Host Intro (Gene Kim)

And the next speaker up is Cat Swetel, who I've admired for years, and I was stunned and delighted when she joined Nubank as a general manager of foundational infrastructure. And so that means she's responsible for all the environments that the mission-critical systems run in. It includes the transactional environment, the regulatory environment, and various tools like observability, monitoring, and incident response at Nubank.

And she also owns, amazingly, the group that is responsible for the greatest computer language program of all time, Clojure, the Lisp functional programming language, which I was geeking about with Eric Meyer earlier today. And she also runs all the bank's Datomic databases. As she quipped, "The amazing database loved by dozens." So she will be presenting her learnings about AI adoption at Nubank over the last three years. Here to talk about all the wonders of Clojure, it's Cat.

Cat Swetel

I'm Cat. I'm here.

My first slide is a disclaimer. Since Yagi said something about Datomic in his talk, tons of you have been coming up to me and asking me about Datomic, and I'm not going to talk about that today.

Yeah. If you want to learn about Datomic, normally I would send you to the tutorial and have you do that with the pro version. But given our setting, I thought I would also link to Maxdatum. And if you haven't seen it, it's very cute. It's like an '80s video game, but it's mostly the cloud version.

But anyway, I feel like people hear what my job is, and they are always really disappointed when I give a talk like this. I'm not going to say garbage collection besides this one time. Okay.

All right. So who am I?

I am a deteriorating body of flesh and bone. That's the number one thing you should know about me. The other thing is general manager of foundational infrastructure, like Gene said. And then finally, three facts that are really important. I'm a Leo rising and a Taurus sun and a Taurus moon.

Some of you know what's up, and other people are asleep because it's 3:00 PM. Okay.

All right. Now, I would like to know who you are — the people that are sticking around the end of the final day.

So who here, you've already started on your AI journey? Raise of hands. Okay, cool. And you're pretty far along on your AI journey. Is that you? Fewer people. Okay. And who's just pumped all the time? Like totally psyched 100% of the time? Okay, sure. One guy raised his hand — like both hands. So I don't know what that means entirely.

Who — and you don't have to raise your hand if you don't want. You can keep it for you inside. But who is also having a little bit of anxiety?

Two-hands guy. You? No? Okay. All right.

I guess I could leave the stage right now, as long as you all leave this talk knowing that it's okay to have more than one feeling at the same time. And I think we are a group of leaders, generally, right? That's what this event targeted. And I think that if we can't hold that fact — that people can have multiple feelings at the same time, simultaneously, even when those feelings contradict each other — I think it's going to make AI adoption in our organizations a lot more difficult. So, if you want to go take a coffee break now, that's the most important thing, I believe.

Okay, so I promised this talk would be really meandering, which I believe it is already. But here is the timeline. Here's my own personal fool's journey of AI.

In 2022, I was getting my master's in technology policy. I was taking a class from President Crow, who was president of Arizona State University, and he randomly got will.i.am to teach an AI course at ASU, which is crazy. But anyway, I'm taking his class, and he's like, "This technology is going to change everything in academia. It's going to change everything in industry." And he gave me this book, which I thought was very sweet, but it's basically like, "This is how computers work, and AI is coming." I was like, "Do you know what I do for a job?"

But anyway, that kind of kicked things off, and I was so excited. And then in 2023, I was like, "Oh my gosh, I'm going to do all this cool stuff." And it did not work. None of it worked. Literally nothing. So, that's kind of what went on there.

I tried several things. One of them, I was like, "Oh, we can use LLMs on the logs from our CI/CD pipeline." And I got a not so nice phone call that was like, "Cat, are you over there lighting money on fire?" So yeah, and the models just weren't there yet, right? A lot of the cool stuff that I could imagine, I couldn't actually do. So that was my year of disillusionment.

And then the following year, I was kind of like, "Oh, shit. Things are really going to happen. Things are going to go on. I better prepare." So that was a year of kind of — what do I need to do in order to prepare for this thing that is coming?

So again, it wasn't the most fun year, right? Because I was trying to think of what are the things that we need to do right now in order to take advantage of the future.

However, I had one win, and it was not cool at all. So we were trying to leave a vendor and go to a homegrown solution — like a big vendor. And in a very short amount of time, in eight months, we were trying to do this. And it was very overwhelming. We had our internal solution. We ended up just using AI assistants to rewrite the queries and stuff like that from this vendor into just normal SQL, which accelerated the movement a lot. And we did save millions and millions of dollars, but it felt very anticlimactic. Like, that's not actually that cool at all. But it ended up saving a lot of money.

And then in 2025, our CEO was like, "Yeah, we're going to be an AI-first Nubank." So we all started kind of moving around that. And then this year, I feel like is the year of vibes. Like, yeah, now we all know it's coming, and we kind of just have to ride this wave and be okay with the fact that we don't actually know where it's going.

Yeah. So I'm going to walk you through part of that.

So David Vélez was our CEO, and he said this. He gathered up the whole company and said, "Okay, we're entering the era of AI first." And it was like, "Yeah, rah rah." And everyone's really pumped. And then people leave, and they immediately are like, "What does that mean? I'm not totally sure. Who is we? What does era mean? And AI first — we are an operating business currently. We already were a thing. How do we re-first ourselves?"

