamation suit when they discovered that, that, that his attorneys had used a generative AI program to submit a court filing filled with errors. And I think, I wanna kind of kick it to Alan because, uh, I talked to a friend of mine who is a lawyer and, uh, she practice his family law.

And one of the things she mentioned was that law school 1 0 1 teaches us, is it Shepherd eyes? Shepherd she shepherd. Well, shepherds is the book you shepherd eyes it as the verb of a shepherd verb, okay?

But, you know, that went over to computers when I was in law school in 19 82, 81, we were already using the Lexus Nexus to shepherd die case law. Um, but look, you know, there's this, there's this term in law fruit of the poisonous tree, right? All fruit of the poisonous tree are bad.

And so if you have cases that are decided on poisonous, you know, on poison, everything coming out of it is, is no. So of course the court had to overturn it. I think the case in Colorado, John, they actually sanctioned the attorney and find him $2,500 or something for, for spouting off nonsense.

But a couple things. Elon Musk isn't the victim here. Elon Musk is the cause.

Elon Musk decided that there were too much, there was too much woke and liberal input into his rock gr AI and removed all of that. So it went too far the other way and started spouting antisemitic Hitler stuff. And he throws his hands up in the air and says, oh, another day in the internet.

No, when you put your finger on the scale, bad s**t hap bad stuff happens. I'm sorry. So let's not feel sorry for Elon.

He's the cause. Oh, no, No, I, I'm not at all. I'm just saying, I find it ironic, this guy who's been pointing his finger and warning us about what could go wrong is in a sense, per perpetuating this problem.

Well, he, he brought it on himself. But that being said, here's the bigger issue, guys. You know, our friend Chris Blaque is railing on here every week about his partner, Lumina and Civic ai and putting ethical kind of guardrails around ouris.

And I do believe it's going to happen, not because of government is mandating it or anything like that. I think just good functionality is we, we are our patients for AI hallucinations, for making up case law for, you know, spouting unacceptable type of, uh, stuff. What we saw here with Gar, you know, the public and, and the market is gonna force people to put these ethical sort of guardrails around AI to, to hopefully get better about these things.

Yeah. Alan, can I disagree with you? Yeah, You, I would've I be insult, Insult?

Well, let, lemme disagree with you because I think there are two parts to this. In one case, and let's remember how, I don't know how many people remember when Microsoft started their own little AI bot or whatever, it became a flaming Nazi within like a day and a half or something like that. Now that had training data, because there's a big difference between the two things we're talking, the, the hallucinations and this issue because with, you know, the, the Microsoft thing, which we have to consider, they were training it from Twitter or wherever else where they were just listening constantly to Nazis, for lack of a better way of phrasing it.

Now, the thing that concerns me and what we should really be taking away with it is we now have an AI where somebody put their finger on the scale. Because what happened was, the reason it was tailored was if you go on rock, rock was actually surprisingly neutral. And the problem is Elon's fanboy did not like the fact that it was sping facts.

Yeah. And they're like, gee, what do you mean that, you know, does, you know, like vaccines don't cause autism, you know, and it's like, how could you say that? And then all of a sudden they're scratching their head, gee, I have to keep these pe these lunatics happy.

And they went ahead and whatever they did, there was an intentional tweak because rock was good. I didn't like some of the things, but it was factual. When I check rock's facts.

Now it went the other way. And that says there was a manual intervention. And that, frankly, that happened because much to disagree with you, Alan, it wasn't like the public did not support, you know, or was not happy with it.

No, the, the public was, this was Elon unhappy, This was Elon. No, But Elon's fanboy on there were not happy with the fact it was giving what they consider too liberal results. And they, so they went ahead and they tweaked it to give less liberal results instead of just pulling facts up.

So I I, I have a theory on that though. Ira, look, AI hallucinations, if you want to call what happened with gr hallucination or AI spouting misinformation, false information, there's two reasons these kinds of things happen. One is, if the AI is trained in a cesspool, which the internet is in many places today, if you train it in a cesspool, don't be surprised if it's spout cesspool type of information.

The second thing is if you intentionally poison it, right? And maybe that's what happened here in rock, it was intentionally poisoned. So it's where, see where you get your training data is part of the problem.

And that is like, hey, if I want to train it on truth social, which somebody might want to, you will get a set of data. If you want to train it on Google data, hopefully it'll be a little bit better. But the reality though is that, again, I still wanna make the point that the hallucinations in the legal case, that is just people not understanding AI answering, you know, like these LLMs trying to answer a question and trying to give a good answer without having a factual answer, uh, compared to training data, which has pulled it out.

So I think there's, I think there's a big difference here between these two cases that we're talking about. One of them has a responsibility of the expert to do her research, to make sure the data that she is presenting to the courts is accurate and to rely on an AI to spout cases when you go to, when you, if you sit at bar, you're sitting at the bar, you know, to, to just prove that you have the knowledge base to do this, and you're not getting an AI to assist you in your bar exam. And I would expect that, I mean, if I was the, that, if that was my lawyer that ended up doing that and had that, that happen, probably should fire them.

And I think that there needs to be, from you looking at us as people that are using AI and how we use AI and individually, how we are trusting that information, we can't, when we pull that information, we have to trust but verify. Yeah. You know, it's interesting.

I'm, I'm glad you brought this back up, Kimberly, because there's, there's somebody who keeps a database of this issue and there have been, I think 156 cases in which lawyers cited fake cases generated by ai. But then I, then I think about, you know, where this is going and at the same time, yeah, right. It's not surprised that the number is escalating me despite the f Well, no, but never underestimate how lazy lawyers are.

Let me just finish my thought though. Oh, that's what I was saying, Alan. Look, I've been there, I've done that.

But here's the thing. I mean, we we're, we're talking about all these things, but there's a number that jumped out at me yesterday is $4 trillion if Nvidia just went over that in the market cap and Microsoft's next. So we're gonna see an escalation of AI gone wrong.

AI doing My own, Meg. Well, I know, but I'll give you, so look, whenever you think of lawyers, I think of doctors maybe because I did a little of that kind of stuff. But, you know, we reported earlier this week, um, Microsoft released a new tool for diagnosing, for diagnostic medical diagnosing for x more accurate than human doctors.

That's great. But what's gonna happen when the, when the it misdiagnosis and someone dies as a result, it's a little different than the pillow guy citing bad cases and not, you know, I would, I would point out two things. One is it's frigging awesome that people found 167 instances where AI went wrong.

I mean, maybe there's thousands more, but maybe there's hope that we can actually find these things. And then b my biggest issue with all this AI stuff is all these things are designed to be behan and they keep coming up with answers instead of just saying, I don't know. It's okay to say, I dunno.

Well, fair, there's, you gain more credibility when you say that when you answer, I don't know. To a question. Well, well for a legal brief you can say, I don't know.

And I think one of the problems is, and let's face it with Mike Lindell, first off, it's a very public case, and you also have a client who's broke. And if I was his attorneys, I would spend as little effort as possible, hopefully not violating the law. But to your, you know, to Kimberly's point, absolutely correct.

You expect the lawyers, I mean, frankly, at some point, I think somebody's gonna release their own legal LLM that goes through documents that fact checks their, all their LLMs. And they are Doing that now. They are starting to do that Now.

You know, my, my son worked on that in his law reports Coming through. My son just graduated law school up in Boston, and at his school, they have a legal technology lab, and they're actually working on that very topic, um, as well as others. Yeah.

But it still comes down to the point where, yes, any lawyer should have fact checked that the fact a judge ruled on it, he was, or she in this case, was relying upon the attorneys to submit factual documents instead of, now we need judges, well submit. But the judges look, the judges To verify what they're saying. The judges don't read these things.

The judges look at the facts, they tell their clerk, this is how I want you the decision to go. And it's the clerk's job to go do that kinda work. And, you know, these are kids who just graduated law school for the most part.

Anyway, Hey, we gotta take a break. This's an interest. Another interesting topic that I'm sure we'll be revisiting on the gang.

Let's turn to security though. Let's talk about software supply chain. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry.

com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more. com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com to learn more.

com. Home of security bloggers network. Hey folks, we're back and we're talking about software supply chain security, which has been a top of mind issue for a lot of folks these days.

com where the question was put to C-level executives about whether or not they had any visibility into their software supply chain, and half of them said they didn't. I, I looked at this whole thing and I kind of smiled and I went, wow, C-level executives actually know they have a software supply chain. So, um, you know, is this becoming a board level issue?

Or, you know, is this something, um, that's a moment in time because there's just too more noise in the system or what's going on here? So With regard to whether or not it's a board level issue, the answer is, uh, for over or board does oversight for oversight purposes, the board should at least know to ask the question, what are we doing with supply chain related issues? Because supply chain attacks have become a major problem.

Companies have been ruined because, for example, there was a bit wallet application where they uploaded malicious software into their core code base. And I think what's happened is over time, and a lot of people, if you're not in the programming space, don't really realize, but PE code has not like code used to be written originally. Maybe there were a few library programs around to implement some basic functions.

But lately, what software has becoming why you can start implementing generative AI to write software is because programmers have started not to write their own code, but to pull code segments together from lots of different places. So that's number one. And a lot of people are not tracking well, where they're pulling these software packages from, and that's a serious problem.

And that's why you have supply chain, because the code you pull from could be malicious, the code you pull from can have bugs and all this sort of stuff that you don't know until you know. Now the issue is, once bugs are reported, you know, there's no understanding of where they came from or where your own software came from. And if you're impacted, this is why a soft SBO m software bill of materials has become a critical discussion in the cybersecurity space as well as in the coding space.

Because you need to understand where these packages are being pulled from to know whether or not you have potential issues. So for example, if you're a defense contractor and you're pulling code in from China, that's a problem in all likelihood, as an example. However, the other things are, the other part of it is you might know where you are pulling your code in from, but you don't know where your vendors are pulling their code in from.

And then, so you have first level problems of writing your own codes, second level problems of pulling in other code, you might purchase third level problems where you're pulling in code that other people have written. Then you have the, where are these people pulling their code in from? And in some cases, supply chain attacks have been down four or five levels of coding where smart criminals have gone in and figured out, wait a second, I know this program pulls in this library of code, and that program pulls in that library of code, and that program pulls in that library of code.

So I'm gonna embed something, a malicious piece of code five layers down that I know will eventually be pulled into the software of my target. And that's why these software bills and materials are critical, but it's also why it's so difficult because you, if you have LLA or generative AI pulling software together, and it, it has to track, and maybe it can track better than people, but this is a concern that it's a ma, you know, you use the term octopus, but it's a massive octopus with countless arms and arms of arms and arms that boards need to start putting some governance in place. But it's a serious issue.

And if you don't have a firm understanding, and even if you do have a firm understanding of where software is coming from, you're still likely not fully visible, and this needs to be addressed, then I don't know if the study goes as far into the depth or it was a trivial marketing study, but it needs to be addressed. I'll say it that way. I mean, you know, the, the gist of the survey was if you don't have visibility, you're much more likely to suffer a breach than if you do have visibility.

I don't necessarily buy into that. I think there's visibility and vis, and then there's visibility, right? Well, I, yeah, I mean that's, well, I think at least if you're starting to look, you'll come up with the more obvious ones.

But like I mentioned, when it's five layers deep, that's where you're gonna start having the problems. And, you know, this is where, again, governance like, it, it, it's a pain to change how people program and force them to write their own originals code again. But in some levels that's, I I, that's, and until it does, you're gonna have these issues and you need to have better Cases.

That's another, I have another solution. Look, we do, we live in a world of Franken code, right? Apps today are built by stitching together components.

All of these, almost all of these components are downloaded from different repos. There's your choke point before you download a component from a repo, the repo has to sort of bless it and say, as far as we know, this is the, the latest version of this component. It's been verified, you know, vulnerability wise, it's not been tampered with.

