Latest / The Adults In The AI Room / Trent Fitz on Mastering AI and Data Architecture in IT Organizations - The Earley AI Podcast with Seth Earley - Episode #042

Transcript

Trent Fitz on Mastering AI and Data Architecture in IT Organizations - The Earley AI Podcast with Seth Earley - Episode #042 · The Adults In The AI Room

• 0:00
  SPEAKER_00: This is the Early AI Podcast, hosted by Seth Early and Chris Featherstone.
• 0:05
  SPEAKER_00: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible with artificial intelligence.
• 0:14
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science, your digital transformation journey with design and deployment of innovative technology solutions, as well as CMS Wire.
• 0:24
  SPEAKER_00: Now, enjoy the show.
• 0:30
  SPEAKER_02: Terrific.
• 0:30
  SPEAKER_02: So welcome to today's podcast, the Early AI Podcast.
• 0:35
  SPEAKER_02: As you are uh aware, um we are uh we have a great guest today.
• 0:41
  SPEAKER_02: And uh my name is Seth Early, by the way.
• 0:44
  SPEAKER_02: I'm here with uh Liam Kunik.
• 0:46
  SPEAKER_02: Say hello, Liam.
• 0:47
  SPEAKER_00: How's it going?
• 0:48
  SPEAKER_00: Co-hosting today.
• 0:50
  SPEAKER_02: Yeah, Liam is uh newer to the organization, recent uh graduate in uh uh digital marketing, and uh he's been doing a lot of behind-the-scenes work, so he's helping out in front of the camera today.
• 1:02
  SPEAKER_02: So great, great to have you.
• 1:03
  SPEAKER_02: Um and we are very excited to have our guest today, who uh and we're gonna be discussing things like the evolution of IT infrastructure monitoring in the context of AI and uh machine learning operations.
• 1:18
  SPEAKER_02: Uh, we'll talk about the importance of AI uh ops maturity uh and the data collection behind uh AI ops.
• 1:28
  SPEAKER_02: We'll talk about uh AI and machine learning in the IT organization and how we need to think about topology and awareness of uh the tools and the technologies.
• 1:38
  SPEAKER_02: We'll talk about some of the drawbacks of application monitoring in and of itself, and look at a bigger context of AI ops machine learning uh uh maturity.
• 1:48
  SPEAKER_02: Uh so our guest today is a uh digital expert, a digital guru with 20 years of experience in the tech industry.
• 1:56
  SPEAKER_02: Uh, he's an expert in global marketing, product strategy, business development, cloud computing, cybersecurity, and AI.
• 2:02
  SPEAKER_02: Uh, he has led uh projects at companies such as IBM, Cellpoint, TrustWave, and various startups.
• 2:10
  SPEAKER_02: He's currently the uh product strategy and technology marketing leader at Xenos.
• 2:15
  SPEAKER_02: And uh Trent Fitz, welcome to our show.
• 2:18
  SPEAKER_02: Thank you so much, Seth.
• 2:20
  SPEAKER_02: I'm happy to be here and I appreciate the time.
• 2:22
  SPEAKER_02: We're just talking about the fact that you're in Austin, as is Liam.
• 2:26
  SPEAKER_02: I'm in Boston.
• 2:27
  SPEAKER_02: Uh, what's your weather like today?
• 2:29
  SPEAKER_01: Uh it's about 20 degrees with the windshield.
• 2:31
  SPEAKER_01: It's nine.
• 2:33
  SPEAKER_02: Wow.
• 2:33
  SPEAKER_02: Okay.
• 2:34
  SPEAKER_02: So I didn't expect that.
• 2:35
  SPEAKER_02: So we're we're similar, but uh we have a lot of snow.
• 2:39
  SPEAKER_02: Uh, but it's expected in this this uh neck in the woods.
• 2:42
  SPEAKER_02: So one of the things I'd like to start uh the show with is kind of your perspective on misconceptions around AI.
• 2:48
  SPEAKER_02: What's fact?
• 2:49
  SPEAKER_02: What's fiction?
• 2:49
  SPEAKER_02: What's the future of um of uh applications?
• 2:52
  SPEAKER_02: And you know, what do you come across most frequently in your travels when you talk to executives and customers where they're just not quite getting it right?
• 3:00
  SPEAKER_02: I I almost I know when we were talking before, I almost framed it as what are they getting right versus what are they not getting right, but you can frame it uh either way.
• 3:09
  SPEAKER_01: Yeah, so uh we're still at a pretty nascent stage with AI.
• 3:13
  SPEAKER_01: I mean, it's it's been around for you know 50 years, but um it obviously because of uh Chat GPT and large language models and and the kind of stir they've caused, uh, it's generated a lot more interest recently.
• 3:29
  SPEAKER_01: It it has become a more tangible uh you know uh application for business use, uh, but there's still so much confusion.
• 3:39
  SPEAKER_01: Um mostly what people know is chat GPT uh or the you know corollary Google or or Microsoft uh options for that.
• 3:49
  SPEAKER_01: Uh obviously open AI is the topic of a lot of conversation.
• 3:53
  SPEAKER_01: But uh I think the the general sentiment right now, rather than you know, trying to say who's getting it right or wrong, I think people are feeling their way around.
• 4:02
  SPEAKER_01: They feel like they're going to miss the boat if they don't get projects going, if they don't start investigating how it can help their businesses.
• 4:10
  SPEAKER_01: Um, at the same time, they're they're really just trying to figure out what's what.
• 4:14
  SPEAKER_01: And there are so many terms, you know, around artificial intelligence, even drawing a distinction between artificial intelligence and machine learning, and where do those fit in with natural language processing and uh large language models, et cetera.
• 4:27
  SPEAKER_01: People are still trying to just figure out how all these parts and pieces work together.
• 4:32
  SPEAKER_01: And and right now it's a honestly, it's a huge opportunity for vendors to kind of use smoke and mirrors to sell anything.
• 4:41
  SPEAKER_01: You know, it's a it's AI washing.
