Andrew Bialecki: 00:00

If you grew up in the era of tech in the twenty tens, you've got a big bias that's gonna work against you because you're used to working a certain way, and you need to reunderwrite everything about how you think, how you work.

Wade Foster: 00:23

Hey, folks. We're here with Andrew, who is the cofounder and CEO of Klaviyo. Klaviyo is a Boston based marketing automation software platform and is a pretty unique company. They've bootstrapped Klaviyo in its early years, got to a million in ARR before ever hiring any employees or raising any venture capital. You know, many, many years later, went public, in, 2023.

Wade Foster: 00:47

And today, Klaviyo serves over 140,000 customers, all across the globe. And so stoked to have you here, Andrew. To kick it off, the thing I was curious to hear you talk about is your the founding story of Klaviyo is pretty unique in that you didn't raise any money for a long time. I think it was just you even for, like, a good while.

Andrew Bialecki: 01:11

My cofounder and I, thank god, you need cofounders. Right? We can talk about that.

Wade Foster: 01:13

Right. So you had you had a co had a cofounder. Walk me through, like, what are you doing before Klaviyo, and then why why'd you start this thing, and why'd you do it the way you did?

Andrew Bialecki: 01:24

Yeah. Sure. So my so pre KLAVIO okay. So put a little bit of a chip on our shoulder here. But my my undergrad was in was in the sciences, so I was a physics major.

Andrew Bialecki: 01:37

And I remember getting into CS. I remember actually the way I got into programming was I was working at this observatory. So MIT has like a satellite dish, like an observatory, right? And it's a bit old school where like all the scientists hang out at the observatory. You don't really need to do that anymore because everything's digitized.

Andrew Bialecki: 01:54

But anyways, so their observatory is in like New Hampshire almost. So it's kind of middle of nowhere, which I mean is good. It's good for, you know, light pollution, all this kind of stuff. And anyways, I was given this like data cube and like, they're like, hey, need you to like post analysis on this thing to figure out how these two galaxies are slamming into each other so we can learn about that. And I was like, wow, this work is really interesting.

Andrew Bialecki: 02:14

Like, the pace is a little bit too slow for me. The people were excellent, but it got me into this, archaic programming language. I can't remember what it was called, MIPS or something like And I was like, okay, I got hooked on that. And so I took a couple of CS classes. But then when I graduated, I was like, well, I wanna get into tech.

Andrew Bialecki: 02:31

I had interned at Microsoft for a summer. I had a great time in Redmond, but it was too big of a company. And I was like, alright. Well, I just wanna I wanna work in a tech company. And I remember the time, like, you know, Facebook and Google were hot.

Andrew Bialecki: 02:44

And so I applied. I remember it's like, just know, now the term is ghost. I just never got any reply. Right? I was like, okay.

Andrew Bialecki: 02:50

Guess I'm not good enough for that. So, yeah. So I basically before Klaviyo, I joined a bunch of basically earlier stage, you know, we, I now we call them startups. You know, the first company was doing kind of, you know, large scale data analytics, but this is like before we had distributed computing. So it was basically, like, buy the biggest Dell server you can.

Andrew Bialecki: 03:09

Right? And, like, see if you can jam, you know, all Starbucks's or Walmart's data onto it and make it reasonably performant. And then I worked at a couple of start ups, you know, outside of in in Boston. One was doing some analytics and some marketing stuff. So you can kinda see where the story is going.

Andrew Bialecki: 03:25

Right? That company actually great company, great set of founders. They became mentors to me and, like, got acquired by HubSpot. And I worked at a start that didn't work out. Was also a great, great team, but just like, know, never found product market fit.

Andrew Bialecki: 03:40

Yeah, and then by the time we got to Cladio, was like, all right, I'm really into data. I've been doing a lot data infrastructure stuff. And like, let's just get it going. And so my co founder Ed was, he was finishing up his MBA at MIT. So I was like, okay, this is the thing, you need a co founder.

Andrew Bialecki: 03:56

And I'm really curious on your like, you know, hey, do you need a co founder? And if so, how do you find them? And like all this good stuff. But so Ed was somebody I'd worked with before and he was getting his finishing his MBA. And I was like, all right.