The check sum is right, blah, blah, blah. There is a very natural choke point at the repo, wanna call it a repo firewall, maybe, where we can make sure that the components that the developers are using are safe. And at the same time, that information goes right into your sbo.

Well, well, let me actually say how that, how that did not work. Because there was a case, and again, the bit Waller case, I think I mentioned, where what happened was there was a common software utility, like, I think it was a compiler or something, and the guy who wrote this compiler that was used by millions decided, I give up. I don't want do it anymore.

So the person basically let his domain expire. Somebody else reregistered the domain, got the email that was sent to verify the person's identity. They took the compiler software, embedded malicious software into it, recompiled it, it was then a legitimate piece of software from a supposedly trusted person, and it made it all the way up into commercial products.

But that, but that, that piece was not in the repo that that's the issue, right? It was, we saw that. I wonder, did it happen?

Was it, was it open VPN Irie years ago? It turned out there were like two people that maintained o open VPN in the whole world. And, and it, you know, it, it got a vulnerability and introduced into it.

And the idea behind these repos is there's, I think there's more girth going. Anyway, hey, we've gotta end it today. I'm sorry.

It was a great discussion. I wish we had another hour, but we don't. Kimberly, IRA, John, Mike, of course.

Thanks for joining. Thank you for joining. We hope you've enjoyed Textron Gang today.

Stay tuned for Textron tv. We'll be back Monday with even more fun and more, more, more ai, more security, more DevOps, platform cloud and everything else. Have a great weekend, everyone.

I'm Ellen Shimel. We're out. Hey everyone.

Alan Shimmel back here on Techstrong tv. My next guest is Sebastian Pack ett. Yes.

I hope I got that right. Sebastian. Uh, Sebastian is VP of AI strategy at Coveo.

Welcome to Tech Drunk tv. Sebastian, it's great to have you on here. Thanks, Ireland.

Great to be here with you. Yep. So, Sebastian, even as big as AI is today, and it's all of rage, I don't meet a lot of VP of AI strategies, and I'm not sure many people in our audience do either.

Give us an idea what, what, what exactly does it mean to be that you're a VP of AI strategy, and how did you get this job? Hey, uh, it started, uh, had the beginning of, uh, of this 2000, uh, so I got my PhD in AI actually in 2006. Uh, then I worked for like eight years in a more consultant role.

Um, and I joined Coveo in 2014, um, where I, we started to deploy to the cloud. We had our, uh, usage analytics and all of all, uh, all this new data. So I started a machine learning group at Coveo to train on, uh, on the data we were gathering.

And I started the team at Coveo, built it from scratch, started developing some ML models, uh, that were there to improve search relevancy, um, and, and from there grow, uh, from an applied scientist to a director to, uh, v vice president at Coveo. Um, and now my, now my role is really to define the AI vision for Coveo and more importantly, recently where we position ourselves because this market is now really crowded as, as you know. Yes, it is.

It is. That's a great, that's a, a, a great story. And, um, you know, we didn't mention, let, let's dig into Coveo a little bit more, if it's okay with you, Sebastian.

First of all, let's start with the website. What's the URL? com.

com. VEO. Just the way it looks on the screen here exactly, just so its spelled on the screen.

Um, give us, if you can, for people who haven't heard of Coveo, I mean, you gave us a little bit here, but give us a little bit more depth, if you will, about, you know, what, what Coveo does, solutions problem solving, how to engage, stuff like that. Yeah, sure. Um, so, uh, for the last 20 years, so a lot, uh, uh, way before the ai, uh, hype, uh, COO was, uh, already a market and industry leader into enterprise search.

So we've been doing enterprise search for two decades now. Um, and what we provide is a relevance platform where we can index the, uh, unstructured data that an enterprise has within all their enterprise silos. So most enterprises have hundreds of applications they're using.

Their data is scattered all around, um, the place, so Kao can index it. What's really a different, uh, differentiator for Kao is that at the same time that we index the content, we also index the access rights so that we can provide a unified, uh, secured search for our customers. Um, we have big customers in, uh, commerce where, uh, we provide the product search and discovery.

So when you go shopping on some of our customer's websites, it'll be covered providing the search results. Um, we also have, uh, a knowledge line of business where, uh, for, uh, self-service and, um, the, uh, workplace environment, we can provide a search and question answering and, and place deflection services. So this is where people go on the support website, they ask a question because something is broken, something is not working as expected, or they just need more information about the product they just bought.

And we can provide direct answers to them. Um, and if they go into submitting a case, we even follow what they are asking for, and we can, uh, answer them directly so they get the, the end user get their answer faster. And for our customers, they, they don't have to pick up their phones.

So it's, uh, it's easier for them. It's what we call the case deflection. So the user is happy because he found the information by himself more rapidly, and the company is happy also because, uh, it reduces their support cost.

Um, so in terms of generative AI and, and, uh, what we provide is a fully managed solution for question answering. So something that can be a service that can be deployed in less than than a day. Uh, you, you can have a generative answering, just like what you see right now on Google.

When you ask a question, you have the answer right below the, the search box. This is exactly what Covill offers, but not for the internet, for each enterprises. So on our customer's website, when people search on their websites, they can get this slide the same, um, generative answering, uh, capability.

And for agents, uh, or any generative AI applications, we provide APIs so that, uh, our customers or SI partners can build solution on top of Coveo. So if I summarize, I think Coveo is really the information access layer for the enterprise knowledge, enterprise information, so that other applications like search, like, uh, generative solutions, like agents can have access to the enterprise knowledge in a secure manner. Uh, and, and it's also really scalable and can scale to, to any number of users or agents.

Love it. Thank you for that. I appreciate it.

All right, so let's start to some recent news. Kvi Kvi recently announced you guys launched an MCP server, and that this MCP server is available in the, uh, agent force from Salesforce, right? Um, this is where they're building all of their agent functionality, and it's not just Salesforce's agents.

It's, it's kind of like a marketplace for a lot of people's agent technology. Um, why Agent force Sebastian? And, you know, you know, I, I think, I think our audience knows what an MCP server is.

Let's say some people out here don't know what an MCP server is. Why don't we start there and then tell us why Agent force? Okay, so, uh, sure.

Um, so MCP is for model context protocol. So it's an open standard, uh, that was released actually less than a year ago, and so late 2004, 2024, uh, by, uh, entropic. Andro, Yeah.

Yeah. And it's becoming a standard, but it's really nascent. And so it started to really pick up in, uh, the beginning of the year.

Uh, and the, uh, what it allows is to, uh, give a standard protocol for agents to get access to external tools, data or context. So it allows agents to communicate with external systems, um, and Coveo being one, uh, providing search services, uh, for the agent. So an agent can call a Cove OMCP server to, uh, access the search capability and get back the information from the enterprise.

Um, what Salesforce announced last week was that Agent Force was now supporting MCP, uh, servers, or will support it really soon. Um, and CO is one of the first, uh, initial partners that, uh, Salesforce announced with, um, we already provide, uh, what they called an agent force action, which was the, the previous way to connect an agent force agent to caveo. So it's already possible to use an agent force action to connect an agent force agent to the overall enterprise knowledge.

So things that are in Salesforce or outside Salesforce, um, and provide this information within the agent force agent. What they announced is the MCP support and coveo, it will work with Salesforce to provide it through their, their, um, agent marketplace. Um, excellent.

And what's, what's really interesting for Coveo is being this, uh, um, information access layer and, and why Agent Force, because Salesforce is a big ecosystem where it's a big partner of caveo. We have already a lot of, uh, common customer, um, and, uh, we were already providing for, uh, for Salesforce agent or Salesforce self service portals and search capabilities and question answering capabilities. Um, and, um, but Coveo also, uh, is providing connectivity to Salesforce, but it's also providing connectivity to other agent tech platforms.

So like Microsoft copilot, like Amazon Bedrock, like, so all the big players out there have their agent tech platforms. And, uh, really wants to position itself as the inf as the agnostic information access layer for an enterprise so that all these agents can have access to the same information, uh, really easily. Um, so that, uh, for an enterprise, most enterprises actually use many of these platforms.

So all big enterprises will have a Salesforce, uh, implementation, Microsoft one and others, uh, and SAP one. So, and we will connect to, to these different, uh, ecosystems and provide, uh, the enterprise knowledge to all of them. Sebastian, this I think is very helpful to our audience.

I've got one more line of questioning for you. So you're right. Uh, MCP very quickly has become a defacto standard.

I, I did see recently the Linux Foundation came out with like this a to a project, the agent to agent project Google, uh, contributed some open source, uh, or open source and contributed some, some software to it, some of the other big players involved, you know, some of the biggest names in tech, but certainly MCP has quickly become the standard, as you said, come, comes from anthropic. It seems like there's a lot of MCP servers out there. There's probably multiple MCP servers available on Agent Force even.

What, what's special about the Coveo one versus other MCP servers? Um, I think right now, uh, every enterprises that have APIs will provide MCP servers to, to wrap their APIs and make them accessible to agents. So I think what makes the CO MCP server different is not necessarily the MCP server itself, but it's the capabilities that COO can provide to our customers.

And this is everything I thought about before survey, uh, secured enterprise search, and where you can have access to all your enterprise knowledge, not only the knowledge within a specific ecosystem like Salesforce or Microsoft. Um, so I think this is this, uh, this broad access to, or, uh, broad access to the information within enterprise is really the strength of coveo. Uh, so we do have connectors to connect to all the enterprise systems.

We provide one access point, which is, which is any unified index to access all this information. And this is way better than what we've seen out there. So, as you said, there's many multiple, there's many, uh, MCP servers and all systems have theirs MCP server.

But if you want your agent to actually have access to, I don't know, 10 data sources, it's way easier to contact one MCP server at coveo than to contact the 10 different sources and ask then your agent to figure out which information is the right one for, uh, answering any, uh, customer or any user question. So, um, I see this as the same trend that was, uh, 15 to 20 years ago about federated Search bet, uh, and unified search. So if you contact 10 different sources, it's really hard to come up with a, a good relevance, uh, for all.

There's these different sources. Having all of this within a unified index is, is really valuable, and you get really, uh, more relevant results out of it That's available on Agent Force itself. com though, you can probably grab it there as well, right?

Uh, yes, we will, of course, uh, ize, uh, um, where to get access to it. And MCP is really, uh, interesting. It's really new.

Um, they just released, uh, in June 18, uh, the specification for, uh, identification and secure connectivity. So, uh, this is really a, a stepping stone, making it, uh, available and making it, uh, usable within an enterprise setup. Um, so you were saying when, when is it gonna get the ground running?

Uh, I think we will get there. Uh, currently it's, it's still a lot of experimentations company trying it out. Um, uh, so we provide a code, they people can, enterprises can deploy them, can try it out.

Um, but uh, now that we have a identification and security protocol, uh, now it's gonna be, uh, more enterprise ready. So I think that absolutely, this was a missing pieces. Absolutely.

Sebastian, thank you so much for coming here on Thanks Junk tv. Good luck to you and Coveo. You know, this, obviously this whole MPC and AG agentic AI thing is, you know, evolving day by day, minute by minute, hour by hour.

So we'll be watching, come back again and keep us posted. Okay. Thanks Alan.

Thank you. Sebastian Packard, VP of AI strategy at Coveo here on Tech Drunk tv. We're gonna take a break.

We'll be back with some more. Hey guys, thanks for the throw. We're here with Pablo Ry, who's the CEO for space lift, and we're talking about, well, they've just picked up at $51 million in additional funding for infrastructure automation platform, and we're gonna dive into what that means exactly.

Pavo, welcome to the show. Thanks for having me. I think everybody is familiar with, you know, the various infrastructure as code tools that we have, and they're pretty sure that that's some form of automation in their minds, but is that really enough?

