theCUBE + NYSE Wired: Practitioners Series | Lawrence Fitzpatrick, OneMain Financial

theCUBE.net

Clips
More from theCUBE + NYSE Wired: Practitioners Series

Lawrence Fitzpatrick

CTO

OneMain Financial

play_circle_outline Managing Rapid, Non-Deterministic AI: Guardrails, Continuous Curation, and Living Product Operations

play_circle_outline Emphasize "speed to learn": iterate quickly and validate by doing

play_circle_outline Agents acting on behalf of humans raise performance, evaluation, HR questions

Info
Transcript

Lawrence Fitzpatrick, OneMain Financial

Lawrence Fitzpatrick

CTO OneMain Financial

In this interview from theCUBE + NYSE Wired: Practitioner Series, Lawrence Fitzpatrick, chief technology officer of OneMain Financial, joins theCUBE's John Furrier to discuss how enterprises can build the confidence and operational discipline to deploy AI at scale in a non-deterministic world. Fitzpatrick frames the core challenge as AI's fundamental unpredictability — unlike traditional IT, where automation handles break-fix scenarios in predictable patterns, AI requires new guardrails and evaluation mechanisms before organizations can trust it at scale. He ... Read more

explore Keep Exploring

How have organizations experienced and responded to the cultural clash between humans and increasingly non‑deterministic AI tools, and what are the trade‑offs and measures (e.g., guardrails, evaluation, monitoring) needed to build confidence to deploy them at scale? add

How should organizations change their mindset and approach when building and deploying these kinds of solutions so they can learn quickly and move from pilot to production at scale? add

How should organizations approach evaluating AI agents' performance and budgeting for their token costs and impact on workforce planning? add

bolt Powered by CUBE AI

Lawrence Fitzpatrick, OneMain Financial

search

John Furrier

>> Welcome back, everyone. I'm John Furrier, host of theCUBE here at theCUBE's New York Stock Exchange Studio. Of course, we have our Palo Alto Studio connecting Silicon Valley and Wall Street. This is part of our NYSE Wired program, a CUBE original, also open community of leaders connecting with each other. This Practitioner Series is the focus of real world people solving real world problems. Larry Fitzpatrick, CTO of OneMain Financial. Welcome to the podcast. Thanks for coming on.

Lawrence Fitzpatrick

>> Thank you.

John Furrier

>> You guys are part of the Mayfield dinner tonight. They gather all the practitioners. It's been one of our hottest areas when we do videos with leaders who have been there, done that, and are on the frontier, literally of AI. So thanks for coming on. So I guess my first question is, you've seen waves, you've participated in them over your career. This one is more fun. It's faster, high velocity. What's your thoughts on where we are and just as AI is continuing to thunder away?

Lawrence Fitzpatrick

>> Yeah, yeah. No, it's a really exciting time. You have the normal technology injection, takes a while to figure out how to use it. Initially, people are doing incremental things, but the business model changes happen. And that's when you really get the leverage. But this time, I think there's a couple of things that happen that are very different. One is it's super fast. It's changing incredibly fast. And the second is it's non-deterministic. And that is something that organizations really have trouble adapting to, and I think it's one of the big bottlenecks for adoption. I can kind of-

John Furrier

>> It's not predictable.

Lawrence Fitzpatrick

>> No.

John Furrier

>> It's not a workflow kind of thing you can just lay down and process it.

Lawrence Fitzpatrick

>> Right. I mean, the way I think about it is we've had about 10,000 years to get cohorts of unpredictable humans to operate in a way at scale where we feel comfortable with them, and we use things like payment, legal, employment, peer pressure, training. We don't have any of those devices for a machine that has enough autonomy and can be non-deterministic. And on the technology side, we've been in the build and run, and the run has really been, you can automate away most of the run. And in fact, it's really just very deterministic looking for break-fix. And so the run side of technology is very different. It doesn't really apply in this world where the technology can be unpredictable. And so what we're really working through is, how do we get comfortable with the unpredictability of the tools? What are the guardrails that we need to build? What are the mechanisms that give us the confidence to be able to deploy this stuff at scale without doing any damage-

John Furrier

>> Yeah. I mean, that's a great point, Larry, because the enthusiasm is high.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> The next layer up in the attitudinal scale is confidence.

Lawrence Fitzpatrick

>> Yes, exactly.

John Furrier

>> So that's where we have this. So I have to ask you a question because that is interesting, because humans are not deterministic.

Lawrence Fitzpatrick

>> No.

John Furrier

>> So how have you seen this culture clash collision play out in the past few years? What's observations? Humans would love non-deterministic tools that worked on their behalf, but then, okay, trade-offs. First was hallucinations. Now it's other stuff, goes out there and does things. What's your observation on the balance? Is there balance in the force, as they would say?

Lawrence Fitzpatrick

>> Well, I think the big thing for organizations is how do you develop comfort? And you develop comfort when you have some form of predictability. And so how do you get predictability? And so I think a lot of the energy for most enterprises is in trying to figure out how to put evaluation into the mix. It's like if you have a very large workforce. You don't just ship them out and have them do work and don't monitor what they do. So we've got to change the way we monitor how this stuff works. And I think because the build cycle is getting so much shorter with this technology, or at least that's the promise of it, that traditionally in IT, you want to spend 80% of your investment on the build side and 20% on the run side, maybe it's 70%, 30%.

