News
More from Google Cloud Next '24

Paul Lewis

Chief Technical Officer

Pythian

As gen AI dominates, Pythian looks to help businesses with a complex transition

Now that the generative artificial intelligence revolution is in full swing, affecting everything from data engineering to code development, analysts are beginning to fear that the gen AI adoption bubble will pop.“Most CIOs made the mistake, and especially in the cloud world, of saying, I’m cloud first. I’m cloud only … they put all their eggs in a single basket. [The] gen AI mistake will be putting all your eggs in a single basket,” said Paul Lewis (pictured), chief technology officer of Pythian Services Inc., who compared the AI wave to the rise of cloud computing. “You want to go where there’s options

Info

Paul Lewis | Pythian

Paul Lewis

Chief Technical Officer Pythian

(upbeat music)

>> Good afternoon, Cloud community, and welcome back to day one of Google Cloud Next here in Las Vegas, Nevada. My name's Savannah Peterson, joined by my fantastic co-hosts, John Furrier and Rob Strechay. Gentlemen, looking sharp today. Feeling smart.

>> Yeah. >> Yes.

>> We got a great keynote analysis, a lot of a precursor of the conversation we're about to have.

>> Getting all the data. We're bringing all the data out.

>> It is a data celebration. >> Yeah.

>> I feel like. And we couldn't have anyone better to talk to us about data than Paul. Paul, welcome to the show.

>> Thank you very much. I appreciate it.

>> Savannah: How's it starting off for you? How was the keynote? What's your initial impression?

>> The keynote was amazing. What I really loved about the keynote is this obvious shift from last Google Next, right? The obvious shift was GenAI was applications. You're going to build these new applications to create amazing business value. In fact, you might be a GenAI first organization, shifting that to, you know what, that's a lot of energy. Maybe it's agents. Let's talk about smaller snippets of things that I can create value in much shorter time with technology that's at hand, right? And you saw a bunch of announcements to say, "Here's data agents, here's security agents, here's workspace agents, employee agents, customer agents." That's kind of the shift, right?

>> Yeah, and it's really about the use cases and how people are looking at this. 'cause, and we, even with our partner, ETR, have been looking at a lot of the data, and they're looking at the different use cases. And a lot of them were talked about this morning, you know, from code development to marketing and content. What are you seeing from a organizations as they lean in to AI? Because they're trying to justify it in a big way.

>> Yeah, well, it's much easier to think about the internal use cases first, right? And even much easier if you don't even go outside of IT. So the very first set of use cases is Code Assist, right? They're already going to Stack Overflow to get code. Here's another way to do that where I don't have to go to Stack Overflow. And it's already connected to their IDE, already connected to the code repository. So those are easy out-of-the-box functions, sort of number one. Number two is productivity to a workspace. So let me draft an email, let me draft a document. Let's do an image. Those are kind of the easy out of the box ones.

>> Talk about your role of context before

we go further. >> Sure.

>> Because you have a unique background in operating side of IT, multiple generations of innovation, and your company's in a unique position. Explain what you guys do for your company. And then your role now as you look back at your previous role and responsibilities and you see the GenAI wave coming at you. Want to dig into, like, what's coming, and what can we learn from the past or not learn from the past or throw away from the past. But start with what you guys do first and then we'll get into the wave.

>> Sure. Pythian is a data and cloud services company. So we spend a good portion of our time with the primary data sources of our customers. Tens of thousands of databases, data warehouse, data lakes, where they're not just doing their insights, but their operational OLTP database things, right? Their core assets, we'll call them. And we've managed that for 26 years. I've been here three years. Prior to that, I was at Hitachi for a decade. Very OT-centric, right? Nuclear power plants, and bullet trains, and lights-out manufacturing plants. What it means to stop a train and make sure the health and safety of a person. And then 17 years in financial services and actual CIO of 5,000 workloads. Highly regulatory environment. You know, a consumer of technology. And that's in many ways still where I come from.

>> You've seen the movie before

many times. >> Yeah.

>> This complexity. >> Yeah.

>> The next big thing is coming. Is it a really a big deal? I mean, obviously the gen of AI thing is... What is the real... Scope the order of magnitude impact of this next wave?