And what's the difference between what folks are doing with those tools and what Space Lift enables? Yeah, that's correct. This general thought is that infrastructure as code was supposed to solve all the problems with the infrastructure, right?

And it's already partially an automation tool. You can copy, you can version, you can share the code, uh, of your infrastructure. But what happens in reality is that it still lacks the collaboration layer, the security layer, the auditing, uh, ability for larger teams to work simultaneously in making changes, right?

So it is getting much closer to being a proper solution for emerging infrastructure as code at scale. But you need still need, especially at scale, something to have an overview of the whole infrastructure and make sure that people don't step on each other's toes. And I think part of the issue is we wanted to enable developers to provision infrastructure themselves in the name of productivity.

But when I look at it, um, a lot of times, well, a, we just kind of reinvent the same code over and over again because we don't have a way to share it, and b, they make a lot of mistakes. And it seems like that's where a lot of the vulnerabilities come from. So, um, and we reach some level of pain where everybody's kind of willing to consider it a different approach.

Yeah, that's exactly correct. Um, so what's happening is you can't really give ability to selfer to all the engineers unless you have the right guards and framework in place, right? So, uh, sounds great on paper, opening up everything, having everyone spin their own infrastructure.

But a, it's super risky because you don't know what's gonna happen. You don't know what's, who's gonna spin what and when. So it's very costly because you have no overview over what's happening to the infrastructure and what changes, uh, were being made.

And also, you don't have any way to properly audit and see the lock of all the changes. So you do need to have a layer on top and build the, I I like to call it like a, a, um, you build like a, a machine, you set up the machine with the right guardrails, uh, and protections. So the rest of the company can do self-service, but under certain conditions and policies, right?

So nothing bad can happen to the infrastructure. We of course, have been all talking about the rise of platform engineering as late. Does this kind of play into that whole methodology?

Oh yeah, of course, of course. So, uh, the idea is that when people use solutions like space lift for orchestration, you can have one kind of centralized platform team that sets up the right best practices for the teams and then kind of exposes and enables teams to use the reus reusable components, the automations, uh, while the level of security and compliances there, right? Because the platform is able to set it up so that nothing bad happens to the infrastructure.

Who's taking the lead on this in these organization who kinda wakes up one morning and says, we need to rethink our entire approach to managing infrastructure? Yeah, it's usually, it's usually those who deal with everyday pain of managing infrastructure at scale. Uh, at some point, you know, there's this balance between going very fast or trying to go super fast as the organization while maintaining control and compliance.

And initially most of companies care a lot about how do I go faster, right? Like, how do I deploy faster? My application are being deployed very fast daily, how can I make sure that infrastructure is being deployed equally as fast?

The issue is that the more you deploy and the more frequently you deploy and the more complexity you're getting, and the more people are involved, it's taking longer, there are more mistakes. Uh, everything is pretty much crumbling. So usually what we see is that a leader of the infrastructure division world realize and see for numbers that maybe took his or her team a day or two to deploy something, and now it's taking a week.

And it turns out there are so many manual processes, so many checks, so many rollbacks and everything happening that it's just breaking the whole team because they need to spend more time on this. They can spend time on other things than what does it mean? Do I need to hire more people?

Like how do I make sense of it? And then there's a lot of pressure, of course, from the management to deploy safely and also as frequently as possible, right? Because the application developers are complaining.

So it's usually, it's building up over time and is the leader usually of the infrastructure team that realize we need to solve this problem. We're, of course, are all living in the age of AI now, and everybody's scratching their head a little bit to determine whether or not AI will make the provisioning of infrastructure and the management thereof better or worse. What's your thought?

We believe, and I believe we believe at Space Lift, that it'll make things better, right? Because there's so much complexity and there's so much, you know, nitty gritty in the infrastructure, that if you're able to fine tune AI to work well, infrastructure as code and infrastructure in general is gonna be of great benefit, right? Imagine now an engineer might need to spend countless hours figuring out what happened to certain resources.

If you're able to train, you know, an AI agent to do it for, for to developer, the agent could do it much faster. So I think that the question is not is it gonna help? It's like, when are we gonna get to the point where AI is helpful enough so that the end user is not afraid of using it, right?

Or some like middle ground where it's supervised by the human. So AI is doing a lot of work on the infrastructure, but there is a human, at least initially, that approves what happens to the infrastructure. Okay, well, $51 million may not be, you know, open AI kind of money, but it's nothing to sneeze at.

What's your thought for how you guys are gonna apply that and what needs to be done? Since inception, we've been a single product company, right? We are very focused on day one operations.

So infrastructure provisioning with this new run of funding that we have two new products that we've been working on for a while now that we should be able to release by the end of the year. Uh, all of them are for the same area, so infrastructure management. But we would like to cover day two operations.

So, you know, maintenance of infrastructure updates and so on, as well as day zero. So we're planning to invest in new products, of course, at the same time, we're investing more into the go-to market of our core product, uh, the, you know, the orchestration platform. Uh, and of course to be able to further develop and invest in the existing platform and existing product, and then add to new products, that does require a lot of resources, human resources, financial resources.

So this is where we'll be putting the money, uh, to work. You know, we, we do see that there is still a lot of potential in the market. We do want to become, uh, uh, we would like to call it like an infrastructure operating system of choice in the future.

Uh, and that's, we're investing the, the resources. So, so invest, uh, going up market, sending other customers, expanding our existing product, but also building and releasing new products. How hard is it to set all this up?

'cause I think part of the issue that a lot of DevOps teams will encounter is they'll conceptually agree with the idea, but they don't have a lot of time to go fix it, so they live with it. So, uh, in a platform like yours, how long does it take to get people started? Well, it's actually pretty straightforward.

So you connect your GitHub GitLab repo with, uh, space lift, and that's pretty much where we are in, I think the, then the question is how much time and how quickly you're gonna adapt space lift to work for you and for your infrastructure. So it's usually honestly a matter of days, um, when the team kind of sits down and sets up space lift in a way that is beneficial and useful for, for them, right? So it's usually a few days that, uh, it's enough to at least get to the first point, right?

Like the first aha moment or, or value. And where does the platform itself run? Is it a cloud service or is it something I can deploy on myself?

What's the, what are my options? Our options expanded significantly in the last 12 to 18 months. So when we started the company, you could run this either fully on the public cloud operated by space lift, or you could also use private workers where you put the agents or execution on your, in your own infrastructure.

Uh, what we implement in the last 18 months, a now you can also self-host the control plane. So you can fully self-host space lift either on AWS or Azure or GCP. Um, very soon you'll be able to self host it, uh, fully on-prem air gap.

So this also coming, and we're running a POC with one of larger customers, also as a first company in the space. We are federal certified, so we'll be, we're available on federal marketplace. And very soon, first, customers will be a able to use our FedRAMP, uh, instance to, to operate.

We're, we're seeing a lot more government, uh, interest in our solution. We're also seeing all of our customers serve the government, which is why we went and we decided to get, uh, our product firearm certified. Of course, in this whole space, there's a lot of debate about, um, which open source infrastructure is code tool to use.

There's Terraform Open Tofu, there's also a couple of proprietary options, but, you know, I think it's been a year or so since that whole fork in the road came. And so where are we right now? Or in your mind, are these two communities kinda co-joined or are they separate, or are they maybe even gonna come together someday?

I don't know what's gonna happen in a year or two. Uh, for now, I can just tell you what we're seeing in the market. Uh, more than half of space of customers are running primarily on open tofu.

So we're seeing an attraction of open tofu. Uh, we're getting signals that the large enterprise, like Fidelity investments, uh, are moving completely to open tofu. Uh, there are a few other companies moving to Open Tofu, so we're seeing more and more traction with, uh, with Open Tofu.

We didn't know what's gonna happen or IBM is gonna decide to do about, uh, Terraform. Uh, it's an interesting thing where people and users switching to open FU does not happen overnight, right? It's not that you release something and everyone says hooray, and it moves off Terraform and moves to open tofu.

It's more of a, you know, gradual process. And the more examples users and practitioners see that large enterprise are moving to open tofu, the more they adapt. So you start creating the snowball effect that the more I see, oh, this big company moved, this big company moved.

Well, maybe I should move as well, right? You're gaining more credibility. So, uh, I view it as a very long term play.

And, uh, the battle or, or the war, if you wanna call it, is, is still pretty much undecided. So what's that one thing you see DevOps teams doing when it comes to infrastructure as code that just makes you shake your head a little bit and say, folks, maybe we need to be a little bit smarter than that. A lot of manual processes, kind of, uh, there is a lot of baggage.

We're, you know, we very often I do go and visit the customers or the prospects, and I start talking about how they run the infrastructure and just people are often justus to doing things certain way and in very often in non-optimal way. So, very recently, I was a very large financial institution, and there was this particular thing that took seven or eight days for a single, um, even DevOps engineer to set up an environment just, just to simplify, right? With space, it'll take them six hours, but because they've been used to this very manual process where you need to ensure the safety and security of the infrastructure is taking them seven, eight days.

So I think the, the biggest part for me is sometimes people are just used to doing things certain way, and I might not be aware that those things could be automate and don't match faster in a much more secure and reliable way. Folks, you heard in here all good things and bad things related to DevOps. Kind of start with the infrastructure as code tools solve that problem, and all kinds of interesting good things will happen downstream from there.

Pavel, thanks for being on the show. All Right, thanks for having me. All right, and back to you guys.

And Steve. Hi everyone. Um, thank you very much for joining the Techstrong AI Summit.

Um, my name is Dr. Katie Paxton fair, and I'm gonna talk about, um, kind of where I see some of the big AI threats coming from. Quite often in the media, we've really heard about AI being like a force multiplier, maybe hack bots you've heard of, and a lot of people are worried about AI really coming to hack queue.

Um, now I don't know whether or not AI is really gonna hack you, but I think it might just get you hacked because there is a real growing challenge of securing the AI slash ml supply chain that I think are not enough organizations are really thinking about and talking about as they kind of go into the new AI era. So my name is Dr. Katie Paxton, fear, I'm a principal security researcher at Traceable by Harness, and this is my AI threat top five.

So these are the five risks. So when I speak to organizations and how they're adapting to the new AI era, um, this is kind of some of the threats that I can see as a hacker myself. So the first one is really on de rely developers relying on AI generated code.

So this could be your developers using tools like maybe windsurf. You've probably heard a lot about vibe coding. So that's non-developers being able to use, uh, things like cursor in order to actually generate entire applications from scratch.

Uh, a lot of the time now we're actually seeing that developers are not even professional developers. They're maybe just, you know, they can be your sales staff, they could be your marketing team. They're not professionals.

And with that, we're seeing a lot of threats being reintroduced into co bases. The second risk we've got is about data platforms. Now, a lot of organizations nowadays really understand the value of data, and perhaps even more so in today's AI era that we've got, there are these platforms that ingest data and can give you a lot of insights into what that data shows and how you might wanna change your business based on the results.

Unfortunately, with so much customer data, it makes it a real target for attackers. Uh, these platforms are often targeted. Um, they are often, you know, people are getting things like phishing emails.

It's one of the key ways attackers are actually targeting organizations. Thirdly, we've got AI tooling within organizations. There's a lot of tools that organizations can now use in order to be more efficient or to really get the benefits of ai.

And it's important to recognize that these two can be a risk of both a security risk, but also a business risk. For my, for my favorite, fourthly here is suppliers and suppliers and suppliers of suppliers using ai. You know, nowadays your security is not restricted to the brown bounds of your organization.

It encompasses a ton of different, um, products and services that you use. And while you may not be using AI in these ways, there is no guarantee that one of your suppliers isn't using AI generated code or using a data platform, and your customer data or your data as a company may be in there. And finally, it's how organizations are implementing AI into their products.

So, thinking about, you know, nowadays every application has AI in it. What does that actually look like? What data is being set?