Yeah. So there was kind of a lot of anxiety and questions.

Yeah. So now enters Cat Swetel's theory of wool management.

You can laugh or clap or something. Yeah.

Okay. So I think that managers serve a couple main purposes, and the first one is people development, right? But I think potentially even more important, and really important right now today, is contextualizing strategy. So, David Vélez gave us this big, kind of huge objective. This is where we're going over the course of years. And then each one of us — like CTO, I report to the CTO, I'm a GM — so I need to translate that locally. What the heck does that mean? So it's up to me to answer each of these questions that folks were asking.

Okay.

So how did I go about doing that? The question that I got a lot was, "Era — that sounds very long term. We have stuff to do today. So when will we do this?"

So the first thing was that I had to really make sure that people understood: you're doing this today. So if other stuff needs to leave your plate, okay, but this is happening right now. And the way that I framed that to everybody was to say, "Hey, look at our roadmap for the next year. We're going to be provisioning all these new environments. We have international expansion. This is the smallest this company will ever be again. So if we want to make changes, now is the time. Right now. Every single day, every week that we wait, the surface area that we have is getting larger. So right now, this is the time." So creating that sense of urgency.

And then as far as where to start, what to do, I posed the question to them this way. If we were founding an AI-first Nubank today — so we didn't already operate, we're founding today — what choices would we make? What components would we use? What kind of infrastructure would we provide? What approaches would we use? How would we structure our teams? How would we create spaces where AI can be basically a teammate?

And you will notice I said today. We're founding an AI-first Nubank today. For me, I felt it was very important not to ask those folks to imagine some future. Oh, in the future, Nubank will be AI first. Or, oh, I've got to imagine where all this stuff is going. No, just try to imagine the ideal today. And this is a concept from design. Yeah. So that's what I asked them to do. I'm not asking you to imagine where this is going. Just — if you were to found this company today, what approaches would you use?

I feel like everyone else has been like, "Here's all the statistics from my whole company," and I'm pretty much going to do the opposite of that right now. I'm going to tell you about one person at Nubank, and her name is Lenisy.

So — I gathered up my team. We talked through some of these things. I said, "Here are some things that I'm really concerned about." They shared with me, like, "Oh, these are other things we're really concerned about." And Lenisy shared that she was concerned, rightly so.

So the team that she's on, one of their primary responsibilities is monitoring all of the software and infrastructure limits at Nubank. So this includes AWS service quotas, rate limits, just theoretical limits like partitions in Kafka or something like that, right? So all of the software and infrastructure limits across Nubank, because we scaled very quickly. We went from zero customers to over 130 million customers in the span of 12 years. So limits — that's something we think about a lot.

And as we're thinking about AI, Lenisy is like, "Oh my goodness, there could be a lot more activity in our technical ecosystem, and that could put a lot of stress on limits." And at the time that we were having this discussion, it took three or four engineers most of their time to be actively monitoring these limits.

So, kind of every day there's this process where these monitoring systems say, "What is the limit? Is it the same as it was yesterday?" Because some of those limits are dynamic. "What is the consumption today? At what rate is the consumption growing? Is this a critical situation?" All this kind of stuff. And it does that over all of the components in our fleets. So, big surface area. That probably wasn't going to be sustainable.

So Lenisy got this idea — that's going to be a big problem. And she, despite being product operations, she said, "I think I can solve this problem." It just happened to be that one of the critical engineers was going to be out on leave. All these things kind of came together. And she did it.

She's not an engineer. In fact, when she came to work on my team, we're standing at a bodega having a beer, and she's like, "I really want to work on your team. I don't think I'm technical enough." I was like, "Oh, come on down."

And in this case, she didn't need to be, right? She figured out a way to take all of this — honestly pretty shit work from these three or four engineers — and she found a way to make it happen automatically. And it's not just that she made it happen; all of that work — new opportunities were opened up to us. So it used to be kind of just in this dashboard, and different teams were looking at that and surfaced in these different places. Well, then it was so easy to have all of this data just right there and kind of wherever we wanted it, that she said, "Hey, can I make this part of our operational metrics review? Can I make this part of these other rituals? It all just happens automatically and good to go." I said, "Yeah, knock yourself out." So now, this data that's changing very quickly as we're on this journey is now in front of our faces and across different audiences in a way that's much less painful than it was before.

So I think that's just really cool.

And then, so what happened to those three or four engineers? I axed them, right?

I did not. Three of you are awake. I did not.

Okay. So, we were able to just move them to another place within the global infra team, which is obviously quite stressed right now with all of this activity happening.

But I think it's really interesting — and several of you have talked about, "Oh, my teams are getting smaller, my teams are getting smaller," as you add agents and stuff like that into the mix. In my business unit, at least, we're going the opposite direction. Teams are getting bigger. They're just kind of big blobs that are morphing around a lot, and these few people are over here today, and then they get swallowed back up by the blob again. And these few people are over here, and then they get swallowed up by the blob again. So we've actually condensed multiple squads into just these weird amorphous blob squads. And it does seem to be working. People can really share a lot of context, and we're not seeing in those blob squads the divergence — because we are seeing in other places where people will diverge, and there's duplicate work, and now that it's AI-enabled it's going so quickly — and not seeing that so much in the blob squads.