Andrew Bialecki: 04:07

So I cornered him in this like MBA conference room or MIT conference room. It was like, all right, Like, will you just I was like, practiced my pitch. I was like, will you just work with me while you're finishing your MBA? Cause I'm like, look, there's basically, there's no risk to you. Like you can still get your MBA and go take whatever fancy job you're gonna get if that works out.

Andrew Bialecki: 04:23

And two is like my impression of MBA students was like, you're not really working that hard anyways. So like you definitely have some free time. I have no idea that's actually true because I don't have an MBA, but I was like, that's my impression anyways. So I was like, why don't we just hack on this thing for like the second half of your second year of doing your business degree? And if it works out and we don't hate each other at the end of it, then we'll keep going.

Andrew Bialecki: 04:41

And if it doesn't work out, then no harm, no foul, right? It's just a big experiment. So I remember like convincing him to do this. Now he would tell the story that he's like, well, actually I wanted to start a company, so was actually really good timing. Yeah, I mean, that was kind of off the races for us.

Andrew Bialecki: 04:57

We were just packing in a MIT conference room and we spent five or six months and got enough traction or whatever in the early days. We were like, alright. Let's play it forward for another six months, and off we went.

Wade Foster: 05:06

Yeah. So how did you did you pick email? I mean, you know, it's like, of all the spaces, like, there's you know, like, you got Mailchimp, you got Constant Contact, you got ExactTart. Like, you can sort of run the gamut of, you know, billion dollar email companies that already exist when you all are starting.

Andrew Bialecki: 05:24

Yeah. Well, so actually, like, the so when we first first first started, because I because my background was, like, in, you know, databases, data infrastructure. Mhmm. I'll say, like, we're we're kinda sitting at the union of two ideas. One, I've always been very interested in this idea of, like, how do we audit automate, like, you know, how do we scale and automate ourselves?

Andrew Bialecki: 05:43

I mean, I think, like, aren't most of us as, you know, maybe probably people, but certainly, like, programmers, we just, yeah, you naturally love that because that's why you write code is to automate something that you didn't wanna have to do. But I always like taking that to the extreme, like what would it take for me to build like a surrogate for myself? Because while I worked at those other startups, I'd spun up a couple of my own little side projects. It was fun to build, but I realized like, man, like talking to customers and helping them understand what you're building, like in theory, if your product's perfect and you do the perfect marketing and product marketing, like people just get it, but in practice, that's like not how it works, right? I'm sure you remember, I mean, I remember showing my parents, like, I don't know, how to use like Gmail for the first time or how to use whatever search engine or how to use down to like ChatGPT.

Andrew Bialecki: 06:22

So I mean, like even these products that we think are so intuitive, like they still need some voiceover helps. And you're like, okay, well, the only people intelligent of the time are like humans. So let's do that. I was like, man, wouldn't it be awesome if you could build a system that could take some of the stuff that you and I know and just give it to the rest of world. Now with LMs, that feels dumb obvious, but back in 2012, we're starting it felt like it's like, well, how would you do that?

Andrew Bialecki: 06:45

So my first obsession was like, okay, if you're gonna do that, we need to find a way to like replicate a brain. So was like, literally like, what are the data structures that sit inside of like our head? I don't you know, so there's obviously a bunch of neurons and stuff, but I like, well, let's talk more about the outputs. So they have these kind of tendencies of being a little bit like a we think, like, data warehouse, data like, OLAP system being very analytical. Like, you run hard queries, do thinking and large scale data.

Andrew Bialecki: 07:08

But they're also like have properties of like these transactional systems where they're fast. Right? I mean, if you ask, he's like, hey, when's your brother and sister's birthday? You just like instant record retrieval. Right?

Andrew Bialecki: 07:18

You just get it. Right? And not in seconds, like in milliseconds and, you know, kind of at scale. Right? So I was just obsessed with, like, what would it take to build a bunch of data structures or really hybridize these like this idea of a fast database and one that's, like, really good at thinking into one thing.