So let's start with number one, developers relying on AI generated code. Look, it's fairly obvious. Developers don't wanna write tedious code, right?

Developers often say they think AI to be kind of like a calculator. It's a tool that makes their job easier. It allows them to kind of automate very simple tasks that are just time consuming.

Or as someone on Reddit, put it, use it like an intern, let it do all the work. You don't, don't feel like doing, but check its work. And if you look at the stats, you know, people are using AI generated code more and more and more.

Um, this is becoming really, really common. So what can we actually do about it? I think everyone's first reaction is to go ban AI generated code, like ban ai.

Stop it. You're not having cursor, you can't have windsurf, you can't have anything. I think it's wrong because I think that just leads to developers using AI kind of unpermitted by their organization.

Instead, I think you have other options. Invest in basic SaaS scanners in the build. Um, really make sure that you have got that in the pipeline with developers.

Give developers the tools they need to work. You know, give them the ability to use AI generator code. Give them ideas.

Now, developers shouldn't be using AI to write code in a language they're not familiar with. Um, but at the end of the day, if they are, we can encourage the use of prompts that feature security and make sure when we are actually deploying that code that, you know, it's, it's going through a proper peer review process. A lot of people will say they do, um, you know, your, your typical reviews, code reviews, but in fact, that's just someone seeing you looking over it and go, eh, it looks to out, right?

So really encouraging, you know, developers to take a step back and actually read their code and make sure they understand that. Number two, data platforms ingesting all of your data. I think this is very, very attractive for a lot of organizations because it allows you to put all of your data in there and connect and then get those insights from that data by just uploading it.

Instead of needing to hire a data scientist or an entire data team or having to buy expensive software. Instead, you can put everything into the platform and then use, create or use interactive dashboards, prepare reports, and they can really start to understand the trends. And one of the kind of key advantages here is that many of them are low-code, no-code systems.

They do not require, uh, experience. They do not require expertise. They require an understanding of the data.

And unsurprisingly, with so much data, attackers specifically target these. Um, whether or not that's something like we saw Snowflake recently and it's gonna be your more typical, um, you know, malicious download link, run it, and then the malware gets executed. Or it's in a traditional kind of like phishing attack where people were explicitly looking for API keys for things like Snowflake.

Data ingest platforms in general are huge targets. They contain so much sensitive data, and because they're used by a lot of different teams within an organization, often it's not gonna have a lot of security oversight, or it may be completely invisible to security teams. You know, with info stealer malware, this is where we're seeing things, right?

We've got malware that specifically target these data platforms, API keys, um, or even the accidental commit of API keys to these platforms. These are all being used as a way to get into data ingest platforms. Three AI tooling within organizations within an organization.

There is lots of different, uh, ways that people use ai. In fact, I would hazard to guess that most people listening here have had a meeting where an AI has joined in order to, um, take notes really common. So things like note taking applications, but also, you know, if you do a lot of long documents, a lot of people will throw it into chat GPT and say, Hey, summarize this.

Some people will use speech generation or they'll use it for content writing. And it's really important that when we're actually thinking about these tools, we want businesses to be more effective. We want to give people the tools that they could have.

And then it probably won't be surprising for you to hear that I'm saying don't ban ai. Again. I think, you know, it's very tempting to outright ban.

It's very tempting to go, you are not allowed to use this stop. The problem is, is that if you ban people, they're probably gonna do it and just not tell you. And that's far worse for your security.

As we say at traceable, one of the best things you can have is visibility into what your applications are doing. And the problem is, if you ban people from doing it, they're just gonna do it sneakily. They'll do it in a way you don't have visibility on.

So set limits on things like AI note takers. You know, you should not have meetings where customer data is being shared or intellectual property is being shared. Um, and ensure that really whenever you're using a third party tool, that you have those, um, SLAs in place to figure out what happens if there's a breach.

How are you gonna protect your customer's data? Next up, we have suppliers and suppliers of suppliers using ai. Look, no application lives in a vacuum.

It doesn't matter what you work in or anything. A lot of people still kind of have this perspective that applications are monoliths and they just exist on their own. This is not the case.

No application lives in a complete vacuum. Every single one has dependencies, they have containers, they have builds, and you need a continuous solution that really understands not just, you know, a single application, but an entire suite. And this is particularly true for integrations.

You know, when we talk about API security, APIs are used everywhere. This is not something where you can just do a search for API and just find it, you know, there are huge lists of public APIs that anyone can use and take advantage of, and APIs will use other APIs. You've got this API supply chain.

So even if you go, well, I'm not using ai, this isn't important to me. You don't know that. Maybe you are using one of these like APIs and they're using ai, you know, you have got a risk of that third party and it's really, really important that you know that third party and you are aware of that risk.

Um, and that's partly, you know, why having third party API visibility so important. And finally, we've got implementing AI into your products. It is so easy to add ai.

You know, you choose your preferred ai, maybe you go for philanthropic, you go to chat GP t you go to pilot, you top up your account with some credits and uh, you implement like a really easy to use restful API. You can even get AI to generate it for you. You don't need to know how to do that either.

Or you can use an existing library built into the language. Um, one thing that we're really seeing is obviously the rise of agent ai. This is the idea that AI is not just a chat bot anymore.

Now it's able to act autonomously. It's able to say, okay, you know, you ask, Hey, sorry, bye. Like, I wanna fly to the us right?

And it will go and look at flights, it will look at hotels for you, and then it will go and be able to actually go there and pay for it. And really, the intention here isn't to do this securely, it's to do it really well. And that's a very difficult attack surface because you have so many different components.

You have the AI chat bot that you are using. You have your AI agents that are going and accessing, you know, the website for, um, uh, the, the, uh, airline or the hotel. Then it's using maybe their APIs in order to connect to, um, and actually make payments that's connect to a payment provider.

Like this is not a simple attack surface. And certainly I think most organizations are very quick to jump on the, okay, we wanna have ai, we have AI everywhere. We really understand the value of this.

But it's not that easy when you do handle AI input into applications. Just do not trust the input. Um, look, if you look at kind of the sensitive data that goes through, you know, I think customer data, right?

It is crazy how many people just trust it. And if you look at the stats, the majority of people say part of what they use AI for is literally just understanding data. Um, which is crazy to me because the AI can't necessarily do that in the way that, you know, a human can.

So my advice, um, is, you know, perhaps unsurprisingly, uh, really look for when you see restricted data types. So this is like a screenshot from the traceable product. So we do this automatically, um, but look for things like prompt injection, look for things like sensitive data.

Uh, there's the OS LM top 10, which is a fantastic project that really helps you understand the risk of, uh, L LLMs and applications. But you've gotta always validate inputs, right? You cannot trust what an AI gives you the same as you can trust a human.

So TLDR zero trust is a philosophy is gonna be more important than ever in the next few years, right? It is more important than ever that not just you are secure, but your third parties are. And it is a must that you should include third party attacks in your risk assessments and security policies.

If your security tool doesn't give you visibility into those third parties, it is not the security tool you wanna be using. AI isn't going away, it's not a phase. It's here to stay and we need to figure out how to use it.

This is not something that we can sweep under the rug anymore, right? We need to jump on this and make sure that we are being proactive. If we get too caught up in AI security like jailbreaks, we're really not seeing the bigger picture because it's not just about jailbreaks, it's about the entire AI ecosystem.

And five, ai AI infrastructure is gonna be targeted more in 2025 and beyond. And I'm gonna leave you on that really cheery note. Uh, thank you very much everybody.

I hope you have a great rest of the conference. Hey everybody, welcome back to the Open Source Summit. We're here with Ben and Ruben and we're talking about something called cmera, an open source project that's in the Linux Foundation that, as far as I understand, it puts a set of APIs in front of your network that makes it easier to invoke.

And who knows, maybe those network and DevOps people can get along. Gentlemen, welcome to show. Thank you, Allen.

Alright, now your company is Cable Labs, correct? And they are heavily involved in this project. So give us the backstory of how it came together and, and and what it does and because it's been around a little bit while, and then maybe what's latest and greatest.

Sure. So Cable Labs is a nonprofit that does r and d for the cable industry, and our members pay us to get access to all of our patents in any code we write. So our members are companies like Comcast Charter or Rogers Vodafone.

And what we have going on there right now is a big initiative. I'm in the software group. They're, uh, called Network as a Service.

So, and that's when we got engaged in Kamara, um, about a year ago. And so what that means is we're trying to figure out, um, how to bring a wire line perspective to the Kamara project. It started off as mobile first, and there's a lot of great APIs to interface, uh, with a mobile network.

The those same APIs can, uh, can, can sit on top of a wired network as well. So fraud detection is one example. If you're making a purchase on your phone or if you're on your home wifi, uh, financial institutions will be interested in making a call to be able to tell what level of confidence this is a fraudulent transaction or not.

So that's one example of a Kamara API that could be used both on the mobile front and on the wired front. Alright, Now, GSMA has been around a while. It's, it's memorable.

So how did you guys kinda get involved in this whole thing? Well, actually, um, GSMA get into this project with done open gate initiative, right? Um, open the Open Gate initiative, take part of kymera as, uh, the main, the, the, the main flight to, to development the kind of code that tried to standardize these APIs, right?

The, the GSMA, uh, job in this ecosystem is to enable and convince and expose all the benefits that API at as, as the kyara have, um, make it, uh, made it for the ecosystem to make more seamless connection between the layers of the value change of an enterprises or a, uh, a channel partners and a and a mobile operators. With these exposures, APIs is going to be more seamless for everybody. And the work of GSMA with the mobile work on support and with all the bands and with all the white papers and knowledge that we can bring to the table, we support this camera initiative with the open grade initiative around the world.

So will this make not just the, the networks themselves a little more accessible, but can I maybe bring the various teams together around an API? Because there's developers, they're familiar with writing code, there's DevOps engineers that are managing infrastructure, but the network's always been kinda off to the side here a little bit. So is this gonna at some point, converge a little bit more than we've seen in the past?

Yeah, I think so. And, and you, you point out a, an interesting distinction where you've got, um, we call it northbound and southbound on the API. So northbound is a third party application developer wants to consume these APIs, could also be a network operator themself could consume their own APIs for some sort of support, uh, may maybe monitoring what whatever they want to do.

They can use that also. But then the southbound side is once you get the intent, they, they're called intent based APIs. You intend to interact with the network.

The southbound side is then the network config behind it. That has to happen to make that possible. For example, if you have an application developer writing a video streaming app, and they want to allocate dedicated bandwidth for a session, I need this amount of throughput and this amount of latency and jitter, they can create, uh, use the quality on demand API and create a session.

And then on the southbound side, it makes everything happen to reserve that dedicated bandwidth session, whether on wireless or a wire line, uh, connection To his point, I feel like the applications we're trying to build are more distributed, they're more latency sensitive than ever. And are we gonna make it easier for developers to understand that? Because I think a lot of times they write applications and they think the network will just magically do something.

They're right. Well, that, that's, that's the beauty of, of this, uh, camera once that the developers community join this kind of effort in order to enrich that kind of code that that, that could, uh, have some kind of, um, uh, advance, uh, uh, project to, uh, to these APIs, right? With this, with this, uh, saying this APIs has, has been, um, uh, has been said, this API process get it a, some kind of certification, right?

That the GSMA is running when when the certification, the developers contribute something kamara gets right in the code, what hundreds of APIs or the priorit, the prioritization of APIs launched to the market, then get it into the certification process of GSMA that can, um, put it the stamp that say, Hey, this is a certification of GSMA with a ca with a camara code, with the developers that are, join this kind of community and enrich that kind of code. And with this, the ecosystem of the APIs of developers, channel partners, um, um, mobile operators with aara coordination in the coding is going to expose this DSMA open gateway to all the operators around the world. And that will be more consistent versus having 300 companies then implemented kamara.