So, go back to your place of business and form a blob.

So my other big leadership aha.

Almost one year ago, I started doing yoga. Who does yoga? Okay. Again, there are literally dozens of us. Okay. So a year ago, I started doing yoga, and I am relatively strong, so I'm working on this arm balance, where you put your hands on the ground and you balance on your hands, okay? And I'm like, "Oh, I got this. Look at me balancing. Hey, I just started yoga." But my teacher comes over and she goes, "Do you know that if you just shifted a little bit forward, you wouldn't have to muscle it out. Your joints would be aligned, and you can just balance. You don't have to use your muscles to hold you there."

And that was this moment for me where I was like, "Oh my gosh, what are all the things in my life that I'm muscling out?" And that's the thing that I keep seeing over and over again on this journey — if I can just find that point where I have people aligned, telling them, "Hey, urgency. We have to start today. Here are the areas where I'm concerned. Can we address them?" And then people like Lenisy go and figure things out. So I think it's very critical to not try to muscle it out, or that's how it's been for me, right? I'm not all up in everyone's business every day. I'm just trying to find that point where all the joints are stacked, and we can balance and be good to go. So that was my other big leadership aha.

And then the other thing that is just so cool to me about Lenisy's story is that she did have this huge desire when she came to my team. She wanted to come to my team because it is so deeply technical. She said, "I want to learn about this." I was sending her to AWS summits, all of this stuff. But she was trained as a product manager.

And now she's participating essentially as an engineer. She's consistently one of the top users of tokens in my entire 350-person business unit.

And I think it's amazing. That just fills me with delight that people like Lenisy can be participating in this way. I love it.

And the other thing that is so cool to me is that I don't see a world in which Lenisy is ever on call. I have been on call — I don't know how much of my life at this point. Some of you may have heard me get paged at this event. I have been on call for so long, and it sucks, and I don't want other people to have that experience of being called in the middle of the night, like, "Oh, we need more disk on this cluster." Like, no. That's terrible. I don't want anyone doing that. I surely don't want my friend Lenisy doing that. And now I see this world where maybe our craft will be democratized. More people can have this wonderful experience of creating something, and they don't have to deal with the bullshit that the rest of us have done, and that fills me with delight. I don't know about you, but I am pumped about it. And I see a world in which this democratizes much more than just software development, and that also delights me.

A bunch of people over the past week have said, "Oh, is this AI play about growth, or is it about efficiency?" When these people can do more, are you going to ask them to do more and grow your product surface area, all this kind of stuff, or are you going to have fewer people, just make things cheaper?

I would like to suggest a little bit of an alternative framing there. So I think that growth and efficiency — what if we were to think about that in terms of access? So if we can do more things and provide more things and make those things cheaper, we can provide those things for more people, and perhaps for people that could not afford to have access to that before. So our CEO, David Vélez, he said this: "Becoming AI-first means accelerating our flywheel by scaling to offer higher quality products at lower comps."

And I think if that promise is even halfway realized, that would be incredible.

All the financial services companies love to throw around numbers like this. Five hundred million people will gain access to the digital economy in the next four-ish years. Right now, what it means to gain access to the digital economy is you get an account that essentially does not serve any of your needs. So you go from being unbanked to underbanked. What an upgrade.

And I think there is a world in which these tools can provide services to customers — services that they actually need. I didn't have a credit card until I worked at American Express. That was just not part of the way that I grew up. I remember my mom had a store charge card, and that was it. Try buying a house if you've never had a credit card before. It is not fun.

But who has access to the information about how to become a full participant in the economy? It's people who are already participants. Their parents are participants or whatever else. And I am so excited by the future where perhaps people can get the education and access to the products that they need in order to be full participants in the economy.

So the time is now, people. This is it, right? Yeah. Not because your business is going to go under or whatever else, but because it is a magical time to be participating in this industry.

And it's ours. We are here, right? We, the people in this room, we are making the decisions about how this is all going to play out. And I'm so thankful for this community in this room. I love to look at the Agile Manifesto website. Have you ever looked at it? Do you know what the background is? Sorry, Ken. It's these guys, and they're all wearing the same pleated khakis, but just in slightly different colors. Three of them are literally wearing the same belt. They are all programming in Smalltalk. Pretty homogenous ecosystem, okay? But we look around this room — we're all coming from different backgrounds. People look different, right? That is so exciting to me.

So I'll just close by saying that I'm really thankful for the honor of your time. I loved it that Eric said the other day, "Time is the only thing of value that you have," and it really is, and it means a lot to me that you have chosen to spend this time with me. I appreciate that.

In my last ask for help — I would like stories of good vibes where this technology is doing something good for the world, for the people that inhabit the world. So if you have any of those stories, please share them with me. Okay. Thank you.