Andrew Bialecki: 07:33

So we actually that's what Clico first was. Our first customers, I remember our first customer was a parking lot. They just wanna do analytics on like who was parking there. And then we got into a bunch of other businesses and actually fun story from way back when is we actually, you know, we spent a lot of time with like the Mailchimp team and CampaignMar, great crew. They were what they were really interested in was like not so much what we did for messaging, but just the fact that we had all this data.

Andrew Bialecki: 07:57

And it was like, we're really good at slicing it up and this kind of stuff. And so the way we got to email was our customers were just using this kind of like database kind of analytics system we built. And I was like, okay, well, what are you exporting all the data to? Or what are you using it for? And they're like, oh, we're putting in, you know, exact target, Mailchimp, Costco, all this kind stuff.

Andrew Bialecki: 08:13

And, yeah, I mean, I I was like, okay, well, that's cool. I remember, you know, we worked with those systems and what we found was they actually weren't very good at, like, accepting all the data inputs we had. Right? I'm sure you find this a lot with like, you know, Zephyr was like, hey, it's not these two systems don't necessarily fit. Right?

Andrew Bialecki: 08:28

And you're like, you sort of have to figure that out and that's hard and you sort of wish that their data models were more compliant. So we realized that was gonna be a problem. And then I remember talking to couple of, I'm like, Hey, it's gonna be kind of hard for us to integrate in a way that's really good. What if we just built this marketing thing? And then I felt like the race was on to, okay, we're this database company.

Andrew Bialecki: 08:46

Can we build marketing faster than the, you know, whatever, the marketing company can figure out how to become a database company, and off we go.

Wade Foster: 08:53

Got it. So you you effectively almost backdoor your way into this observation that there is this big wedge, this big gap in all the email platforms. And if you get really good at that, you're able to carve out I wanna say niche, but it's clearly not a niche. It's clearly a core capability that marketers need to run their companies.

Andrew Bialecki: 09:17

We thought like, I think in general, I very much believe that, know, have mental model that's like for between a business and its end user customer, whoever it is. Right? We mean this very abstractly, like it could be a software company that has a bunch of users, right? Could be really long retail brands, could be there, you know, people that are buying their, you know, the stuff that they create, right? They make, but also it could be like a teacher to a student, anywhere where there's this like one to large end.

Andrew Bialecki: 09:44

I think the problem we observed was there's so much software had been built where it was like, yeah, if you put a person in between those folks, right, to sort of mediate that relationship, like there was software for that, right? That's normally what we think of as like classic CRM. But we thought like, well, what if you put software in the loop and expected it to do all the work? Nobody had really thought through that, or we had very, very basic stuff, one size fits all, or it was really hard to like build these like automation. So yeah, we just looked at like, I don't know, like messaging was just one media type right?

Andrew Bialecki: 10:14

To start with and off we go. And yeah, to your point, mean, there's many businesses, we think of them as consumer businesses that have, the number of employees to like the number of customers is this like massive differential. And you're like, so how do you solve that problem when you obviously can't have somebody call each one of those people? Well, that's somebody's gotta go build that software.

Wade Foster: 10:33

Got it. Alright. So I wanna switch gears. You did something that was pretty unique at the time, though we're starting to hear more and more folks kinda taking this approach. You all were at million in revenue, I think, before you ever hired anyone.

Wade Foster: 10:49

And you know, I think you rewind the clock, and basically nobody is doing that. But now in the era of AI, like, you hear all these teams bragging about how small their teams are and how big their their their revenue count is. Maybe two questions for you. Like, do you think if you had access to these tools, you would have gone even further before you hired somebody? And then maybe the second thing is, do you think that like, how big of a trend do you think this will become?

Andrew Bialecki: 11:20

Yeah, so our story on like why we didn't there's sort of the I'll give it two ways. Like one is I had a lot of respect for people that had built small businesses And even large, I mean even larger. Mean, go back into the seventies and eighties in the history of tech. I mean, those businesses were profitable like pretty soon out of the gate. It's only a really recent phenomenon, right, say in the last twenty years that there's this like, oh no, it's more, it's venture fueled growth, right, or capital fuels growth.