>> It will be big. It's hard to say it is big now. There's a reason why it's at the peak of inflated expectations, right? There are some bad news forthcoming. However, I believe it will be revolutionary beyond cloud in terms of its actual impact in the business. Now, will it be application by application? No, I think the answer here is mostly feature-based, embedded, right? I'm going to build things. I'm going to buy from the marketplace, and I'm going to consume from my big technologies, the Salesforces, the Atlassians, the service desk, they're going to come with embedded GenAI features.

>> 'Cause they got the data. >> Yeah. They have the data. It's already controlled. It's already highly secure. It's already private. Those are good things.

>> All right, so we are talking before we came on camera about data, the reality of where we are. Are things in place? And you said the duck's got to be lined up. I said the duck's aren't even on the pond yet. (panelist quacks) So what ducks need to be lined up for an organization to truly start, you know, reaping some of the rewards from pouring more cash into building out embedded AI, AI-enabled things, AI for this, things helping AI? What's the ducks that need to be lined up?

>> It's definitely the crux of the issue, right? And which is why it's going to take a little bit longer. The reality is there's a bunch of prerequisites still there. I don't have all of my knowledge bases in a single entity, or at least not accessible by a GenAI model. I have databases at the edge in multiple data centers, in multiple clouds, in multiple SaaS products, which I can't even get the data out of. That's a problem. And we're measuring these things in petabytes, if not exabytes of data. I can't just migrate that data somewhere, right? I can't just make it available for use.

>> Right. >> Right? And therefore getting my sort of data ducks in a row, not just accessibility, but security, privacy, when I can use it, how I can use it. And really, what are the architectural and economic limitations to that, right?

>> Well, what are people doing now? What do you see as, like, the baby steps or as they say, three feet in the cloud of dust? What are people doing to like get moving? We heard from Jensen at GTC for NVIDIA.

>> Yeah. >> Accelerated computing. Everything's being accelerated. What are some of the best practice? What are people doing?

>> They're starting enterprise search, right? The unstructured knowledge bases, at least, already exist. They might not be well tuned, but they exist. I can get to my drive, I can get to my wikis, I can get to SharePoint, I can get to these things, and therefore I can, you know, I can have a better enterprise search both for my internal IT staff or even for my customers. And that's where I going to start because I have that data. Once I need to start grounding, right? Once I need to start augmenting it with my own data dictionary and my own customer information, that's where it gets a little bit more complex, right? Because I now have to have a data dictionary and metadata of my company to say, "When I say price, I mean this specifically."

>> Yeah, and I think that to me

is one of the things >> Super good point. Yeah.

>> we've been talking about a lot is the metadata and having that global metadata across there. The role of the DBA is, and having started my life on that side and financial services, when you look at the role of the DBA today, it has become uber complicated, to put it. Now they're called data engineers. And now they're trying to figure... How do you see that evolving in this market, given where data is just spread out all over the place, but you need the right data in the right place to action these AIs?

>> Well, we have 500 DBAs and we see it real time, this evolution of skillset. What DBAs used to be was ALTER TABLE, right? ALTER TABLE, and then they manage the actual database service itself, right? They patch Oracle as an example. But that's now not good enough. They need to have multiple roles. They need to be worried about getting data in. So they need to be data engineers. How do I migrate data in, how do I get data in? They need to be data platform people. It's not just the OLAP, it's the OLTP as a single device. So the database engineers, not just Oracle database, but it's also Snowflake, it's also BigQuery, it's also Databricks. They have to have a much more foundational knowledge of security. In fact, they might be a security analyst or security officer because privacy and security of that data, that's the nuggets of gold, right? That's where those bad actors are trying to get to. So now they have this convergence of all of those other IT skills. ALTER TABLE simply doesn't cut it anymore.

>> Yeah. >> I'm curious, so you're based in Canada, correct?

>> I live in Toronto. Yep.

>> Yes, so when we're talking about regulatory environment, we're talking about privacy, we're talking about governance and collaboration here, you're dealing with some of the biggest companies in the world, hands on. Is there a difference, or I'm curious to even hear if you have any observations, between Canada and the United States in terms of our approach?