• 4:43
  SPEAKER_01: Everything is AI these days just because it's something that people are interested in investigating.
• 4:48
  SPEAKER_01: So we're just at such an early stage.
• 4:51
  SPEAKER_01: There are companies doing real things with it, obviously, but um, but for the masses, we're there, people are still finding their way around.
• 4:58
  SPEAKER_02: I like that term AI washing.
• 5:01
  SPEAKER_02: You know, I used to uh write that uh back in the early days of AI and machine learning, um, by definition, AI didn't work because as soon as it worked, they called it something else.
• 5:13
  SPEAKER_02: And I forgot who coined that phrase, uh, but um uh so word processing was an early iteration of AI.
• 5:19
  SPEAKER_02: We don't call it AI now, we call it word processing, right?
• 5:23
  SPEAKER_02: So so but now everything is AI, right?
• 5:26
  SPEAKER_02: So anything like algorithm and anything that a uh a vendor puts forward is is AI.
• 5:32
  SPEAKER_02: So so where do you see the challenges in getting education uh uh through to you know the right education and and awareness through to executives and decision makers and even line of business managers?
• 5:44
  SPEAKER_02: So you know, this because there's so much confusion and noise, you know, what do you what do you see as kind of the challenges and opportunities to to um get people on a good uh footing?
• 5:57
  SPEAKER_01: I mean, it this is a common challenge for I'll just call it new technologies.
• 6:02
  SPEAKER_01: Um, you know, back 15, 20 years ago, when people talked about cloud computing, there were plenty of naysayers at that time.
• 6:10
  SPEAKER_01: This is, you know, this is a fad, this is gonna go away.
• 6:13
  SPEAKER_01: And obviously in hindsight, it didn't.
• 6:15
  SPEAKER_01: I think that artificial intelligence, you know, when you go look at things like like mobile devices, look at cloud computing, artificial intelligence, I think, has the potential to be bigger than than any of the kind of revolutionary technologies that we have.
• 6:31
  SPEAKER_01: Um, and so I think that um it's it's right now it's a matter of of educating customers.
• 6:39
  SPEAKER_01: And this is a common process when you have a new area.
• 6:42
  SPEAKER_01: Um, so when when you know I've worked for a number of startups, and and when you decide you're gonna go build something, you have to decide am I gonna attach myself to this category that people understand and recognize, and I'm gonna build a better thing in that category, or I'm creating a new category.
• 6:59
  SPEAKER_01: And in this case, it's kind of all a new category.
• 7:02
  SPEAKER_01: AI has crept into things that people do understand.
• 7:06
  SPEAKER_01: They don't understand necessarily what AI is doing differently.
• 7:09
  SPEAKER_01: Um, but there's a a lot of education, and this is a natural part of the evolution when you have a new space like this.
• 7:16
  SPEAKER_00: Actually, doing a class project last semester.
• 7:18
  SPEAKER_00: Um, and one of the topics that I was doing was presenting kind of like a watered-down version of what AI was, it was for like a general IT class.
• 7:26
  SPEAKER_00: And one of the questions I asked everybody in the class was, okay, um, it was like this kind of interactive form thing.
• 7:34
  SPEAKER_00: I said, Okay, what's the first word that comes to your head when you think of AI?
• 7:38
  SPEAKER_00: And or what type of uh, I guess, AI program do you use?
• 7:42
  SPEAKER_00: And everyone was just chat GPT, chat GPT, chat GPT, chat GPT, chat GPT.
• 7:47
  SPEAKER_00: And one thing that surprised a lot of people that got a lot of reactions was when I brought up how the um Google kind of I'm not a robot captcha thing, um, when you basically will be select images of a motorcycle in um a variety of images, and I was telling them, well, hey guys, I don't know if you knew this, but you're actually training AI models, like computer vision models.
• 8:10
  SPEAKER_00: Um, I think that's the right term.
• 8:13
  SPEAKER_00: Um, but I think that's kind of something that's really interesting.
• 8:16
  SPEAKER_00: Um you hit it on the head when you said, Everyone knows about Chat GBT, but once AI becomes integrated and it's something that people don't even understand is right beneath them or under their fingertips in certain situations.
• 8:27
  SPEAKER_00: Um, the education aspect I think is really important, even like a student level or younger people that are, you know, growing with this emerging technology.
• 8:37
  SPEAKER_01: Yeah, and I I don't want to gloss over the fact that there are um companies with online marketplaces that are absolutely leveraging AI in fairly mature ways to understand how to sell more things to the people that are buying from their marketplaces.
• 8:52
  SPEAKER_01: And I won't name them, but you you guys know the the major online marketplaces.
• 8:56
  SPEAKER_01: Um, this is something that's been in the works for for years and that there's some maturity there.
• 9:02
  SPEAKER_01: The the buyers that we sell to are in uh infrastructure and operations, uh, and these companies that their whole business model did not originally begin, didn't begin as uh you know technology.
• 9:21
  SPEAKER_01: They weren't technology companies, they were forced to become technology companies because pretty much every company does now to be to remain relevant.
• 9:29
  SPEAKER_01: You have to be online, you have to be able to interact uh digitally, you have to be able to sell your products digitally.
• 9:35
  SPEAKER_01: Uh, and so so as a result, there are some large companies, mostly large companies who are pretty sophisticated with AI, but you know, I'd say 98% of companies out there are still just trying to figure it out.
• 9:48
  SPEAKER_01: And and they have the same questions that those college students might have.
• 9:54
  SPEAKER_02: Yeah.
• 9:55
  SPEAKER_02: So so when you think about um some of the things that are really missing uh in when you talk to those IT organizations, do they appreciate I mean most most of the time when I talk to IT organizations, it is about the algorithm and less about the data and the content.
• 10:16
  SPEAKER_02: And you know, there's also this, you know, when I look at Chat GPT or any generative AI, and people listen to the podcast know this.
• 10:24
  SPEAKER_02: I think of two different ways.