John Furrier

>> Yeah, yeah.

Lawrence Fitzpatrick

>> I think the shift is going to happen where you're not just launching something over the wall into production, you're getting something up and running. And then you're going to have to curate it in production as if it's a living thing that you're going to constantly have to look for, well, where does it do things we didn't expect it to do? What insights are we getting from the dialogue? For instance, it may have a dialogue when you call up three different ways, and we may find through analysis that one of those ways happens to be more effective than another way. So we want to take that learning and bake it back into whatever prompt construction we're doing in order to make the thing operate. So I think it's going to shift to operations in a very big way, kind of like we do with humans.

John Furrier

>> Yeah, yeah, I totally agree, and then it'll just be flexible and agile and elastic.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> I have to ask you a personal question and put you on the spot for a minute, because one of the things that's come up a lot in my past year of talking to experts is, everyone has either some frameworks and mental models that they've been leaning on to frame the systematic impact. Because it's a systems revolution.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> I mean, I've been saying that for years. Now everyone's admitting it and agreeing. It's obvious it is. It's not just, "Hey, build an app and here you go, App Store." It means on the systematic. Is there frameworks and mental models that you use to understand certain problem sets or the way to think about what's coming, because you can see these unknowns, so how is it deterministic?

Lawrence Fitzpatrick

>> Right.

John Furrier

>> I'm just curious, is there-

Lawrence Fitzpatrick

>> A few things. One is what I already talked about, which is the, you've got to change the way you think about building these solutions and shift your mindset to run. How do you run and curate what's going? The second one is, and most enterprises take a long time to deliberate doing something and they spend a lot of time on the specification side, in this case, we have to shift our mindset to speed to learn. And that means you learn by doing. And so we've got to find ways to actually execute faster. There's no way we can study this for two years and come out with a solution, and the longer we wait to actually do, the longer, the farther behind we're falling because the world just keeps changing on-

John Furrier

>> I love that speed to learning.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> So you essentially got to get your hands dirty-

Lawrence Fitzpatrick

>> You have to....

John Furrier

>> so to speak.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> And get in there and get muscle.

Lawrence Fitzpatrick

>> Yeah. And I think the way to frame the speed to learn is you have to go into it saying, "I have a potential business case, and my goal is to get to production at scale." And if you don't keep that in mind, then it can turn into a little pilot where you don't really get anything out of it.

John Furrier

>> Yeah, it just creeps and leaks out and just-

Lawrence Fitzpatrick

>> Yeah, and there's no close, right? And it may be that you say, "Hey, I think there's a hypothesis here that we can drive a top line or a bottom line return by making this investment." And you may find three, four, five months in, you can't and you'll shut it off. But-

John Furrier

>> Yeah. I love that. I have to throw this at you because you remind me of something so funny. Well, at least to me it's funny. The cliche in the old days was, "Transformation journeys, let's talk outcomes." Okay. And when I say cliche, because they meant let's talk outcomes. I mean, you just basically said, "My outcome is run stuff in production at scale."

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> That's a preferred future state.

Lawrence Fitzpatrick

>> Right.

John Furrier

>> Okay, that's cool. You can keep that as a north star and just-

Lawrence Fitzpatrick

>> It's got to be working backwards.

John Furrier

>> So yeah, the notion of what outcomes were before and how that got done, what's the biggest shift? Because in the old model was, "Hey, here's our north star." Get a meeting, you do stuff, get a team, go out and do a pilot, collect everything. What's been the biggest shift or change in how to go after the outcomes of the thing? Because I think selling outcomes and designing outcomes is actually more relevant now than ever before, because you can actually get to them faster.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> And not have to give up or abandon or political challenges and organizations. "You didn't show us the money fast enough."

Lawrence Fitzpatrick

>> Right. Right.

John Furrier

>> So things die for multiple reasons.

Lawrence Fitzpatrick

>> Yes. Yeah, I think the hardest part is to get organizations to reduce cycle time. So instead of having very large meetings, lots of meetings to discuss what you're going to do, one of the other things we say internally, and we're trying to get people to exercise this way is demo, not memo. It's like, go ahead and put something together and let's see what we can do with it, and let's figure out how to get it into some use as fast as we possibly can. Part of it is accelerating how you do change management. And that may require some restructuring in terms of your organization, whether that's carving off part of your operation to be recipients of some of the innovation, so that they're partners in this process with you. And so you get the feedback real time with real business activity. I think those are important-

John Furrier

>> Demo, not memo. So take me through, what was the old way? What was the memo like? Someone would just bang out a memo.

Lawrence Fitzpatrick

>> Lots of PowerPoints.

John Furrier

>> Send an email? PowerPoints? Okay.

Lawrence Fitzpatrick

>> PowerPoints. PowerPoints.