But did it slightly differently. Well, I, as, as you say that you, you, you, you, you, you hit it is is going to be a seamless process for everybody. It is going to create more value of, of the, uh, uh, in the, between the, the layers of the, of the ecosystem.

And it's going to, um, prioritize the monetization of the 5G environment, right? In order to, to, to have more, uh, uh, more and better products to the market. They are, uh, uh, better coding, better, uh, knowledge of the developer's community to enrich this kind of a, of APIs, right?

It's, it's, it's a, it's a complete project that tried to involve all the ecosystem around the world and as, as, as, as Ben said, tried to get it, uh, um, as seamless process north to south to south to north, and to try to get it west, east, east, west in a interoperability with the operators, right? That is, that is going to be the, the, the, the main objective at the end of this. Yeah.

So what's next for Kamara? You know, you're in charge of the skunk works, what's going on? Mm, Yeah, so, um, Reuben mentioned East West APIs.

I wanna talk about that for a minute because that is something that is actively going on right now. Um, so the North south we talked about already North is coming in from the application developer south, making the configuration on the network itself. East West is now between different operators.

So you might have a Verizon, at and t T-Mobile, Rogers Vodafone, uh, an application developer does not want to have to write to all of these different network operators to do the same thing, to check the avail availability of a cell phone number, for example. That should be one API call. So East West is now federating and connecting these different, uh, companies together in a, in one federated, API and Ericsson and Vonage announced a, a, a split off company called aduna, which is working to achieve that.

Exactly. So that's, that's on the mobile front. And me at Cable Labs, we're working the same thing with our members trying to do a similar thing to federate, uh, the API calls and encourage all of our members to join Kamara and participate, uh, in the project.

How do the other folks watching this get involved? I mean, do I have to work for a telecom company or is this a community where you're inviting folks and where do they sign Up? Yeah, so it's on GitHub, all the code is available, all the APIs that we define, we define the APIs and it's up to the network operators to then implement those APIs.

So we define the endpoints and the data schema behind the APIs, uh, and it's public. You can go look at it there. If you want to contribute, then you do need to sign a license agreement with, with the Linux Foundation and to join an actual, if you wanna do like a pull request and push some code up to the project, then you do, then you have to join there, but you, it's free for anyone to go take a look at it.

com/kamara, project Kamara or Kamara project, Kamara Project. I'll Start. Right?

Last question to you. We are all talking about AI agents, right? And there's gonna be, I don't know, millions of these eventually sitting on our networks that are all trying to call something through some sort of API is the current infrastructure that we are basing everything on really designed for that.

And, you know, does, do we need to kind of put a layer of isolation or abstraction between all these AI agents that are calling APIs so we can build out the infrastructure without necessarily like affecting the software? Well, that's, that's a very interesting question of these, um, the respo, the, the response is, at this point of time, we need to get it into some kind of a step, right? We are going to get it into try to monetize the 5G with the resources that we already have, but now it's coming to six G and, and, and that is some awarding and, and, and, and papers and news that, uh, GSMA is bringing to the table with the, with the, with all the white papers and information that we, that, that, that we have it for the AWA Congress and all the events that we organize.

But the response, the answer for your, for your question is yes, but we need to get it into, into, into some kind of step by step to support that kind, uh, of traffic or notification, uh, depending of this API's calling. And, but, um, but the responsible of all this is all the telecommunications, um, change operators association, the developers, partner channels, everybody need to having, uh, uh, that in, in the top of their mind that we need to join that kind of, uh, uh, of technology step by step, but as soon, as soon as they can, right? This is, um, get it involved budget and forecast and everything like that, but we need to get it into the first step to try to be prepared for their coming, right?

Because, uh, for this open gateway initiative, uh, we saw that it's something that need to be done, right? Is, is, is, is a seamless process and, um, is, is is going to benefit the ecosystem in technology in a seamless process with a correct monetization. I'd like to add one more thing, um, on the AI front.

Kamara did recently kick off an initiative to explore MCP, the model context protocol and try and figure out how do we make all of the API calls, um, so that AI can interact with them in an easier fashion. So we are actively looking into that as well. All right.

And we were talking about a two A here at the show. Is that on your radar screen as well? Yes, we're, uh, We are actively working with A two A and using AgTech AI to try and connect, um, these, these APIs.

All right. Yeah, folks, there's an old joke that says, you know, what's the one thing that a developer and the person running servers can agree on? It's the network guy's fault.

Hopefully that won't be as funny as it used to be. Gentlemen, thanks for being on the show. Thank you.

Thank you for having us. Thank you. All right, and a pleasure, and we'll be back in a minute.

Hey, everybody, we're back at the open source summit in Denver, and we're talking with Aloc from Chronosphere, who's head of product innovation, and we're gonna be talking about a thing called logs of Palooza. And I believe this has something to do with logs, and we'll jump in from there, but explain, you guys rolled this out at the show. It's brand new, as I understand it, we're gonna make logs easier to manage and maybe less noisy, Right?

0 is our ability to offer the customers control over their log data. That's the big new thing. So if you, if you, you know, if you, our, our entire platform, one of the differentiators about our platform is the ability to control data volumes through insights about what data is used, what data is not used into making informed decisions about what you should therefore collect and how you should collect it.

Or we didn't have have that for logs till today, and now we have that for logs as well, which is the big new thing. Are people log hoarding? Are they like saving so much of this stuff that they don't have to save and then the cost of storage goes up and they get yelled at by their CFO for how much storage they're consuming.

But what's the core problem? I mean, I think you hit the nail on its head. I wouldn't say log hoarding, but log hoarding sounds like they're doing this by, by choice.

I think it happens by accident. Uh, it's difficult to always make good choices about what to collect, what to not. Uh, administration teams are responsible for logging tools, not those that are using it.

The tool people that are responsible for the tools always have two sort of things they're trying to optimize for, help the engineers get the data they need to solve an instant, but also keeps cost, keep costs down in that they tend to rotate towards trying to keep more things and throw them away. And as data volumes go up, this problem gets more and more acute, more painful, uh, for them over time. Is there some way to identify which logs are more important?

Is there, do I put a classifier on them, or will the logs tell me somehow that this one matters more than that one? As a, as a user or a customer of our product, you don't have to do anything to pre prior to using the data. You don't have to do anything to your data to determine what's more useful or not.

So the ma the nice thing about our product is it gives you that insight based on your engineer's actual day-to-day usage of the product use, usage of the data. So if your engineers run queries or run que, is it just ad hoc searches or, uh, build dashboards or have alerts based off of those logs, we will know which logs are being used and which ones are not. And we can generate information or insights that give this information back to these admins that we talked about, admins, tools, owners, whatever you wanna, whoever, whatever the title of those folks are in the company, to give them quantified information about what's useful or not based on usage patterns of their teams.

But they don't, as admins or tools owners have to do anything to, to get that information prior to get the, getting the log. Zig no tagging, no marking, nothing like that. And that provides the added benefit, therefore, of streamlining the signal to noise ratio, because I can, the, the platform itself is helping me do that Exactly.

So you can, with that information, you can make pretty, uh, quick decisions as to what you need and don't need. Some of it's very obvious where you've got very low utility scores where you really, obviously no one's looking through this data. And you can look at historical time periods, go back 1530, perhaps longer, uh, long, uh, uh, uh, further, further back in time and determine was this data ever used?

Um, and then that's easy data, you can just drop, but there's other data that falls into sort of this gray area. You need it, but you don't need it in its full fidelity. You don't need all the errors in full fidelity.

You don't, every error that comes in that looks the same, but you need to know the number of those errors that happened so that data can start getting treated a bit differently. You can calculate the number of errors, but not keep all the logs. Those kinds of things are possible within our product with those insights, but also the ability to actually act in this particular fashion.

You can make those changes to this data coming into, into the product. Yeah. Um, there's always that one log or two or three or however many it is that wasn't significant at the time, but then, I don't know, there's a cybersecurity incident six months from now, and some regulator wants to look at that log, can I go find that and rehydrate that in some way that makes that still relevant?

Or how do I kind of deal with those kind of one-off anomaly cases? Yeah, I think that particular example that you mentioned, like a compliance use case where you have to keep something, I, I would split into two types of use cases. Something, a compliance use case where you need to go back to an old log.

When I say drop sample or get rid of logs, I mean to say you don't have to keep it in your performant, more expensive logging store. You may still choose as a user or a vendor to store it in an S3 bucket or some sort of object store. And yes, there's definitely a way to then rehydrate that or pull that back into your performance logging store or just query it straight into, into, into that bucket.

Uh, so that can be solved generally through that kind of a strategy. Um, however, there's situations where, you know, you might need a log where it's really necessary for incident resolution, but you maybe you didn't collect it. Those are the trickier situations, and that's, again, there is no silver bullet for that one.

Uh, you've gotta be careful about how you use a tool like chronosphere or any tool to choose to keep things, not keep things. So I don't want to claim that, uh, you know, solves those kinds of obscure problems. Uh, I think there's, you have to take care using any tool when you're making decisions about what collect, what not to collect.

Yeah, Of course, these days you can't walk down the street without somebody telling you about their great new AI thing. So am I gonna apply AI to log management and what's that gonna look like? So I'll talk to you about where, you know, what we're thinking about ai.

Um, so chronosphere is primary goal is all about making developers a lot more efficient, and that's how AI will continue to be surfaced in chronosphere. And when it comes to logs is a prime example, is being able to do things like log summarization or getting meaning outta the logs. That's a very obvious place where you'll start seeing things in Chronosphere, where it gives you insights and more insights into what these things mean, especially when they're very complicated, difficult to munch through for any one human being.

But we won't be AI washing our product, uh, and stamping AI in every AI feature every time we improve something within, within the ui. Our where do AI agents fit in the DevOps team and the workflow? Are they gonna be essentially members of the team or are they gonna like, augment DevOps engineers or a little bit of both?

It seems people are trying to figure out where that entity sits, and is it, you know, uh, a thing I give a task to and then it's autonomous and comes back? Or is it something that I kind of use that I'm invoking alongside my activities myself? So I'll, I'll give the, uh, I'll give my answer, keeping keeping in mind that this, uh, space is evolving very rapidly as we speak.

So what I say today may be, you know, again, evolution is pretty fast in this space, but generally what we've seen from customers, I'll speak to what we see from our customers on the whole, uh, it's more the augmentation of the developers or the DevOps engineers. So AgTech, you know, ai, uh, from the perspective of instant resolution or SREs, is about helping them become more efficient. Uh, so in that sense, you could say they're a member of the team, but for very dedicated tasks.

There isn't one particular AI that does, uh, agent that does it all. There is, Hey, we want to be integrating with, uh, this particular tool. I want to be able to do this particular task.

So you'll start seeing those things pop up. And in some sense, that's becoming quite necessary for teams because there's a lot more work to be done, uh, at all times. So this is actually making teams more productive.

Uh, but that's what we're seeing from people. It's not, what we're not seeing from our customers is AI doing it all. Uh, we've not yet seen full fledged end-to-end troubleshooting root cause analysis.

Here you go, and then rollbacks, and then that's a, it's a bit beyond what we see right now. People are now starting to go place where they see AI as a way to augment, uh, augment their work, make themselves more productive. And that's how I see agents as well.

It doesn't sound like SREs are gonna be replaced anytime soon, but I wonder if the opposite might be true, where the number of companies that can afford to invest in higher SREs has been somewhat limited compared to the total population that could. So might the overall number of companies that can invest in DevOps and DevOps engineering increase in the age of ai, as the cost become more accessible and the technology becomes more accessible. Uh, very likely, very likely.

I, I do see AI as like with any technology, as it becomes more and more mature, it becomes democratized in some sense. So what was otherwise only possible for certain large organizations or well funded organizations, not just ai, this is not a response just to ai, you'll start noticing that companies start adopting what a big company does over time, because now it's cheaper or things like that. And same thing I see with AI today.