• 10:25
  SPEAKER_02: One is personal productivity, where you can rely on the large language model to give you answers and to help with your work, you know, help you with an outline, edit, you know, a document, uh, create some new content, uh, et cetera.
• 10:39
  SPEAKER_02: We still have to give it the right prompt, and we still have to have our uh uh judgment, our human judgment applied, and we have to make sure it's factually correct.
• 10:49
  SPEAKER_02: But that's kind of using the large language model for its capability of making predictions around what language and what terms and what words should come after a prompt.
• 10:59
  SPEAKER_02: But the other way to look at this is to say, let's not rely on the large language model, let's rely on corporate data sources, the corporate source of truth, and use the large language model to interpret the query and then run that query against the corporate knowledge and then use the large language model to process the result response to make it more conversational.
• 11:24
  SPEAKER_02: Um, when you talk to your targets, your folks, are they even thinking about uh that distinction?
• 11:30
  SPEAKER_02: And are they thinking about uh a source of ground truth for uh for uh generative AI?
• 11:37
  SPEAKER_01: I think they're just starting to think about this.
• 11:39
  SPEAKER_01: Excuse me.
• 11:43
  SPEAKER_01: They're just starting to understand that you know they're so so RSpace is something called AI ops, which is artificial intelligence for IT operations.
• 11:54
  SPEAKER_01: This started around 2015, 2016, and it was a term that was coined by the industry analyst firm Gardner.
• 12:04
  SPEAKER_01: Um we're only now getting to a point, so some seven, eight years later, where customers are understanding uh that there are more aspects of this than the algorithms.
• 12:20
  SPEAKER_01: In the beginning, the the you know, the geeky people got really into the algorithms and understanding this, and and you know, they they liked having that knowledge and they get attached to the algorithms.
• 12:32
  SPEAKER_01: And so when when you were talking to customers, they were asking whether you use this algorithm or that algorithm, and they weren't talking about business outcomes.
• 12:41
  SPEAKER_01: And so, so this this is part of the educational you know stage of this, is that you have to start with where's the data?
• 12:50
  SPEAKER_01: Where's the data coming from?
• 12:52
  SPEAKER_01: And and most uh IT you know organizations, infrastructure and operations uh practitioners, they know that they have a lot of data.
• 13:02
  SPEAKER_01: They don't realize that just data laying around everywhere is not doesn't mean it's suitable for feeding AI algorithms.
• 13:09
  SPEAKER_01: Uh, and so there it's a it's a very common problem.
• 13:13
  SPEAKER_01: Most organizations currently they've got data in silos around the organization.
• 13:18
  SPEAKER_01: There's there's little uh concept of data architecture or the need for pre-processing data for artificial intelligence.
• 13:29
  SPEAKER_01: And so, really, the reason we're here, I I said six or seven years later, but I realize it's 2024 now.
• 13:35
  SPEAKER_01: And so really it's eight or nine years later.
• 13:39
  SPEAKER_01: We're here at this time because it took this long for people to understand this and realize that there's more to it than just what algorithms are you using.
• 13:49
  SPEAKER_01: And and it's a it's been a long growing process.
• 13:54
  SPEAKER_02: And what are some of the elements?
• 13:55
  SPEAKER_02: So when you talk about uh AI, ops, and um, and uh, do you make any distinctions?
• 14:02
  SPEAKER_02: AI, machine learning, ops, uh would you make a distinction, or is that basically the category?
• 14:08
  SPEAKER_01: So it it really the term was a misnomer.
• 14:11
  SPEAKER_01: Um, you know, the Gartner analysts that that I speak with regularly, you know, from the beginning have said this was this is mildly interesting.
• 14:21
  SPEAKER_01: When they created the the acronym AIOps, it did not stand for artificial intelligence, it stood for algorithmic IT operations.
• 14:32
  SPEAKER_01: But you you can't have an acronym that starts with AI and expect people to not assume that it's artificial intelligence.
• 14:39
  SPEAKER_01: So eventually they acquiesced and said, okay, it's artificial intelligence for IT operations, but really it's machine learning.
• 14:46
  SPEAKER_01: And and machine learning is a subset of artificial intelligence.
• 14:50
  SPEAKER_01: AI is like the Uber category that encompasses everything.
• 14:54
  SPEAKER_01: So all of the tools in this space, it is really machine learning that manifests itself in several different ways, uh, including now starting to do natural language processing and and large language models.
• 15:12
  SPEAKER_00: So would you say it's just as important to kind of educate people on the um importance of, I guess, organized and properly structured data as well as you know just the capabilities of AI.
• 15:24
  SPEAKER_00: I feel like a lot of companies right now they kind of just will say, Oh, I saw this really cool thing online.
• 15:30
  SPEAKER_00: I just want to implement it.
• 15:32
  SPEAKER_00: Let's uh let's throw money at this team.
• 15:34
  SPEAKER_00: And then they kind of realize it becomes a little bit more of a, I guess, a bunch of tangled knots than they really expect it to.
• 15:44
  SPEAKER_01: It is.
• 15:45
  SPEAKER_01: And and so, you know, along this journey, there's the concept of data architecture, but then you have data scientists who understand the requirements for artificial intelligence.
• 15:58
  SPEAKER_01: They have to work with data architecture people to get the right outcomes that they need to be successful with AI.
• 16:05
  SPEAKER_01: This is a big process.
• 16:07
  SPEAKER_01: And so, you know, vendors like ours, like mine, we we help customers with this, but the very first step is educating them on uh this is not a widget that you buy online and put it in your environment, and all of a sudden it's you know, reliably producing insights.
• 16:24
  SPEAKER_01: So it's it's it's uh a process.
• 16:28
  SPEAKER_02: Talk a little bit about uh the elements.
• 16:31
  SPEAKER_02: When we uh initially spoke, I talked about some of the uh maturity models that we work on, and we were working on one for um AI ops, machine learning ops, and various dimensions of that, which included things like governance and decision making, uh, whether you know processes were were instrumented to determine outcomes and understand ROI, uh the data and information architecture.