Andrew Bialecki: 11:47

There's nothing wrong with that model. Felt to us like it's like, well, if software is kind of like it's, I mean, especially compared to like physical companies, it's just, hey, there's probably relatively less you have to build. If you build a really good product, then people will spread it word-of-mouth, that reduces your sales and marketing costs. It just felt like it was possible. Anyways, I had a lot of respect for the kind of the tech greats, right?

Andrew Bialecki: 12:07

That's gonna start everything out in the 70s and 80s. And then my mom's side of the family just had a small business they'd run and run profitably. And I was like, well, why can't we do that? Like it feels like that's possible. So that's part of it.

Andrew Bialecki: 12:20

The other thing too was, remember that first summer, so I mentioned it's like, we were working out of this like, conference room at MIT. And I remember we went and tried to raise some capital for the summer, right? Because a bunch of the venture capitals as like a little bit kind of an early scalp kind of program, we're doing these things where they're like, Hey, we'll give you like 10 or 20 grands, right? 10 or $20,000 and you get free office space and we won't take any equity, right? And we thought that was amazing.

Andrew Bialecki: 12:46

Like, wait, I'm sure you remember in the early days, we're like 10 or $20,000. Do you know what we can do with that amount of amount of capital? Right. We're like, I could pay our server bill for like the whole year. Right.

Andrew Bialecki: 12:56

And I remember we did all these little pitch videos and pitches. And even though we had some customers and even though even sell a little bit of revenue, nobody thought we were doing was a good idea. Right? I mean, a little bit to your point of like, ah, this is a solved problem. Why are we doing this?

Andrew Bialecki: 13:07

Yeah. So I was like, well, geez, if we can't get the like easy chunk, you know, like chunk chunk of like, you know, no, no, no strings attached equity. How are we gonna go raise a million dollars or something? Right? So anyways, there are two major benefits of that.

Andrew Bialecki: 13:23

The first was we just work to automate everything. So you learn a lot about like what will scale and what won't. I mean, when you've got a 100 customers and there's two of you, I mean, we had a routine of every morning, we'd pop open our like shared inbox and we would work through support tickets, right? I mean, you remember these days. And you just, you'd work through them, right?

Andrew Bialecki: 13:44

And literally we were not allowed to start working on the new stuff, right, building a product until we finished. And you very quickly figured out what the common, so you'd write, I ended up writing like three or four docs in the morning or that afternoon, I'd be like, I gotta go fix this, you know, this feature, make it more tune up or something. So people will stop asking about the stuff just literally so we get more time back to build. Right. But it's had these wonderful things that we talked about being customer first at Klaviyo, but like just being tight with your customers.

Andrew Bialecki: 14:10

I mean, all of sudden, like people loved our stuff because we were so responsive in a way that like, I think it's just other companies just weren't. Right. So that was great. And then the other thing so it got us to be very customer first. And the other part was, as I said, automation anything that we could possibly eliminate as a job to be done.

Andrew Bialecki: 14:28

We had to do. It was existential for us. Right. So we eventually, you know, I remember we eventually, you know, the reason we raised capital was we got into this thing where, okay, so we got to, you know, we were generating enough revenue that we weren't taking a salary yet, but we wanted to, I was like, oh, it would be fun if we had other people to work with and like we could go faster, we could build more. But you know, I mean, we wanna give people, okay, people were joining Kladeo salary, right?

Andrew Bialecki: 14:53

So we were literally hiring people based on how much we grew the previous month. And I'm sure you remember like your bank account, it's like it would do the sawtooth thing where it's like, it would go up over the course of the month as you get all your little like payments and it would drop when you had to pay your run payroll. And we were just like, we just need a little bit more cushion than that. Anyway, so like, but And now it's your question of like, hey, what happens going forward? I'm curious what you think.

Andrew Bialecki: 15:19

I love the discipline, practically we probably could have started recruiting and hiring folks a little bit sooner, but I love the discipline to create on those two things. It forced this very customer first, you know, you have to know, you know your customers, you spend every day with them. You can't abstract that away. We built a better product as a result of that. And it forced this automate everything, anything that could be automated and it forces like great prioritization.