>> There's definitely large regulatory differences between the two, yeah.

>> Break it down for us.

>> And it's hard to say whether either one of them are advanced in fairness comparatively. But if you look at the last executive order from the White House, they are setting up an obligation of the federal government, each one of the agencies to say, "Thou shalt have and implement a set of privacy requirements, a set of security requirements." Each one of them has to have a chief AI officer,

a chief data officer, >> Right.

>> a chief security officer. They now are as focused on, sort of, the security regulation and legislative boundaries as any other private entity. That's an assumption that's going to move from just the government to private entities. There's now an expectation that every organization from person up will have that same structure.

>> Yeah, and I think one of the interesting things, and I think, I mean we were chatting at the last Google Next, which feels like a year ago, but it was literally only like eight months ago.

>> Paul: That's right.

>> So I think one of the big things that also comes around that is how you work with your partners, and people like Google, and how that comes together, because security, you know, privacy and all of these things. How do you work with your customers to help them understand that Google is a good place for them and to work with you on that?

>> It's very helpful by having a deeper relationship with Google as an example. So Pythian is an MSP, we're also a reseller. We also help on the engineering side of some of their technologies. We'll test ALLERDB as an example. We will be involved in the product, code of marketing stuff. I'm, my team is actually a pretty big consumer of GCP. So we get to use our own selves as an example of how to deploy complex architectures into that cloud environment. And we can say, "Here's the framework we used. This is what makes sense for our 25,000 databases and 400 customers. It probably works for you too."

>> Talk about the digital transformation impact.

>> Yeah. >> We've been talking about that, Rob, for decades. It's become cliche that there's a pre-GenAI hype and/or changeover, which we've been talking about.

>> Right. >> And I think it's legit. I think the bubble will burst, but it's still revolutionary.

Totally agree. >> Yep.

>> But all that talk about digital transformation pre-GenAI was about the data lake, Snowflake,

Databricks. >> Right.

>> Maybe BigQuery's thrown in there a few times. Spanner. So, okay, cool. In comes GenAI, what has that changed in terms of the role of the CISO, CIO, CXO, and the development teams? 'Cause now you've got more stuff going on that's generative, not so much pre-programmed

or pre-stage. >> Right.

>> How does the digital transformation equation or journey change, get killed or rebooted, reset? What's your take on this?

>> Excellent question. I have the opportunity to talk to at least a hundred CIOs and CTOs every single quarter. I'm on a few boards, but I do a lot of roundtables. And I definitely have a deep appreciation for the change in the CIO's role for digital transformation over the last five years. The CIO is now at the table, the CEO, they're making the growth business decisions. They're responsible for the digital transformation program, which meant the rest of the IT team was focusing operating IT. And now there's this gap between the VPs and the CIO because they didn't get to experience that. But now the CIO is saying-

>> John: They didn't get to experience what?

>> The leadership table. >> Oh, okay. Got it.

>> They weren't part of the growth program, they were just operating IT. So the next set of CIOs might be 10 years away instead of five years away,

as an example. >> Yeah.

>> But this digital transformation program, while, what is it, 76% of companies have a program, but only 8% that are successful, right?

Very small amount. >> That's dramatic.

>> Right? >> Yeah.

>> Very small amount. Because the goal is saying, "I need to change the way I'm selling my services to how the consumers wish to actually buy." That's a customer journey question, right? That's not enable data. That's not even just monetized data. That's saying, "I have a certain set of customers that are going to some other competitor. What do I have to change holistically to make that happen?" And some of those holistic changes are simply the difference between more and better. Most companies are more features, more functions, more products, more things, more stuff, where customers are actually asking for chaotic environments. "I don't want a 30-year relationship. I want a one-month relationship. And if I don't like your app, I'm going to delete the app and download a new one." That's a very different customer journey, which means I have to rethink how am I'm going to implement technology. One side, I have to monetize data, right? How do I get a better insight to my customers based on the data that I have, and then how do I create using GenAI or new talk with my data or new enterprise search or new, you know, creative means to create a different journey for that new expectation, that new segment, that's going to some other cloud native environment. That's kind of the big difference between the two. Digital transformation used to be just monetized data, right? To get that journey, now it's monetized data actually create new interaction.