• 16:56
  SPEAKER_02: What are the other elements that that you guys look at when you're thinking about uh AI ops and your uh approach to getting a handle on what's going on in the organization?
• 17:10
  SPEAKER_02: Maybe talk about kind of the current state of what you walk in and see, and then what are the different elements that you try to uh tease apart in order to get a handle on uh on uh AI machine learning operations?
• 17:25
  SPEAKER_01: So just at the highest level, there are really three categories of things.
• 17:29
  SPEAKER_01: The first one is this data collection and kind of preparation, we'll call it.
• 17:36
  SPEAKER_01: Uh some sometimes it's just called data cleaning, sometimes it's data architecture, but we'll just call it data preparation.
• 17:42
  SPEAKER_01: The second part is the analytics, it's the the algorithms that you use and what insights you're trying to produce using the machine learning algorithms.
• 17:53
  SPEAKER_01: And then the third thing is what are you doing with that data?
• 17:57
  SPEAKER_01: And so this is where there are key advancements being made in automation, which is another element of AI ops.
• 18:06
  SPEAKER_01: And so, so that's the really the maturity model that we teach is let's figure out where your data is, let's figure out what your data looks like, how we needed to get it in the right state to be leveraged properly, and then let's apply the appropriate algorithms for for the outcomes that that you desire.
• 18:24
  SPEAKER_01: And then how do we leverage that data to be as efficient as possible in achieving the requirements that that you have?
• 18:33
  SPEAKER_02: Could you walk through kind of a specific, a more specific example of you know what an environment might look like and uh kind of walk through what those different uh pieces might look like if you have some uh examples that come to mind?
• 18:46
  SPEAKER_01: Yeah, absolutely.
• 18:47
  SPEAKER_01: So uh organ, you know, we we usually sell to large organizations, uh, but but what what large organization means these days is is dramatically different from what it used to be, because even small and medium-sized organizations can have a large uh IT operations estate.
• 19:06
  SPEAKER_01: Everything is digital now.
• 19:08
  SPEAKER_01: And so uh when in the space we're talking about, which let me just back up and say that originally what Xeno started doing is IT infrastructure monitoring.
• 19:20
  SPEAKER_01: So it's just monitoring your servers and your networking gear.
• 19:24
  SPEAKER_01: And it was a pretty discrete set of things that that needed to be monitored.
• 19:29
  SPEAKER_01: And importantly, they were largely static, they weren't changing day to day.
• 19:36
  SPEAKER_01: So so cut to today, what an IT environment looks like is this crazy haystack that is is very difficult for IT operations staff to visualize.
• 19:47
  SPEAKER_01: It used to be you could, you know, there was an application called Visio.
• 19:51
  SPEAKER_01: You get in and you map out your your network infrastructure, and that's what it was.
• 19:56
  SPEAKER_01: And unless something broke or they expanded or some unusual circumstance, that's what it looked like all the time.
• 20:03
  SPEAKER_01: You could see it in one picture.
• 20:05
  SPEAKER_01: So so then you know, come now we're talking about not just networking systems, but servers and storage and load balancers and and even security gear, etc.
• 20:15
  SPEAKER_01: And so the number of things that need to be monitored has dramatically increased, but also with the advent of initially it was server virtualization, but then there is um you know software-defined networking, software-defined storage, then cut to containers and microservices like Kubernetes, uh, serverless uh technologies.
• 20:41
  SPEAKER_01: These things are coming and going all the time.
• 20:44
  SPEAKER_01: There, it's the whole idea is that they're on demand.
• 20:47
  SPEAKER_01: When some piece of of work needs to be done, you spin up the resources that need to make it happen and then you turn them off.
• 20:56
  SPEAKER_01: This is incredibly difficult to not only visualize, but to manage.
• 21:00
  SPEAKER_01: And so this is the the huge turning point for you know this need for something like AI ops, because human beings can't simply draw it on a sheet of paper, and now I can see what to check when something goes wrong.
• 21:12
  SPEAKER_01: And so that's the beginning of this use case is there are so many disparate types of of infrastructure.
• 21:20
  SPEAKER_01: Uh, and and the monitoring strategies that are currently in place for most organizations are something that's just organically happened over time, where we bought these networking systems and we we bought a monitoring tool to monitor those.
• 21:39
  SPEAKER_01: We have these storage arrays, we have monitoring tools to monitor those.
• 21:43
  SPEAKER_01: Most organizations will have 30 or 40 monitoring tools.
• 21:47
  SPEAKER_01: And so that means the data they're collecting is in their own sort of processed format.
• 21:53
  SPEAKER_01: It is in silos, so it cannot be analyzed together.
• 21:57
  SPEAKER_01: And so that's really step one of this journey.
• 22:00
  SPEAKER_01: Is how do we get this data, get it normalized and into one uh data lake, basically, so that now all of this can be analyzed together.
• 22:11
  SPEAKER_01: And to give a tangible kind of use case here, it's hey, this application isn't working.
• 22:17
  SPEAKER_01: And if you use it, you know, like an online retailer or a marketplace, a minute of downtime can cost millions of dollars.
• 22:24
  SPEAKER_01: And so this application stopped working.
• 22:27
  SPEAKER_01: What's the problem?
• 22:28
  SPEAKER_01: And most organizations are not good at figuring that out.
• 22:33
  SPEAKER_01: It is still, well, let's get the networking team in with the server team and the storage team, and all of these different teams come together in a room and they say, Well, it's not us.
• 22:42
  SPEAKER_01: Well, it's not us.
• 22:43
  SPEAKER_01: And so it's it's the mean time to innocence.
• 22:45
  SPEAKER_01: This is the joke that we use.
• 22:48
  SPEAKER_01: Okay, so mean time to resolve an issue is this MTTR is the standard kind of uh measurement.
• 22:54
  SPEAKER_01: Uh, MTTI is the joke that we use.