Andrew Bialecki: 15:46

I think most people like embracing constraints, is an action I just love. So, I think a lot of people are learning that now. It's like, hey, I don't have to do this. And then, you know, you know, I mean, we could talk about like the the other side of venture capital. I mean, I talked to a lot of folks.

Andrew Bialecki: 16:01

By the time we raised VC, the fact that we were profitable, we have to worry about a runway. People are like, wow, what did that feel like? And it was like, was great. We just never really worried about it. I think people are realizing now it's like, are actually some really nice attributes of a company and, you know, I don't know where the VC is not free, so to speak.

Wade Foster: 16:17

Yeah. I mean, there's I your your story, like, creates a couple memories for me. So, you know, we we had a similar mantra of, you know, don't hire till it hurts. You know, it was the three of us, and we would wake up in the morning, same thing, where we'd have email inbox. You know, we're working out of Gmail at the time.

Wade Foster: 16:36

And so there's a little bit of just like, do you have this ticket? Do you have this ticket? Because, like, you don't know who's replying to what. And so we had a definitely, like, a couple experiences where we're like, hey. Is there a product out there that, like, could help us do support?

Wade Foster: 16:47

It's like, we didn't even know what ticketing software was at the time. And it turns out, oh, yeah, of course, there's this whole category of like so we end up buying one. But we would work till I remember we were doing support until like 3PM in the afternoon. And eventually, were like, Okay, this is great because we're learning so much about our customers and what's a problem with this. But also, we're running out of hours in the day to go fix their problems.

Wade Foster: 17:12

So that was when we were like, Okay, we do need to make a hire. We were at the same mantra. We want to spend so much time just talking to the customers, automate as much as we can, and then go. And then the second thing, to your point of you know, where is this heading? Like, I I I believe that more companies could have and should have been built this way in the last decade.

Wade Foster: 17:35

But I just don't think it was like people knew how or that it was that popular. Like, most of the discourse out there is promoting building tech companies in the way that Silicon Valley and Venture Capital sort of promote them. They're like the biggest thought leaders and teachers of how companies get built. But it's very biased to that way of building. But the reality is, yo, you look at most b to b SaaS, and I'm kind of like, I think it could have been built this way.

Wade Foster: 18:05

It's like it doesn't seem like it had to have been built by throwing humans at the problem for everywhere. And I'm actually really excited to see AI kind of like I think it's sort of the two things. It's like the correction of the market where people have battle scars from that, combined with AI is creating this catalyst where folks are like, wouldn't it be cool if we had a lot of revenue and not that many people? And I'm like, yeah, that sounds pretty cool. A 100.

Wade Foster: 18:36

Speaking of AI, tell me, what was kind of your, like, you know, what was your sort of like moment with the advent of LLMs? Like, where did you start to, like, very much pay attention to what this trend was doing?

Andrew Bialecki: 18:52

Well, I remember when it's, I've been following along with the various iterations of like the GPTs, as they came out, and it's always, I think it's fun when you do tech and you can kinda see what technology's capable of, and if you're not literally on the front lines, like building them, you sort of don't know what happens next. Like sort of have we hit a plateau or does the curve keep bending? It's just hard to tell, right? So you have to sort of just wait and see. And so you just kinda keep evaluating as things come along.

Andrew Bialecki: 19:21

But I remember, obviously when chat GPT and, you GPT-three came out, you know, I was like, wow, this is, really interesting. I don't know, being a bit of, like, a math and science guy, like, yeah, really wanna understand how it works. And I remember getting Stephen Wolfram's book, like, you know, what is ChatGPT? Like, sort of gets into little bit, it's, the linear algebra, and, know, it was bringing back to the early CS days of like, know, it's obviously I was like, oh, okay. So it's like kind of a very good like Markov chain.