>> Yeah. >> That requires GenAI.

>> John: That's a product conversation.

>> Yeah. >> It is.

>> That's not an operational thing so much as it is both operational and product development.

>> Exactly. >> And, by the way, generatives run time product development. (Paul laughs) So you're generating product on the fly, you got to enable an entire new data in the market, marketplace.

>> And back to those prerequisites, that sounds expensive, right? That is a expensive adventure to run and manage

your own models. >> Yes, it is.

>> Tune and retrain your models.

>> John: Or hire someone to architect it.

>> Hire a bunch of people.

>> Who do you hire to do this? I mean, obviously, you guys, you guys are in this business, but this is what I find people are struggling with. Okay, I get it. It's mind-blowing. I can see the leap of faith, that bridge to the future. But what do I do now?

>> Well, I can tell you the biggest error CIOs have made in this GenAI period of time is they gave those projects to the chief data officer or to the BI team. The reality is GenAI is a software engineering project.

>> Yes. >> I have to have pre-processing and post-processing. And I have to make sure it's guided by the principles of my organization. And that hallucinations don't make errors. Prompt engineering is a great example. Prompt engineering in the consumer sense is asking a bunch of questions to hone my answer. Prompt engineering in the enterprise sense is asking a bunch of questions that gives me one answer. 'Cause there's only one right answer.

>> Well, we are, Bob and I were talking about this. Paul, it's a great point.

It's a systems problem. >> Yes.

>> Not a, "I got to analyze the data and do some schema work."

Unstructured data. >> No.

>> Right. >> The database. Okay, we can query it.

It's pre-programmed. >> That's right.

>> Prompt, response. No reasoning, really. So, I mean, Rob, this is what platform engineering, All this we've been talking about.

>> Yeah. >> Software engineering.

>> It's all those data ducks in a row like we were talking about.

>> Yeah, and I think another piece of it is, and I think you just hit on it, is how do you treat this as a product,

not just a project? >> Right.

>> And I think that's the big difference is that companies that are doing AI as their product get it, because that's all they're doing. But when you're dealing with organizations that that's not their primary role, right? You're going in there and trying to help them understand how do you monetize your data? What's the data you want to do, you know, use as part of that? I would assume it's a much different conversation with those customers about how do you really leverage and what is the right cloud to be leveraging for that matter as well?

>> Yeah, a hundred percent. Most of our conversations, especially with our clients that is let's go back to the drawing board. Let's have a educational conversation. Here's what GenAI is and isn't. Here's what you can or cannot do. And then let's look at all your potential use cases, right? And let's put in the two by two matrix. The difference, though, is in the old days, we would never have gone to a customer and said, "Here's Cloud, give me all the use cases to deploy to Cloud." Because it doesn't make logical sense. Here's a database.

>> Savannah: And they wouldn't know that.

>> Yeah, Give me all the use cases to implement it. But we do that with GenAI. Here's GenAI. Tell me all the things we could do to create value for your data or create a new customer journey or all of those things, which they're much more naturally in tune to giving you an answer to.

>> Well, great insights, and this is a great interview as we explore and riff on this. We were using a baseball metaphor earlier, as we always do, early innings, you know, later innings the value will come in. But in one case, Rob and I were talking about like, people are so starved for value, to prove the value of GenAI that you see RAG, retrieval-augmentation generation, become the hottest app. That's just data wrangling. It's a search problem. And so that is like, I don't want to say desperation, but that's a easy way to go with for proof points with my own data. So what are going to be the value points that you see people doing out of the gate besides RAG that's going to be value-based, where they can get a win, a single, lay down a bunt, get on base, to use our baseball analogy? Because that's what people are trying to do. "Hey, boss, look at, we got some movement here. We moved the needle a little bit. We got on base."