• 22:57
  SPEAKER_01: Um, but so so this is the process, and and companies are even even very mature companies are terribly inefficient at root causing issues because it's really difficult.
• 23:08
  SPEAKER_01: Uh, and so as you go from how do we collect this data in one context, the next step is even for AI to be you know really helpful, you have to take your IT services.
• 23:24
  SPEAKER_01: So there is that application.
• 23:26
  SPEAKER_01: We don't know what what authentication systems, what databases, what container, you know, it could be a containerized application, what container orchestration tools are you using?
• 23:37
  SPEAKER_01: It could be the the whole set of things that are just delivering that one application.
• 23:42
  SPEAKER_01: Some of it could be in the cloud, some of it could be in your data center on-prem.
• 23:46
  SPEAKER_01: Uh, there could be hundreds or thousands of components of systems that are delivering that one application.
• 23:53
  SPEAKER_01: So now when that application is degraded or fails, it's extremely difficult to figure out what the problem is unless you have the ability to dynamically map out these IT systems.
• 24:06
  SPEAKER_01: And when I say that, I mean dynamically, again, in the old days, you just know this application is running on this server, it's connected to this storage array, it's connected to this network switch.
• 24:19
  SPEAKER_01: It was easily mapped out.
• 24:21
  SPEAKER_01: When there was a problem here, there were a handful of things that you could investigate reasonably straightforward to root cause the problem.
• 24:30
  SPEAKER_01: In today's world, again, most of the things are not something you can walk into the the server room and see if there are red lights blinking.
• 24:39
  SPEAKER_01: This is all this abstract, you know, foggy cloud.
• 24:44
  SPEAKER_01: Uh, and so so step one is get this data uh right.
• 24:49
  SPEAKER_01: Step two is you have to be able to map the IT services and map them dynamically because they're changing all the time.
• 24:57
  SPEAKER_01: If it's not dynamic, then you know, five minutes from now it's gonna be stale.
• 25:04
  SPEAKER_02: So it's very interesting.
• 25:06
  SPEAKER_02: And you know, when we when we I and I can see the levels of complexity kind of growing uh logarithmically, right?
• 25:14
  SPEAKER_02: I mean it's it's orders of magnitude, more complex.
• 25:17
  SPEAKER_02: And I like the when when you're talking about the fact that there's such a heterogeneous environment of applications and services and and tools and technologies, uh, and and then the fact that they each will have mechanisms for reporting on their health and their status.
• 25:35
  SPEAKER_02: And do you want to talk a little bit about application monitoring and how that was kind of the big thing a couple of years ago, and how that's kind of fallen short.
• 25:45
  SPEAKER_02: And now we need again this more dynamic uh ability to map what's going on to still monitor the health of the applications, but also monitor the health of the uh supporting systems and applications.
• 25:59
  SPEAKER_02: So maybe you could talk a little bit about uh APM and um kind of what the promise was and then where things uh where things kind of fell down.
• 26:10
  SPEAKER_01: Yeah, so APM actually came up somewhere around a decade ago.
• 26:15
  SPEAKER_01: Uh and what the APM vendors said is infrastructure monitoring is hard.
• 26:23
  SPEAKER_01: Really, all you care about is that application performing.
• 26:27
  SPEAKER_01: If the application's performing, then who cares about all of this putting all this energy into monitoring the infrastructure?
• 26:34
  SPEAKER_01: Um, and originally this was built for developers to build highly performant applications, and so they're using this in a development environment to get the most performance they can out of their applications.
• 26:49
  SPEAKER_01: But what they pivoted to was they said, hey, let's sell this as an enterprise monitoring capability.
• 26:55
  SPEAKER_01: The selling to developers was not a strategy that was setting the world on fire.
• 27:00
  SPEAKER_01: And so, especially when developers like to use tools that are free, so they said, okay, if we're going to sell to large enterprises, let's let's make this a monitoring tool, not this thing that's helping developers build faster applications.
• 27:14
  SPEAKER_01: So it it was a good pivot and it turned out to be incredibly successful.
• 27:20
  SPEAKER_01: The message resonated.
• 27:21
  SPEAKER_01: These these companies are still doing very well.
• 27:23
  SPEAKER_01: The problem is if you're focused only on application monitoring, you know, as I described, there's this application, there could be hundreds or thousands of components that are uh supporting that application as a part of an IT service.
• 27:36
  SPEAKER_01: If it's not the application, then you don't know what it is because you're focused only on the application.
• 27:42
  SPEAKER_02: Applications failing, why?
• 27:44
  SPEAKER_02: Right.
• 27:46
  SPEAKER_01: And so I will say this APM is entirely necessary.
• 27:49
  SPEAKER_01: And you'll find it's very rare to come across a medium or large size organization that is not using one of the major APM tools, but they've they've recognized the limitations, and so they know that it you can't just focus on the application.
• 28:05
  SPEAKER_01: And and they're they're expensive tools, so they can really most this is written in analyst research.
• 28:11
  SPEAKER_01: It's not made up by me.
• 28:13
  SPEAKER_01: Most organizations are only using APM for 10 to 20 percent of their applications.
• 28:19
  SPEAKER_02: And what are some of the big vendors that you typically run across uh in that space?
• 28:23
  SPEAKER_01: The the big three are well known, and that's Dynatrace, App Dynamics, which was acquired by Cisco, and New Relic is the third one.
• 28:32
  SPEAKER_02: And then so when you go into these environments, how do you work?
• 28:36
  SPEAKER_02: So I assume you you layer on top of those tools or or underlying those, how how do you kind of uh integrate in those environments?
• 28:44
  SPEAKER_01: That is the answer, is we ingest data from their uh from their monitoring.
• 28:49
  SPEAKER_01: And so so to be clear, there's a whole set of infrastructure.
• 28:54
  SPEAKER_01: Xenos does very deep infrastructure monitoring, collecting the data and then analyzing the IT services.
• 29:00
  SPEAKER_01: Uh, the application monitoring, the APM tools, they do very deep application monitoring that we don't do.