John Furrier

>> And then present it, you're sitting there and then you'd be checking your email and just another-

Lawrence Fitzpatrick

>> Document analysis. Do we have requirements? Do we understand what we're going to get? Do the analysis on the outcome 17 different ways. So there's a lot of thought that goes into preparing to commit to do something.

John Furrier

>> Yeah.

Lawrence Fitzpatrick

>> So I think part of it is getting faster at producing a result and then deciding to cut quick if you can't get something.

John Furrier

>> That's a great point. All right, so let's talk how about your role, CTO. What's your day like? What do you do? How's that relate to the business? Can you share? Because you mentioned a lot of the things that are crossing over with the agents. There seems to be a new C-suite formula and it's all rallying around this trend that we've been observing, which is, I'll just use Jensen Wong at NVIDIA, he talks about extreme co-design. There's a lot of co-designing going on in successful organizations, not from an NVIDIA standpoint, but this idea that unification, horizontal, there seems to be like this systematic co-design. Not, "Hey, HR, you got your swim lane. Stay in your lane." There seems to be a C-suite formula. What's your thoughts on this? Because you're in there.

Lawrence Fitzpatrick

>> Yeah, right. So I think at the end of the day, it's about putting together the teams that are composed of the people who are building with the people who are consuming. And so we've adopted a product operating model, which sounds easy. You can say the words, but it's really hard to keep it going. And so creating really strong partnerships at all levels of the technology organization with our business partners, our product partners, our finance partners, to make sure that everybody's in the game together at the same time. And that creates that co-creation where you can actually have trade-off conversations, because at the end of the day, technology's role is about offering trade-offs. It's like, "Hey, I hear what you want to do. We could do it this way or we could do it that way. It might take this long or that long." And so you can't have that conversation over documents and swapping emails-

John Furrier

>> Yeah. And you also mentioned earlier that the tokens is a finance problem, or not problem, a finance concern-

Lawrence Fitzpatrick

>> Yeah....

John Furrier

>> and then they're also doing work. So potentially an HR, I mean, they are basically working.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> Agents are working on behalf of the human on tasks.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> Did they do the task? Is there performance review? Good judgment. Customer satisfaction. Proficiency.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> I mean, these are checkboxes on a standard evaluation.

Lawrence Fitzpatrick

>> Yeah.

John Furrier

>> Yeah, so what's your thoughts on that? Because that's kind of crossing over.

Lawrence Fitzpatrick

>> Yeah. I think it's going to take some time. I think the hard part about that is forecasting cost.

John Furrier

>> Yeah. Oh, yeah, yeah.

Lawrence Fitzpatrick

>> I mean, when you start thinking about a future of one, two, three years out and we say, we're going to be consuming tokens at a rate of what? We don't know. And what is that going to translate into how our mix shift of workforce changes? We don't know. And there aren't a lot of people out there who you can kind of look. I mean, it takes a little bit of a crystal ball. So we try and keep the focus about two quarters out. Obviously we do budget planning with everybody, but try and keep the focus about two quarters out and learn our way into it.

John Furrier

>> It's interesting. I asked CFOs that question and I get an interesting response. I liked your answer there. Their response was, "We're going to start changing how we budget for growth." So they're not going to try to go to the departments and saying, "Here's your budget, squeeze it. You're not going to get much growth," because the data growth is growing so much faster than the budgets.

Lawrence Fitzpatrick

>> Right.

John Furrier

>> So what's actually happening is the incentive from the CFO is saying, "Here's your budget to run things two, three months," quarters out or whatever. "Here's a growth budget. Here's a transformation budget." So yeah, this is maybe R&D. You want to get more run budget, you got to use this to get more profit. And so they're tying an incentive to the tech teams.

Lawrence Fitzpatrick

>> Yeah, I think that the way the tech teams can solve that is by tying up to a business capability. So to the extent that you look at the business capability as a mix of business operations, business leadership, innovation, and the technology that supports that, then you actually have a business leader who can make a P&L decision around, do I want to invest more in technology and less in people, or more in people and less in technology? Or how does that mix work? And I'm reminded of, I don't know, you probably remember John Zachman, the founder of EA.

John Furrier

>> Yeah, yeah.

Lawrence Fitzpatrick

>> He had a little squeaky voice and he used to say, "50 years ago, there were 25 accountants doing that job and they were the business. Now there's five programmers on a computer. Why aren't they the business?" So you really have to think about that composition end to end. If we don't have that value stream look at it, it's very difficult.

John Furrier

>> Well, Larry, thank you for coming in and participating in our Practitioner Series. And you got a great event with Mayfield Fund, a really well-known Silicon Valley VC firm. I know I've lived in Palo Alto for over 26 years, just got an apartment in New York. So I'm psyched to have them come in and bring their network.

Lawrence Fitzpatrick

>> It's a real pleasure.

John Furrier

>> Yeah. Thank you very much. Appreciate it. Yeah. The mental models, whatever frameworks you want to use, the action's going to be execution in AI and taking the right approach and having working examples of just time to learn. Get out there, don't lag behind. That's the theme here. Continues to be the theme. Get your hands on the AI and get the tools in the hands of the users and the employees and customers. I'm John Furrier, host of theCUBE. Thanks for watching.