Andrew Bialecki: 19:45

Like it's very good at predicting like the next word. Right? And that was super interesting, but this concept, honestly, still am wrapping my head around of like believing in technology that has like emergent properties, right? Where it's good at things that you didn't necessarily program it or design it to be good at is fascinating. And so I think two things, this the idea of the scaling laws and now what we've been able to do with reasoning, plus this idea of like, you can kind of design a system to be really good at one thing and yet it'll just, I mean, kind of by magic, I think we're still understanding, like, the inner workings of how these, like, very large, like, neural networks work is it will, like, have these other properties, these other behaviors that you sort of couldn't have anticipated.

Andrew Bialecki: 20:29

Right? It's really incredible. I mean, it's almost, you know, a little bit of like a human and its ability to learn, yeah, you sort of like, oh, okay, because we're exposed to one thing, it gets good at something else, right? It's not quite the same, but like Anyway, so that was really interesting. I think also this idea of natural language as like the medium is really interesting.

Andrew Bialecki: 20:49

Know, somebody who's programmed and like, you know, I don't know, whatever, like his you know, sort of you learn to adopt like, you know, semicolons, semicolons, you know, parentheses, brackets, all this stuff as like just part of language. But the idea that you can just use natural language, I mean, also takes some real adjusting too. But it's very interesting that we sort of designed this as to be a like, yeah, something that humans can easily interoperate with. And now it's like, hey, you don't actually don't need to know, right, how to write Python or C plus plus or whatever. Anyway, so anytime there's tech that comes along like that, I think like the best thing you can do is just, you gotta go jump in the deep end and just like play with it, right?

Andrew Bialecki: 21:24

So I always try to do two things. One is I've spent some time like just building my own, obviously not JAPI G, GPT-four level kind of models, but like you gotta go see what it's like to actually build something. So I remember when the internet is like, okay, I'm gonna build my first website and that was all PHP based. And I just kinda understand how it works. So you can understand like where the edges are.

Andrew Bialecki: 21:42

Same with like kind of building a model, fine tuning one, you gotta get hands on that. And then you just gotta like play with it a lot. And that's been really interesting because I think everybody goes through the kind of learning curve of like, you try to replicate things that you already do, right? Basic use cases, like search, this kind of stuff. And then you start to give it some like higher order, maybe different problems.

Andrew Bialecki: 22:00

You're like, oh, I've never tried this with software before. And it's maybe not great, but you're like, oh, but maybe it could be. And I think that's really exciting. We're still like figuring all this stuff out. Maybe the other thing I'm curious about how you think about this.

Andrew Bialecki: 22:14

We've been spending a lot of time thinking about this in terms of what does it mean to use this form of AI, large language models, for application development. And one of the things that we talk about is this idea of software, whether it's, let's talk about SaaS, is largely about nouns and AI applications are largely about verbs. So where in the past we're used to building, like you think of like, oh, what's my data model? And I'm gonna sort of build tools that a human will use to manipulate their world. But I'm largely building like nouns, like their tools.

Andrew Bialecki: 22:48

They're things somebody's gonna hold, it's like a hammer or a screwdriver or something, and they're gonna use that to craft something they wanna craft. But now we're talking about like, well, level up a little bit. What is the actual thing, the job somebody's doing? And now we're not, you write things as jobs to be done as they're like, yeah, they're gonna use my tool to do the job to be done. It's like, what if you could just build software that does the job?

Andrew Bialecki: 23:08

Maybe it's human in the loop or it's whatever. I think that is like fascinating. And it's a very cool paradigm shift. And to be honest, like we're just practicing building agents and these verbs and It all feels very new, but I I think it's very very interesting to see where it goes.

Wade Foster: 23:25

Yeah. I think for for us, that that aspect didn't feel as new. But I think that's just an artifact because in some ways, like, Zap has always felt like

Andrew Bialecki: 23:36

is a verb.

Wade Foster: 23:37

Yeah, it's almost like a mini agent. It doesn't have any agentic behaviors, but it does work for you. And so in 2011, when we were starting this thing, we were talking about invisible software, like software that just works for you, and those same concepts. Though, of course, it was very deterministic and very workflow integration oriented. Didn't have any of the LLM properties that are pretty amazing today.

Wade Foster: 24:05

And it is like a different it's a different modality because you have to think about the business model differently. You're not selling a seat anymore. You're selling this unit of work. And it's like, well, how do you price a unit of work? It's just not as those things are not as obvious.