• 29:08
  SPEAKER_01: Uh, and so these things were need to work together.
• 29:12
  SPEAKER_01: And in order to have the entire picture, you have to be able to somehow blend that data together.
• 29:18
  SPEAKER_01: We ingest the data from the all of the APM tools.
• 29:24
  SPEAKER_02: And that still needs to have some normalization done, or are those uh mappings already done because you've installed this in different environments?
• 29:34
  SPEAKER_02: How much work has to be done that's that's kind of one-off and specific to the uh organization versus what are the uh plug and play pieces, just out of curiosity.
• 29:44
  SPEAKER_01: It's pretty much plug and play now.
• 29:46
  SPEAKER_01: In the beginning, we had to understand it and and develop templates for this.
• 29:50
  SPEAKER_01: Uh, but especially as we uh we're starting to migrate towards standards like open telemetry.
• 29:57
  SPEAKER_01: And so that makes it far easier to just have the ability to ingest the data and understand it.
• 30:02
  SPEAKER_01: However, even and again, this is when we start when we started ingesting this data, it was like an integration where you had to normalize it.
• 30:13
  SPEAKER_01: Now that we have better pre-processing capabilities, right, we can ingest this data and automatically put it in the right format, correct things that are you know seem to be incorrect, dealing with outliers, etc.
• 30:27
  SPEAKER_02: So when you start thinking about um, you know, the the uh the things that organizations need to pay attention to uh around, say, governance and decision making.
• 30:42
  SPEAKER_02: I mean, certainly they have to keep the lights on and keep things running uh and and monitor applications and monitor infrastructure and monitor these supporting uh systems and processes.
• 30:56
  SPEAKER_02: What else should be part of uh AI governance from your perspective, even outside of your realm?
• 31:03
  SPEAKER_02: You know, what are the things that organizations need to look at, at least you know, at different levels?
• 31:08
  SPEAKER_02: So, you know, there's certainly at the technical level, but there's also the business process level and then you know the enterprise strategy level.
• 31:15
  SPEAKER_02: Can you talk a little bit about uh some of those elements that need to be considered from a governance perspective?
• 31:21
  SPEAKER_01: I I definitely can, but I'll say this that this is also an area where the world is still learning how to how to control this.
• 31:29
  SPEAKER_01: And so, you know, step one, and if if if if I were running a company and we were looking at these solutions, step one is just how do we make sure that this is only used for good?
• 31:41
  SPEAKER_01: And that may sound silly, but you can see you know what's going on with uh some of the large uh you know, open AI, for example, where that there's a lot of controversy and the things that can be done with these technologies can be very nefarious.
• 31:57
  SPEAKER_01: And and really, this is my biggest concern about it, which is uh not an original thought, but uh it it is AI can be used to create illusions in this world that make it very difficult for any on anyone to understand what's real and what's not real.
• 32:17
  SPEAKER_01: We we don't really face this challenge in IT operations.
• 32:20
  SPEAKER_01: We're trying to root cause problems in an IT infrastructure.
• 32:24
  SPEAKER_01: However, as as this overlaps in the world of cybersecurity, it absolutely could be used to create confusion and even chaos in organizations if you've got bad actors using these tools uh to use you know take nefarious actions with your data.
• 32:43
  SPEAKER_01: Um so at the at the highest level, it's just how do I make sure that I'm not unleashing something in my environment that becomes a risk to the organization uh where doing some, you know, doing something that's not its intended purpose.
• 32:59
  SPEAKER_01: You know, as far as technology goes, that's like the kind of moral strategic layer or risk risk layer.
• 33:07
  SPEAKER_01: In a technology aspect, uh the the risk is simply understanding where the trust should be.
• 33:15
  SPEAKER_01: And and this is something that's gonna have to build over time.
• 33:18
  SPEAKER_01: It happened in in other areas like workflow automation.
• 33:22
  SPEAKER_01: Uh, so if something goes down and I think that I know that the answer is to go spin up new virtual machines, and so I'm gonna automate that process.
• 33:32
  SPEAKER_01: You don't want to do that accidentally 10,000 times and end up paying you know, Amazon or AWS for 10,000 virtual machines you didn't need.
• 33:41
  SPEAKER_01: We we learned how to start automating things and and trusting them after you know some validation.
• 33:47
  SPEAKER_01: And with with AI, you know, technology-wise, this is happening right now is customers know that they're a little bit uh uh on the learning curve.
• 33:58
  SPEAKER_01: They don't just jump in trusting it and say, yeah, let's go automate everything.
• 34:02
  SPEAKER_01: Uh and so from a technology standpoint, it's really just a matter of of balancing the the promise and the the efficiencies that you can get versus the trust factor that still creates some risk.
• 34:16
  SPEAKER_02: Yeah, I mean you make you raise a really good point about um uh being able to optimize performance in cloud environments uh because people are paying on a consumption basis.
• 34:28
  SPEAKER_02: And what can you um talk about?
• 34:31
  SPEAKER_02: How how can some of that how can that uh optimization be achieved and again maintained, uh monitored, and so on?
• 34:39
  SPEAKER_02: What are your thoughts around uh the consumption and uh utilization of cloud services that uh need to be considered from that that cost perspective?
• 34:51
  SPEAKER_01: So, as with anything that AI does, it's something that a human being could do.
• 34:58
  SPEAKER_01: It's just the the scale at which and the speed at which it can be done.
• 35:03
  SPEAKER_01: So, this is really about examining all of your cloud resources, collecting all of the data, just to oversimplify it, about utilization.
• 35:12
  SPEAKER_01: And then AI tools making recommendations about how you can optimize, or even in some cases, automate that optimization.
• 35:22
  SPEAKER_01: And so, you know, when you order cloud resources, there are different classes of just just say servers, compute instances, for example.
• 35:30
  SPEAKER_01: You can you can change those on the fly.
• 35:33
  SPEAKER_01: Uh you can automate that change, you can you can automate the uh the cost optimization.