There's many, many organizations that are trying to make a, you know, jump to the cloud. Many have not yet. And in that world, they need to operate very differently than they do today.

And in that world, the advancements in AI will probably make this a lot easier for them, um, because now it's accessible to them. And it hasn't been. And in, you're right, like SREs are not that easy to find.

They're not that easy to, you know, hire. It is not that many of them out there all the time. So that lack of, uh, lack, you know, the, just the lack of personnel out there available just to, just to plug in and the costs will definitely become something that can be more manageable with ai, um, making them more productive.

We've been talking about the DevOps infinite loop as long as I can remember, but it always seemed like there was a gap between the, I can observe something, but that didn't mean I could fix it. 'cause I'd have to go get a different set of tools and there'd be be a little, there's a disconnect. Will we kind of close that gap more where if I observe something, I can, within a set of policies automatically fix it?

Versus today I've observed something, but then I gotta go, you know, still set up a war room and have a discussion about who's gonna do it. Are you saying with ai could, could we do that? Yeah, I mean, I I'll say this, that it still makes engineers a bit queasy.

Uh, I cannot, can, will the world get there in reality at some point it'll get there to be very honest. But will that happen, uh, in a very foreseeable future? I think that's really dependent on how people accept some of this change.

Less so what the technology can do. Uh, it becomes more of a human question, are we okay with this? Are we okay with this on many levels?

Uh, but primarily, do I trust this? Right? Do I trust that this change won't be terrible for us?

And I think we're not in a place to make that. I mean, I wouldn't be in a place right now to make that call this second. Uh, but knowing technology one day it'll happen.

Is it today? Is it 10 years from now? Is it 15 years from now?

Is it five years from now, two years from now? It's very difficult to tell, but I can tell you that, uh, most of the customers we speak to and several engineers, I won't name names in my life, do not feel comfortable with this final step. Um, to that end, I don't think that will be closed by AI yet, but it's hard to say that one.

It seems like there's a push pull here. One thing is, you know, on one side is it's one thing to be wrong. It's another thing to be wrong at scale.

So if I let the AI do it and it goes wrong, it could be catastrophic, right? On the other end of it, the systems are getting so complex that me as a human, I can't keep up with what's going on anyway. So, you know, between those two diametrically opposed things, which one wins the day?

Uh, in the long run? In the very long run. And I don't know what that timeframe is.

I think what wins the day is, uh, uh, we have to get there. So let's get there. Which means I think at some point AI will be accept, everyone will accept AI making those changes.

Or also, I wouldn't even say ai, some automated system making those changes as opposed to a human being, most likely ai. But I think a question is a question of when, like, is, is it the challenge is not whether it will happen, it'll probably happen. Uh, it's always a question of does it happen within my lifetime?

And it's a very difficult question for me to answer, but I would agree with you that that tension exists. I think some, they'll probably start seeing some companies starting to accept that change and others following over time. Just like with the cloud, everyone is yet not on the cloud, right?

And it's gonna be similar. So last question, short of all those wonderful outcomes in the meantime, might we just see some lowering of the overall amount of stress that SREs and DevOps teams currently ex experience? Because, and we don't talk enough about this, but those teams burn out pretty quickly because there is so many things that can go wrong, you don't know when it's gonna go wrong.

So there's always this constant sense of, you know, is it working? Is it working? Is it working?

Is it again, will AI make? Yeah, And, and will that You were getting into a place of very philosophical questions about human beings. I'll say, uh, I don't think human stress ever really goes down.

Uh, it's a choice we've made as a people, uh, to, uh, however I'll say I think the fundamentals of the technology allow for the stress to go down. I think will the stress go down is dependent on how many more things we make complicated for ourselves in the future. But AI has the capacity to make things simpler in the sense that complex systems, complex actions can now be performed by something else.

However, I'm making no qualifications as to whether the human stress will go down. I think that's a totally different question in some ways. All right.

Hey folks, you heard it here. It's gonna get easier to manage logs and we'll make more sense of them for sure in the age of ai, whether or not we're gonna experience less stress. Well, that's on you.

Hey, Aon, thanks. Thank You. No matter how we define the edge, the special requirements for use in harsh environments drive unique product decisions.

This episode of utilizing Tech brought to you by Soy features Alistair Brad book, founder of Anion discussing Edge Servers with Janice Norski and myself, it pays to start with the outcome to think about the constraints and to build solutions around that. Welcome To utilizing Tech, the podcast about emerging technology from Tech Field Day, part of the Futureum Group. This season is presented by soy and focuses on new technology like AI at the Edge.

I'm your host, Steven fst, organizer of the Tech Field Day event series. And joining me today as my co-host is my old friend, Janice Norski, welcome to the show. Thank you, Steven.

It's a pleasure to be back. Always, Always. It's been a lot of fun.

You know, we've done a couple of seasons here. We've had you on this podcast and the others, and every time you bring in somebody interesting, now the most interesting thing that we're doing this season is we're trying to bring in folks who are actually, um, out there doing the work, implementing, making this stuff happen, you know? Exactly.

And what I'm excited, I'm excited about this series and this episode in particular, because you're right, we do bring in some cool partners. However, it's not too often that you get someone that can talk about the edge in a unique and different way. And, um, I'm super excited about Alistair.

I'm gonna let, I'm gonna give it over to him in just a moment too, to explain exactly what he does and who he is for the company. But Alistair and his team with Anion are building a really unique edge use case that can be used across many different industries. And their vision and their focus about how they get work done and how they work with their partners, I think is incredibly different.

So we're gonna learn a lot about Anion today. We're gonna lot learn a lot about the edge use cases, and then we're gonna find out, you know, where does AI and storage kind of fit into all this? So, Well, welcome to the show.

Alistair, why don't you tell us a little bit about yourself? Hey, yeah, great. Thanks Sam, Steven and Janice for inviting me today.

Yeah, um, my name's Alistair, bro. I'm, uh, the founder of Anion. Uh, been going about 10 years now.

Um, background of Edge, we can talk about what Edge means. It's an interesting concept of what we all mean by edge. But, uh, doing this for 20, 30 years, actually exploring how we collect and manage data and deliver capability to different types of both in the healthcare industry and more recently in defense.

So yeah, that's kind of me and intrigued to see where we end up today talking around those sort of topics. So yeah, let's dive into that, um, because I, I think Edge is a really interesting topic and everybody that we've spoken to, um, within this series and also outsider, different definitions of, of what Edge is. So I don't know, Alistair, can you tell us a little bit about what your viewpoint of Edge is and how do you define it with your company?

Yeah, that's really interesting and probably whatever I define will be different to everybody else. Having, I, everybody has their own view of what Edge is, and probably it probably doesn't really matter to some extent. It probably matters to the company who are designing things and the consumers who are going to use it to some extent.

You just need a natural language that works between you and those customers that you both at least understand or at least can communicate on to us that edge. Well, it could be somebody jumping out of a plane, it could be somebody bo leading a boat and, and walking through water. So my background at the moment is all to do a defense work.

So when we talk about Edge, we actually have called it deployed for many years. So the Edge is more of a commercial construct that's come across more recently. But prior to that, we've been talking about deployed comms within the defense industry for a long time because they have the ability or need to take it with them.

They have to do stuff in these remote locations, and they often don't have any comms. They don't have any power. Uh, they don't have any of the niceties that you get in a data center or sitting at home now where I'm sitting here and I have all of the benefit of a, of a great broadband connection and great it.

They have none of that. So I think that's the challenge for us is that the edge can be right at that, that that tactical space that where effectively what you have with you is all you have, and then you've got phases leading up to that. And, and in the defense world as well as I suppose in, in telecoms and other industries, you've got scaling of people and there's people scale then the types of technologies they need naturally scale with them.

And that eventually, I suppose we say we're kind of not at the edge now, but I don't know if there's a really clear definition of when am I not at the edge. I mean, I'm, are we at the edge now? I mean, all of us are connected to a, a data center somewhere, which I, who knows.

So I think it's an interesting concept of what defines the edge. I'm not sure that or a, does it need to be defined? I don't know.

Maybe for me it does because we build hardware that we say edge hardware. So I suppose I need to define it, but I suppose as a consumer, you just want it to do what it does for the outcome that you are trying to get to at that stage. So maybe it doesn't matter so much, but yeah, it is in, it is something I pose every day, but we ha we use the word edge, but I, I can't define it as well as probably I should do.

Hence I've just muddled my way through that answer to try and explain where I think this edge could be. Um, I suppose, Well, that, that's what I was kind of trying to get at with the, um, in the introduction. I, I'll say this, like edge is as edge does, and, um, the, the reason that it's interesting to talk to you about this is because, because you're not trying to come at it and say like, this is the box, this is where we need to put things into.

You're trying to come at it and say, what, what do you need? What kind of servers do you need? What do they, you know, what, what kind of specifications?

What kind of, uh, configurations And, and, and by kind of coming at it based on the customer demand, rather than coming at it based on preconceived notions of what what it is I think has led to the development of what you've, what you've got. So maybe it would help, do you want to talk a little bit, um, uh, to, to kind of kick things off about the, uh, uh, servers at the edge that you've got? Yeah, lemme, let's go back and explain where we sort of come from, where we are now.

And the reason where we are now, I suppose, is so traditionally in the IT world and my journey with defense in it, in that mix. And we can go back prior to that with, with my work in the pharmaceutical world where we would kind of collect data for phase one, phase two, phase three, phase four trials. Which actually is interesting because the phase one trial is, is essentially a bedside trial.

You, you are in a very small group of patients. First, first dosage of, of a drug to a human is at phase one. You hope at that stage you're hoping you are just going to get some effect, but you don't wanna hurt anybody at that stage.

And then as you go forward with phase four, you are looking at mass trials on a global basis. And you can see the challenges there when it comes to collecting that data. We were looking at this 30 years ago of how do we collect data at each of those stages, but provide them back to the pharmaceuticals or the CROs to allow them effectively to submit that data to the FDA.

So that's where it came from. And then when, when I moved into defense, my first experience of that was actually the first and second Iraq wars, actually more the second Iraq war where we provided the, the comms for the, uh, multinational force, uh, in the south of, of Iraq. And the challenge there was, again, we had this splintering of groups about what type of data and what type of comms they had, both from fixed headquarters to mobile, uh, areas, and obviously a coalition of 44 or so o old countries.

It was, it was a, a very large coalition there. And essentially what we saw at that stage, we were either carrying compact, I think it was compact service in those Dell weren't really even around when we were doing, I think it was compact, that we were either carrying those things on our back, tipping out the dust on the back of a Humvee to try and make them. And generally they did work.

And, and that was an interesting factor is that we realized that that actually commercial comms, even in a desert, in a sandstorm in high heat, you could actually coax its way through. It wasn't always a hundred percent and it wouldn't hit the, uh, it wouldn't hit the metrics that compact or Dell or HP would ask you to, but you could kind of get it to work through. So that was kind of an interesting thing that we learned.

We also learned that portability, and actually we'll probably come back to this topic. We talk a lot in our company about portability and survivability, which those two factors are actually quite interesting because you could just say something's light or heavy. You could say something's big or small, but actually can you, can you carry it?

Can you move it to the location you want with the means of the legis logistics you have at that stage that logistics may be you as a human. The logistics could be a, a C one 30 with a certain size of, of, um, of air you can put it onto. Because if you don't go in that area, they're not taking food or other things they need to take.

So the balance of portability is really important to us. 'cause we need to be able to get our equipment and our capability to those locations where it's, where it's important. So we came back and we looked and we assumed that other people would be doing this.

We weren't a hardware company at this stage. We were actually more into doing software. We were doing loads of stuff with, um, computer associates and BMC and that's where my background sort of was in that middle phase.