>> It's mostly going to be internal use cases, right? So they're going to start with the enterprise search, with their knowledge bases. They're going to start with Code Assist, right? Helping the development staff. They're going to do auto commenting on some of their code. They're going to say, "Evaluate this old legacy technology, legacy SQL, legacy code, legacy infrastructure." And say, "Tell me about this because this person no longer works here," right? Document the things that I have. And by the way, I have a hundred different documents that I've produced by Sally. Sally no longer exists. What did she do for a living again?

Right? >> Yeah.

>> Because I'm trying to hire somebody else. That's the real win use cases now. And it doesn't require sort of the privacy, and security, and governance, all of those extra functions that I need to worry about for external use cases.

>> So, momentum.

Get the momentum. >> Yeah.

>> Hey, boss, can you imagine if we did this, what we could do for this project or build this product?

>> Get the skillset. Once you have the skillset, engineer new projects, then you can start to worry about the hard ones.

>> And get a win. What's some more advice? I love your advice for CIOs right now. And I mean, you're touching a lot of, like you said,

a hundred a quarter. >> Yeah.

>> What are some of the big mistakes or avoidable things that maybe haven't happened in the past, but as someone's evaluating their strategy for this year and looking forward for the next three or so, what would you tell them or what are you telling them?

>> Most CIOs made the mistake, and especially in the cloud world of saying, "I'm cloud first. I'm cloud only." They right out of the box create a technology principle that didn't make a lot of long-term sense, right? Because they put all their eggs in a single basket, GenAI mistake will be putting all your eggs in a single basket, right?

Choosing- >> In terms of foundation model or technology or all the above?

>> Technology suites, right? >> Okay.

>> You want to go where there's options. So go to a model garden that has 30, 50 models. 'Cause I want to be able to use different models over time. They will get better over time. They will get cheaper and more expensive over time. Right? Presume you're going to have a bunch of technology tooling that you might want to displace and replace all the time. And also presume that your data will be divided and conquered across a variety of different opportunities to exist. So know that some of those analytics need to stay there versus central. If I have analytics in a bank branch, it probably needs to stay in the bank branch. So that's not really a cloud opportunity.

>> Yeah. It's architecture.

>> Yeah. >> It's a system. You're building a new operating system.

Operating model >> Exactly.

>> for the corporation.

>> Yeah. >> So it's a whole reset.

>> And you kind of have to think of infrastructure,

applications, data analytics. >> Yeah.

>> It's almost a fourth pillar now,

right? >> Yeah.

>> Because it's distinctly different than those other three.

>> And I think that's why people are so hyped up. That's why I think there's going to be a big bubble pop,

like the dot-com bubble. >> Yeah.

>> Because it's obvious, we talk about it, I can see it happening, but then it's like, shoot, how do we do it? (laughs)

>> Yeah. Infrastructure is the bubble, right? There are very rich infrastructure companies at GenAI. It's going to hit a peak at some point, right?

>> We got to race around now guys, let's go. (all laugh) Our new CUBE AI, got to go funding, let's go. (laughs)

>> Speaking of pillars, what are the three pillars of GenAI adoption?

>> So we want to make sure they're accessible, we want to make sure that they're enabled, and we want to make sure that they're secure, governed. Those are the sort of the big threes of making sure that that data is, all the prerequisites are checkmarked, yeah.

>> Okay, so this has been fascinating, and you obviously have quite the lay of the land, and the pond, and the ducks on that pond, to continue our favorite metaphors of the week. What do you hope that you can say sitting next to us next year that you cannot say yet this year?

>> I'm going to, I believe that we will have real-world major examples within our organization with our customers that have an obvious ROI, right? But not big projects, not big multimillion-dollar things, small, $100,000 quick wins that have obvious in-quarter ROI, that's the goal. Let's have a thousand of those.

>> In-quarter MVP ROI for GenAI on some of these projects.

Love it. >> Exactly.

>> Can't wait to be chatting about it with you in 2025. Paul, thank you so much for joining us for this fascinating conversation. John and Rob, always a pleasure. Your insights and your questions were fantastic. And thank all of you for tuning in to day one of coverage here at Google Cloud Next in Las Vegas, Nevada. My name's Savannah Peterson. You're watching "theCUBE," the leading source for enterprise tech news. (upbeat music)