• 35:43
  SPEAKER_01: Now the the the major cloud providers, they could do this, and and I would guess they have the capability right now to do this for you, but these are these interests are not uh are not in line with each other.
• 35:58
  SPEAKER_01: The the cloud providers uh are maximizing revenue, not trying to help you how you can save on your your cloud costs.
• 36:06
  SPEAKER_02: Exactly.
• 36:07
  SPEAKER_02: Yeah, they they don't have an incentive.
• 36:09
  SPEAKER_02: We've seen that with um subscriptions to research uh publications where some of our life sciences customers, they'll find their book the different parts of the organization are purchasing the same research over and over and over again.
• 36:23
  SPEAKER_02: And while they have a license to it, the the publisher does not have an incentive to say, Oh, you already own that.
• 36:30
  SPEAKER_02: You know, the salesperson is not going to say, Oh, don't worry about it.
• 36:33
  SPEAKER_02: You know, you don't need you don't need to buy from me.
• 36:35
  SPEAKER_02: Exactly.
• 36:37
  SPEAKER_02: So so there's an inherent kind of conflict of interest uh when it comes to provision of those services and monitoring the consumption of those services.
• 36:46
  SPEAKER_02: Exactly.
• 36:47
  SPEAKER_01: Yeah, and there's a whole set of tools called cloud cost management tools, but they didn't start out leveraging AI.
• 36:54
  SPEAKER_01: Um, they're kind of growing along with the rest of the the IT you know vendor world.
• 36:59
  SPEAKER_01: But uh this is something that is uh the there is a huge uh opportunity for a pretty straightforward win in that use case.
• 37:09
  SPEAKER_01: So our our core competency is identifying problems in IT environments, but this whole cloud cost management is its own use case that is uh tremendously beneficial for customers.
• 37:22
  SPEAKER_02: When you think about all the sources of data that are coming into your system, so you're you're monitoring the monitoring tools and you're you know building a dynamic view of the uh infrastructure topology, right?
• 37:36
  SPEAKER_02: And do you want to define those elements of the topology?
• 37:38
  SPEAKER_02: I mean, uh you said it's more than servers, it's software-defined networks, it's all uh a number of other things.
• 37:45
  SPEAKER_02: And and then how do you deal with all of the alerts and alarms that would come from these different systems without uh you know unnecessarily notifying humans or kicking off workflows and so on?
• 37:58
  SPEAKER_02: How do you separate this signal from the noise?
• 38:01
  SPEAKER_02: I know there's a couple of questions in there.
• 38:03
  SPEAKER_02: So yeah.
• 38:04
  SPEAKER_01: Well, I'll I'll start with the last one, which is uh this is a a key element of this, that it's not it, this is part of the sales pitch to the practitioner, to the person who's gonna have hands on keyboards, because they're the ones that know getting getting a thousand alerts in the middle of the night that they have to go figure out what's going on is is a real problem and it's an ongoing problem.
• 38:27
  SPEAKER_01: And so the ability to uh correlate events uh and and deduplication is a huge part of that.
• 38:36
  SPEAKER_01: But it's it let's go back to the the application thing.
• 38:39
  SPEAKER_01: Well, let's say it's it's some uh part in the middle, a router, and that means everything behind it uh is having some communication issue with the rest of the world.
• 38:52
  SPEAKER_01: The ability for a system like ours to say it it's not these hundred things below it, it's this one router.
• 38:58
  SPEAKER_01: So we don't need alerts for all of these things.
• 39:00
  SPEAKER_01: We need to send an alert for that router, and and you know, here's what we think the root cause is.
• 39:05
  SPEAKER_01: This is uh a event correlation is an area where vendors like us, but other vendors as well, have made huge advancements because of AI.
• 39:15
  SPEAKER_01: The the number one thing that AI uh enabled in IT operations was better event correlation.
• 39:23
  SPEAKER_01: And and you know, you can go back to this security world where there were sim tools that were accumulating events from all of these security devices and trying to uh correlate them.
• 39:34
  SPEAKER_01: It never really fulfilled on the promise.
• 39:37
  SPEAKER_01: And this is like sim for IT operations.
• 39:40
  SPEAKER_01: Uh, and sim is security event and incident management.
• 39:45
  SPEAKER_01: Um this is that for IT operations, and now it includes artificial intelligence.
• 39:52
  SPEAKER_01: It it is uh already delivered more than than the sim, which was which was a huge part of every compliance mandate.
• 40:00
  SPEAKER_01: Every organization had it.
• 40:02
  SPEAKER_01: It never really fulfilled its promise uh for 20 years.
• 40:06
  SPEAKER_01: Uh, and and now it's that kind of same thing is is happening again at IT operations, but it's it is aided by artificial intelligence.
• 40:14
  SPEAKER_01: And so this was the very first place where any real benefit was coming from AI in in IT operations.
• 40:22
  unknown: Interesting.
• 40:22
  SPEAKER_02: I wanted to switch gears just for a minute and uh get a little bit uh more uh understanding of who you are and how did you get to your space?
• 40:30
  SPEAKER_02: And uh just give me give us a little bit of your you're based in Austin, I know that.
• 40:34
  SPEAKER_02: Um, and uh what how did you get to what you're doing now?
• 40:37
  SPEAKER_02: And and uh tell us a little bit about uh who you are.
• 40:41
  SPEAKER_01: Sure.
• 40:41
  SPEAKER_01: Yeah, I am based in Austin.
• 40:43
  SPEAKER_01: I put on a sweater about two days a year, and this is one of them um because it's freezing outside.
• 40:49
  SPEAKER_01: Um I started out, uh, my education was in computer engineering.
• 40:54
  SPEAKER_01: So I started out as a reasonably technical person.
• 40:57
  SPEAKER_01: Uh, I became a network architect for IBM.
• 41:00
  SPEAKER_01: Uh, let me back up.
• 41:02
  SPEAKER_01: Back when I was in college, which I won't say how many years ago that was, but it was a few.