But, um, we came back and we realized that actually nobody else was doing this. We were either having very old comms that, that were very, very dated and very hard to use, which was the defense world, or we had the stuff that was designed for data centers and there wasn't really much in the middle. So working with DSGL, which is the, the UK research on for the government and into the MOD, we said, well how about we take commercial kit stuff that say Facebook we're using or starting to use in those days.

How about we take some of this kit and see if we could make it work for defense with, and it becomes a risk balance. The risk is, well, maybe it's not designed to work as long as some of the other kit, but actually the balance of that is it doesn't cost as much and it's easier and it's more powerful so we can do more with it. So we explored these ideas assuming actually that we wouldn't carry on doing hardware for, for very long.

We always assumed that, that we would, either somebody would get better at us or, or, or actually we would just go back and go back into our software services world. But a, we kind of enjoyed it. I think we were okay at doing it.

You know, we, we seem to have forged a a a a different type of approach within the UK and NATO and, and now into the DOD. And it started this vision for us to say, could we put data center technology in a ditch? That kind of was our mantra for a long time.

Could we take, and sometimes actually I have to say to even the people I talk to is that just 'cause we can, doesn't mean we should because yes, I could give you an H 100 in a ditch, but do you need an H 100 in a ditch? Because there are other challenges that come with that. I can give it to you, but you've gotta power it, you've got to call it.

So actually it was interesting how we have to balance what potentially we can achieve versus actually what is logical for what they're trying to do. And that comes back to your point, Stephen, about outcomes. And we call, we talk about outcomes quite a lot because the outcomes should really be the driving force in most IT decisions.

You know, in the end you should be thinking about what you're trying to achieve before you've decided what you've engineered. However, the majority of people in technology are, it are either engineers or people who want to be engineers. So they kind of liked this idea of being involved in describing the physicality and the technical aspects of it.

And they kind of forget about what they were trying to do. 'cause they've designed this amazing when you're like, what were you doing? Oh yeah, I just wanted to send an email.

Why did you design this server then that can send a thousand emails? You just wanted one email. Why did we not design that?

So it's, we we're constantly having to bridge that gap between outcomes and what we can do and make sure that people take the right, the right choices. And that hopefully means that they get the best value for money, the best use, the best portability, the best viable, all of those factors come into a decision. And actually in the commercial world, you don't generally have to make too many of those decisions.

Maybe cost comes into it and performance, but generally, you know, it's a 19 inch rack. It's gonna go in a, in a space that may be changing. You know, I've listened to quite a lot of podcasts recently where we're talking about power consumption.

It was a great one recently with A-A-P-K-I-O and soddy. Um, talking about exactly that factor of, you know, we can't just ignore the amount of energy that it's taking now to do things. I mean, there was a, a quote in that podcast I was listening to which talked about that the, the amount of energy going into just doing some degree of l and m's in the future, it could be more than India could just to do some of these, could be more than what a whole the, that country would.

And that's kind of mind blowing, isn't it? That that, that we're gonna use that much energy to achieve stuff. And I think in the end, hardware will be important.

So things like solazyme will absolutely be important to innovate. We will be important 'cause we need to be able to get the power to it correctly and call it correctly. But I also think the actual engineers who are writing the software are going to be critical because in the end, you know, we, they need to be more efficient and well, we all need to be more efficient, I suppose, in that, in that chain because not one person's gonna solve this.

Um, the question is do we need another LLMI suppose it's a different question, but we are human and we will always want the next best greatest thing. So I think the, the chance of stifling innovation because of climate is unlikely. So it will revolve around us having to innovate to ensure that as we go through that journey, we consume less power, I suppose, and generate less heat, um, going forward.

Um, amazing. I I really love how you connected what I was initially speaking about your, you, your, your company and your vision and how you look at technology differently and kind of put the human factor first, right? Yeah.

Because there's lots of innovation out in tech, right? I've been in it for a long time. Everyone's looking to one up so and so and be the next best thing.

And sometimes you lose sight and you over-engineer and you start selling things that folks don't even really need or you're, you're, you know, oversubscribing on performance or density or power or whatever, right? So I love that you're kind of keeping that people first, human first mentality and designing and optimizing your solution to be what is actually needed. Um, and I think that's why you're great partners with us as well because that's really truly our vision and why we're just sticking to storage.

Um, you mentioned a little bit about, you know, what the people need and what the people want, right? So we're talking a lot about edge, we're talking a lot about ai, that big buzzword and do people really know what they need, right? Everyone's trying to figure out ai, they're trying to figure out the edge.

So, uh, and you brought up power and cooling, you brought up a lot of good stuff, Alistair. So, um, we'd love to get your insight on what you're seeing, particularly with your solution, be it that you guys dabble software and hardware. Where are you seeing AI and, and kind of the needs of, of your customer base and, and how are they integrating all this together?

That's a, that's a great, let's unpick that question 'cause there's, there's a lot in that that, uh, which is really, really interesting and fascinating. So AI in the defense, I mean, we can see it playing out in Ukraine at the moment. The Ukraine war is, is clearly the, the beginning of a new way of doing it.

Drone warfare. But drones are driven a lot by IT systems behind the scenes and the way we consume that data. So actually that is a game changer in terms of the way that, that we're gonna have to think about how that information is both collected.

So sensor fusion is going to become a really big thing. We need to bring all of this information in to mobile, let's call 'em data centers. Really big trucks consuming both video, audio, uh, electromagnetic, uh, human source, other types of, of information needs to all be fused in real time.

And not only does it need to be then taking a model that maybe has been learned on A DGX sitting back somewhere or multiple dgx is no doubt, but it's gonna have to refine that model because actually warfare is never quite as logical as you would expect it to be. Yes, there is quite a lot that you would assume was gonna be part of the model, but you're gonna have to keep refining that model constantly. And I think that, that that's not just the, I mean every industry is gonna have the same challenge.

I suppose the challenge we we have is that we've gotta condense that into, into an object that doesn't have a broadband connection. It's probably got satcom even though it's great now with, you know, starlink is a better than where we ever were before and it will continue to get better, but it's still not directly connected into these cloud ecosystems. So a lot more of what we need has to come with us and we then need to disperse and share that in such a way that we don't have big bandwidth to share that within a battle space.

So we're gonna have to work out ways of how do we share this evolving model from these different sensors in such a way that everybody can learn from that. An example of that would be that you have, maybe you have a, you have a tank, a tank's coming through a from a, from a, from an area of a forest and a drone picks it up and it will take a photo of that potentially it'll take a video of that and then another drone sees that tank. Now if you haven't understood it's the same tank, 'cause the tank may look very similar frankly.

So if you don't know it's the same tank, you now think you've got two tanks. So somehow the AI actually has got to be able to put very small markers and they're using technology around putting it down to the pixel level markers to understand the difference between these objects. And ensure then as that information passes between the different sensor fusions, we know it's only one tank.

Because if suddenly we've got two tanks, three tanks, four tanks, we think we've got a much bigger problem than one tank. So actually that's a really interesting example of where we've got to be able to get better. Now that doesn't particularly need any additional learning.

That type of technology is available, it's just, you know, using vision learning. But there will be additional pieces that need to be generated in real time during that exercise or that battle that you potentially haven't got the time to send back to your big DGX server back, you know, sitting somewhere. 'cause you haven't got, you can't move the data or you haven't got time to make those decisions.

It's real time decisions we need to make. So I think that's interesting on, on that side. I think what we're seeing though, like lots of people, I think most people think they need more GPU than they probably do, but I don't think that's unique to our, I think that's, that's, that's been great marketing from the likes of NVIDIA and people to tell us we all need these things.

And actually I think we've all forgotten about maybe there are other ways the CPU is still in there and sometimes the CPU can be equally useful as A GPU for certain functions, not A GPU doesn't always make it better, you know. And I think actually over time some models are actually aligning themselves to say, well you know what, if you've got a really good CPU, we can take some of those cores and actually you don't now need a GPU to do certain functions. So I think we'd need to educate ourselves as engineers, but also educate then our, our customers downstream to make it clear about when A GPU or A DPU or an MPU or any, any of these additional processing units are valid.

And when actually keeping it simple is the better option. Because the simpler we keep it a it's easier to maintain, it should be cheaper to buy, it's easier to, uh, to, to buy upfront. 'cause you're not on a supply chain.

It includes so many different components that, you know, if you want a GPU, El Musk has probably bought it already and put it into one of his big data centers. So we're not gonna better get that for a long time. And if your system depends on it, you can't then get that and deliver it out quick enough.

So I think that education's gonna be important. And I think actually that the, the software engineers will start writing software that needs less GPU over time for certain functions, certain things we'll absolutely need it. Those core big things that happen, you know, right at the beginning of a, of a, of a model learning.

Absolutely, you know, you're gonna need your your your blackwells and all of that. But I think some things will change. So we're definitely, education is key and I think that's that's key for us as well actually.

'cause we can get sometimes a bit like, oh wow, let is pro GPU at this, it'd be great. Woo, put it, let's just put AI in the title and, and you, and you know, I think we need to also stop doing that a little bit collectively. We are seeing storage, storage is a big thing, hence we partner so well with solid land we storage is getting much more prevalent.

You take that sense of fusion challenge earlier, the amount of information we need to store and, but not only store, we need to go to write it and read it very quickly. Now we've probably pushed our customer far quicker than they had expected. I mean, they've been on, you know, scuzzy Saturn, you know, the fact that we moved into MVM E.

So all of our platforms are now only MVME, um, and generally QLC as well because we want high density, you know, so we want to get as much into our systems as we possibly can at such a high rate. So, you know, PCI four at the moment, five, you know, coming out, we'll, we'll keep moving forward the best we can to mean that when they do hit those challenges, then the stumbling block won't be the hardware that we they've either got or they've bought into our ecosystem. It will be the way the software's written to make sure they take the best they can of that.

But we are seeing a demand for more, more storage in a smaller space. And actually that's always been our challenge, our challenge for many years leading up to the revolution with the E one s, the E one LEE three I suppose, but we're more of an E one, you know, we love the E one. It fits our, our vision so much more than the a standard two and a half type concept.

And we can talk about that, why that is the case. But, um, I mean we can now get with the 120 byte in a half 19 inch system where we do a lot of half 19 inch, the half half of rack size, we can get 10 of those E one Ls in there. So we're over a petabyte of storage in, you know, in that it's, it's, it's, it's amazing what we can now do.

And, and I imagine that we could probably sit here in 12 months time and we'll be even more. I mean, it doesn't seem that there's an end to the amount of, there obviously is an end like Moore's Law when it came to CPUs, but it, it seems that we're on this, we're on this journey that doesn't seem yet to be ending in terms of the amount of storage that, that the likes of heim can now fit into the same physical object. But actually what's more important to me, not only is it physically not changing, which is vital because we want everything to be as small and as portable as possible, but actually you don't increase the amount of power I need to drive it.

In fact, sometimes you give me less power, which is something that you do as well as A and D. So if you look at the, the a and d processes, if you go through looking at the, the 7,000 and the, the second series, the third is the fourth and the fifth actually, when they increase my core count, I actually don't get a big penalty for for, for for power, which is amazing to me because that means essentially I can give my customer downstream something more performant but not then say, just to let you know, now we need a power station wheel behind you to run it. I can say actually you can do that with less power and I'm gonna give you more cause.

So yeah, to us the A MD soine marriage where we use them is amazing because we tend to be able to go forward with capability but either stand still with power or even sometimes even go back and power to us is heat and we need to get rid of that heat. You know, we, we do air cool systems like the traditional ones. We do them in a slightly different way, hence we do a lot of half 19 inch even in the air called fabric.

So you can build interesting stuff, but we also do a lot where we, we have no air calling, so it's all conduction cord or liquid cord or motion calling technologies. Um, and that's when we have a big challenge because we have gotta get rid of that heat. If, you know, the movement of heat through those mediums is, is more challenging than it is if you just put a load of fans over it.

You know, even though, even though fans are not efficient, air is not an efficient medium for conducting heat. You can still get away with just throwing a lot of fans at it and, and eventually the heat will go out. I mean assuming the ambients alright, but when you literally condense it down, so we've been doing conduction cooling with solid D drive ever since the E one first came out within the first week of us getting it.

The first thing we did was rip the heat sink off it, put it inside one of our conduction cool systems. Um, and in and in the defense world, there's a, there's a concept where you can pull the heat out through, uh, effectively called V-A-V-P-X system is where you, you effectively use, um, uh, uh, a way to clamp it in the copper pulls the heat out and then that then is where the heat goes out. And, and we've used that method in our system, in our, uh, conduction systems for five years now, six years now.

Um, with amazing effect. I mean it, it'd be nice not to have to rip the whole thing to pieces, but it's the way it works for us. And, and it means we can get in something about this, you know, about centimeter high.

We can get, you know, six discs in that, in that fabric. Um, and they can pull it out and they can do what they want with it then and, and it becomes effective. It's a bit like having an nin 10 day cartridge.

They have a cartridge of their information, which then if they need to run very quickly, they can pull out the data and they can run, if they need to transport that then, and it's secret, they can transport in certain ways that is easier than if it's 50 discs that are just thrown in a, in a, on the floor. And you've then gotta work out, well, which disc went with which disc, you know, how am I gonna reboot my system with if I dunno. We deal with a lot of that challenges, uh, and the E one s has been, has been the enabler.

We were using M twos before, we never really liked the m twos. It, we liked the, we liked the, the, for the mechanical format, but hated the performance and the general, they were consumer really not really enterprise. E one obviously gave us the best of both worlds, fully enterprise and in a format that meant we could do the same job and E one L.

Yeah, I mean that's just amazing, isn't it? I mean the amount of storage on E one LI know the ruler for many years from Intel from previous years, you know, never quite made it. And I still don't know if the E one will become as prevalent as, um, as other 'cause it, it is a challenge because it's 300 and something deep, so it makes servers bigger and so forth.

We do it differently though. We don't see a, we don't see a server the inside of a server as a one dimensional space. Most, most companies steer it as, as a single, even in a two U system.

It's just effectively the same plane but just higher objects. We actually see it as a full three dimensional space. You know, we're manipulating objects at different planes and the different levels and a different locations to see how much can we cram into the smallest space.

The density is another phrase we use a lot of. We're we're trying to increase the density of everything in our systems. And that's not just storage with soddy, but it's, it's, it, it's everything.

It's the amount of power we could get to it. It's the amount of cause we can get the amount of, uh, of memory. So density is, is critical.

The other interesting thing, we we've we've, we've been playing around with the concept of disaggregated systems for six, seven years plus. Uh, it is become more of a big thing now, unfortunately, from our point of view, when we liked the idea, A, we weren't clever enough to do the inventions that needed to happen. We we're, you know, we're, we're good at our stuff, but we're nowhere near as clever as these amazing people who can do this, these inventions.

So we had the ideas but couldn't deliver on it because the things didn't exist. We worked with Fujitsu n EEC for a while on using light to move PCIs. We did a really great piece of work where we were using, and SAM Tech had a, had a vision of using light to move PCIE so we could disaggregate the PCIE bus remotely.

And what that meant is that we weren't limited then by the density of a single object. We could increase our density across multiple objects and disaggregate the functions we needed. That's become a lot easier now because the likes of CXL has come along.

I mean, I say it's easier, CXL is still quite complicated. 1. I don't think it's really become that proliferated as much as people wanted to.

I think two is starting to probably deliver what most people have wanted it to. But for us, the ability to disaggregate say memory, to increase the memory footprint, but without increasing the, the physicality of the single object and saying to a customer, if you do need more memory, yes there is a, there is an overhead of of a bigger subsystem, but you can make that choice when you need it. You don't have to compromise at the beginning.

So the reason we did this was because we didn't want people to have to say, well my object needs to be this big because I might want a GPU and I may want more memory and I may want this. We wanted to say, well if you don't know you need it, let's make it as small as possible and let's allow you to add those capabilities to only those systems at the moment you need it to deliver it. And that, that's really vital for our world because if they don't have to take it, why give it to them?

'cause if they take it, they've then gotta power it. They haven't gotta maintain it, they haven't even gotta look after it. They then also have to potentially buy the same systems with all of this expensive capability upfront when actually they may only need 10% of their fleet to be able to be upgraded to that level.

Now it probably sound a bit stupid, we should just be saying, Hey, buy the most expensive system that kind of isn't us. We would much rather they bought the right system and had the ability to grow or even contract when they needed to. Because actually it's not like in a data center where you can just throw it at it.

They literally have to sometimes put it in their backpack and carry it. And if they don't have to carry it, why would I give it to them to force them to have to go down that route? So storage will be big for us there as well.

I mean, I know storage disaggregation has been possible for a long time en VME of fabrics. And, but again, I think with CXL it becomes an interesting approach of a, of a, of a standardized approach for us. That disaggregation of everything on that sort of PCIE five and six bus we can start doing.

For us it would be memory and storage would be the two big things with maybe GP as well moving those. So yeah, sorry, very long answer to a question you asked earlier, but anyway, You covered quite a lot of ground there actually, uh, um, I will, will point out that season four of utilizing tech focused on CXL and one of the things that we focused on in there is exactly what you're talking about. It wasn't about necessarily disaggregation as the goal, it was about right sizing as the goal and about flexibility and making systems that were not bound by the strict structures of the size of a dim or, you know, it's about making things right.

Yeah. And, and to me, I think that's really kind of the summary of this whole conversation. It's, it's, it's interesting isn't it, that without constraints people tend to just expand like crazy, but with constraints.

And that's what makes the edge interesting to me is that it's a constrained system. You cannot go beyond this physical size or you can't go beyond this power footprint or you can't go beyond this cooling or, you know, we have to make sure that it's rugged or we have to make sure that it's able to, you know, to, to handle disrupted communications. Oh, and disrupted communications means we have to do processing locally and that means we have to and, and, and there's this whole chain of thought that happens when you are constrained.

That doesn't happen when you're in sort of a data center or a cloud environment where you can have as much of anything at any time within reason. I mean, but, but even now, I mean, look at ai, it's not even within reason. It's just, you know, you can have as much of whatever you want, you know?

Yeah. When it comes to the edge, and especially in the applications you're describing, these constraints make us more creative. They make us come up with clever, uh, novel uses for technology that we already have, like the CPUs that we already, that we already have and, and, and new technology as it comes out.

It makes us see new possibilities. And, and that's what's inspiring I think about this whole story is that, you know, we can do more with less if only we just tried. Would, would you agree with me, Ben?

I Would, yeah. I think that that idea of constraint is amazing. I I love that actually, that phraseology you've come up with actually.

And I think that's what, what, what drives me actually. I think that the idea of not having unlimited options and, and having some constraints actually makes it more, more interesting. I dunno why.

Maybe that's a, that's a part of my psychology maybe, I don't know. But that idea of having a, having restrictions and having to work around them is fascinating. And if you think about where potentially we want to go as a, as a as a world, I mean, not, not within my lifetime obviously, or or any of us probably, but you know, when we look to go to to Mars or other places you think these challenges we've just spoken about, those constraints are absolutely gonna be in there when we start going into planetary, then those challenges are gonna be there.

Even on this world, we are gonna hit these challenges. We know that. Not maybe within the lifetimes of any of us on this call, but there will always be, I think there're gonna be more constraints beside it.

We may, I don't know where the peak of unconstrain ability is in this world of ai, but at some point we will, we will flip over it and, and we will be constrained by the amount of power and and so forth. And I think, um, yeah, I think edge, I think a lot of other industries could learn from edge. Even those ones that at the moment are unconstrained, I think they could learn.

Yeah, I couldn't agree more. Uh, Alistair, I feel like, um, and Steven, you said this too, like, is the industry really trying hard enough, right? Have we really looked at, you know, cooling storage?

Um, we're starting to, we've got some new designs that we just placed out onto the market and everyone's like, whoa, what is that? That's cold plate. Wait, it's cooled on all four sides, not just with the liquid tube.

Wait, it's not dunk dunked in anything. It's not diabolic cooling. What is it?

Right? So it's like, I mean, to your point, Alistair, I think your company and your vision and what you're doing is so ahead of itself, right ahead of the time because you aren't just looking at cooling, the GPU, the CPU, you are looking at the overall architecture. You're looking, how do I make this more seamless, portable, simple.

Um, and again, that human, that human first focus, which I personally, um, really, really appreciate. So, uh, I can't wait to learn more. Um, I feel like we could, we could talk all day about this, um, and, and still be like really interesting.

So I actually think we should, we should have you back on at a later date, Alistair, and, and talk more about, you know, the stuff you're doing with PKIO and some of the other organizations, right? Because we talked a lot about military today and we talked a little bit about, you know, the, the other use cases you, you guys work in. But, um, it would be interesting to give our audience an understanding of how you're doing this differently for others.

So this has been phenomenal. I really appreciate your insight. Good.

It's been a great chat. I've really enjoyed it. Really, really good.

Thank you very much. Yeah, thank, thanks. It's, and, and I have to say too, the, this whole story of constraints and, and, and, and incredible, uh, challenges and, and, and how people meet, meet challenges at the edge.

I mean, this is what we've been talking about at Edgefield Day, um, you know, the whole time through. Um, I, I just love this discussion. We've talked to a bunch of companies in this space that are doing things like this that are kind of rising to the challenges that are placed on them.

Um, it's, it's, it's, it's incredible. So, uh, thank you very much, Alistair, for your time. Thank you so much for joining us.

Before we go, um, I I'm sure that people are gonna wanna continue the conversation with you. Uh, where can they connect with you? They can, they can find me on LinkedIn.

I'm not, I'm, I'm not very good with the whole social thing. Other people are, I'm not, I'm, I'm probably from that generation. I didn't quite embrace it enough as I should have.

Um, but LinkedIn, you can find me there. com and, and reach out to us. Uh, obviously I'm doing a few podcasts with, um, with Solid Dam at the moment, which is Alien to me doing theses, but I'm enjoying them.

Actually, I'm not really into the self-promotion discussing thing, but I'm, I, I love these types of ideas of just talking, which is great. So, yeah, you know, there are ways to get hold of me, but I'm not prevalent on the Twitters and the LinkedIn, you know, or the other play. Well, yeah, I'm not great on those areas, but you can get hold of me.

Well, you, you wouldn't know it. Um, uh, excellent, excellent conversation. Um, and, and Janice, uh, you know, thanks for joining us here again.

Um, what's New with Soy. Thank you for having me again, Steven, this is always fun. com/ai and you might see another sneak peek of what we're doing with Anion there as well.

Excellent. I would love to see that. Um, and as for me, uh, you know, we've been really enjoying this season of, uh, utilizing Tech.

Uh, you will find us, uh, show notes, more episodes and so on. com. Uh, you'll also find us in your favorite podcast applications.

Uh, you'll find us on YouTube. Um, we would love it if you would leave us a comment, leave us a rating, leave us a review. Uh, it really helps us to, uh, direct where we go with this in the future.

Um, this PO podcast, as we noted, was brought to you by Soy as well as Tech Field Day, which is part of the Futurum Group. Um, we can also be found on, uh, the socials. Uh, find us on X Twitter, blue sky mastodon at utilizing Tech.

Thanks for joining. Uh, we've got a new episode coming next week. Uh, listen to the whole season.

Uh, we'd love to to hear from you. Thanks for joining us, and we'll see you next week.