• 41:07
  SPEAKER_01: Um my senior year, I participated with a partner in a Motorola sponsored design contest for a new microcontroller that they were launching that had AI capabilities.
• 41:21
  SPEAKER_01: And so Motorola went to the universities across the country and said, We want you to create a project to showcase the AI capabilities in this new controller.
• 41:33
  SPEAKER_01: Uh, and so so my partner and I participated in this, and we actually got top 10 honors in that contest.
• 41:40
  SPEAKER_01: And so so it hasn't been contiguous, but my AI experience started 25 years ago, right?
• 41:49
  SPEAKER_01: Uh and so uh then I got into to less AI, you know, type stuff.
• 41:56
  SPEAKER_01: Uh I was a network architect for IBM, I became a product manager at at some small and medium-sized companies.
• 42:03
  SPEAKER_01: Uh, and I I found that was my niche is running product teams in these small and medium-sized companies, and that's what I really love.
• 42:11
  SPEAKER_01: Uh, it when when you initially were saying, you know, talk describing my experience, it sounded like a lot of stuff, but really most of the first half of my career was in security, cybersecurity.
• 42:23
  SPEAKER_01: And then this is close enough.
• 42:26
  SPEAKER_01: The second half-ish was infrastructure and operations, and so it's it's not very common to say that you're a cybersecurity expert and an infrastructure and operations expert, but I'm I'm expert-ish at both of those things.
• 42:40
  SPEAKER_01: And so uh that's where I got where I am today.
• 42:43
  SPEAKER_02: And what do you do for fun?
• 42:45
  SPEAKER_01: Uh, what I do for fun, I have kids, so I spend a lot of time with my kids.
• 42:49
  SPEAKER_01: Uh, I find that uh the the most fun, but I run a lot uh and to to get some kind of zen time, I play golf.
• 42:59
  SPEAKER_02: Nice, nice.
• 43:00
  SPEAKER_01: Oh, I love it.
• 43:02
  SPEAKER_02: And if uh if I were if you were to be able to go back and uh give yourself some advice from right after you got out of college with this perspective, what what would you say to yourself?
• 43:12
  SPEAKER_02: What would you tell yourself to?
• 43:14
  SPEAKER_01: Oh man, uh that's a that is a uh hard question to answer.
• 43:19
  SPEAKER_01: I pretend pretend I'm you.
• 43:23
  SPEAKER_01: Well, this is this I tell my kids this, so it's really not that hard to answer.
• 43:28
  SPEAKER_01: That uh I kind of have an engineering brain.
• 43:30
  SPEAKER_01: This is personal, it's not even gonna help your podcast audience.
• 43:34
  SPEAKER_01: Um but I I liked the engineering type jobs.
• 43:40
  SPEAKER_01: I I was when I was younger, I was very risk averse.
• 43:44
  SPEAKER_01: Uh as I got older, I outgrew that.
• 43:47
  SPEAKER_01: And I found that the most success that I had was when I, you know, I hate to say, you know, be silly and say follow my dreams, but you know, to go for it, to to take risks.
• 44:00
  SPEAKER_01: And it's the the ironic thing is when you're young is when you really can afford to take those risks.
• 44:07
  SPEAKER_01: And when I say take risks, I just mean take that job that sounds really exciting and like what you want to be doing, even if it's with a tiny company, and if it doesn't work out, you'll still learn something from it.
• 44:18
  SPEAKER_01: Take that job in in another city or even another country.
• 44:22
  SPEAKER_01: Uh, it's all all I've I've just gotten to the point where to me it's all about life experiences, and there were plenty that I missed out on when I was younger because I was just risk averse and kind of in my comfort zone.
• 44:33
  SPEAKER_02: Yeah, totally got it.
• 44:35
  SPEAKER_02: This has been great.
• 44:36
  SPEAKER_02: Thank you so much, Trent.
• 44:38
  SPEAKER_02: We'll have your uh LinkedIn information uh in the show notes, uh the company website is z-en-os.com.
• 44:46
  SPEAKER_02: Uh, you're simply on LinkedIn, Trent Fitz F A F I T-Z.
• 44:52
  SPEAKER_02: Um, so thank you so much for um being with us today.
• 44:57
  SPEAKER_01: Again, thank you for having me.
• 44:58
  SPEAKER_01: And I just want to say one more, one more kind of plug here, but it's it's really for educational purposes.
• 45:03
  SPEAKER_01: For the past few months, we've been doing this AI explainer blog series.
• 45:08
  SPEAKER_01: So if you go to xenos.com, you can find one that starts with the very first one was here's a glossary of AI terms, and then we just start digging into what each one of those are and just trying to help.
• 45:19
  SPEAKER_01: This is part of this you know educ education process.
• 45:22
  SPEAKER_01: And so I love that.
• 45:24
  SPEAKER_02: I'll be sure to uh push that out on LinkedIn.
• 45:27
  SPEAKER_02: That sounds like a great awesome, would appreciate it.
• 45:29
  SPEAKER_02: Thank you to our audience.
• 45:31
  SPEAKER_02: We really appreciate it.
• 45:32
  SPEAKER_02: If you uh learned something today or enjoyed uh this podcast, please tell someone about it.
• 45:37
  SPEAKER_02: And again, uh thank you, Trent, for your time today.
• 45:40
  SPEAKER_02: It's really been great.
• 45:42
  SPEAKER_01: Thank you, Seth, and thank you, Liam.
• 45:44
  SPEAKER_02: And uh this has been another uh episode of the Early AI Podcast, and we will see you all next time.
• 45:52
  SPEAKER_00: Thank you for joining us on another deep dive into AI innovation.
• 45:56
  SPEAKER_00: Tune in next time when we introduce another industry expert and discuss how to maximize AI in your world.
• 46:02
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science and CMS Wire.
• 46:07
  SPEAKER_00: To learn more, visit early.com.
• 46:10
  SPEAKER_00: That is E A-R-L-E-Y.com.
• 46:14
  SPEAKER_00: Thanks for tuning in.