Wade Foster: 24:23

And I do think we're going to see a lot of innovation and experimentation on just even on that side of the house as technology evolves. Do you recall, like, what your sort of first, like, experience, like messing around with this stuff was? Do you have a use case that you were like, holy crap, this is

Andrew Bialecki: 24:42

Well, I'll give you one, like, kind of risk of outing a family member. So I'll give you a recent one. This is not like my first one, I'll you just a really cool example. So my sister's a bit younger than I am and in college and she's interning for the summer, right? And she's interning with the state government.

Andrew Bialecki: 25:02

And so anyways, one of the things that folks asked her is like, hey, you should go check this out is, so they're reviewing their kind of policy or what their position should be on like cell phone use in high schools, okay? So there's a meaty topic. And so the first part was like, hey, know, could you go do some research and just figure out what the other states do? Like, I mean, do they have a policy? Do they leave it to the cities?

Andrew Bialecki: 25:26

You know, it was just how do people think about this? And I remember it's funny, we were sitting down on a Sunday night and I like, oh, that's really interesting problem. Like why don't we just ask it to all of the LLMs, tell it to go off and do research and come back with an answer. And then like literally the next day, she's like, okay, I consolidate all those answers and like turned in this thing. And I think that was meant to be like the project for the summer, right?

Andrew Bialecki: 25:49

And we're like, so it's day like two or three here and we might've just done it, you know? So I'll probably over something like a little bit, but that is like incredible. Like that level of research that, I don't know, I even remember going up where we had like encyclopedias, right? Like, now it's like, that's just not a thing anymore. It's actually how fast can you consume that information and put it to work.

Wade Foster: 26:12

Yeah. So shifting gears, like, that that like, obviously, that, like, oh, that's a project for a summer turns into, like, literally, like, an hour of work or whatever. Walk through, like, how you have helped Klaviyo make this transition. Because, you know, you've got, you know, thousand employees, plus employees, like, working on this problem now. And, you know, there's a whole, like, recalibration of how you all have to work.

Wade Foster: 26:43

Every organization is going through this. And, you you know, walk through sort of how you've thought through, like, the change management, the staging, the, like, rein reinnovation with this new technology as, a core primitive of how we can do things.

Andrew Bialecki: 26:56

Yeah. Well okay. So let's talk about it from two lenses. One is, like, how how we work. Like, what does it look like for a company to you know, or a team or organization or whatever.

Andrew Bialecki: 27:06

What does this mean for them? And then what's the impact on customers? Like, hey, how does this help you level up your products? So internally, I'm actually a huge fan of the grid that you kind of made of like, hey, what does it mean to be good at this new thing? I think oftentimes these things can be very abstract and, you know, it's you sorta you wanna say it's like, yeah, want you to lean in and kind of explore and experiment, but that's that's kinda hard.

Andrew Bialecki: 27:28

Right? Because, you know, some folks are gonna say like, okay, that's awesome. Like, I'll go spend all day playing with this thing and kind of explore it. My style is like, you do some of those deep dives and also like, okay, let's just do like kind of a thirty minutes or an hour daily. Just force it into your routine a little bit so you can kind of get a feel for it.

Andrew Bialecki: 27:42

Right? And maybe it's in different context. And some folks will like, it'll give it a go and maybe they wanna have success, you know, or they don't won't see the value, right? Because it's raw. You and I were used to like, yeah, tech's not there, but like actually version 1.1 or two point zero, like it'll get there, right?

Andrew Bialecki: 27:57

So I think the first is like just to find like, hey, what does good look like here? I think also another prerequisite, you just have to kind of decide, like at some point it's like, I'm a big fan of like organizations that say like, look, believe in diversity on a lot of things, but there's some things we don't actually wanna be diverse So we talk about our values, right? Like we're gonna be ambitious, we're gonna be customer first. There actually, there's maybe different ways to do it, that's fine. But we all believe that this matters, right?

Andrew Bialecki: 28:27

And I think this is kind of one of those moments where, you know, it's kind of a fun moment because it's not obvious. I think there's some folks like, some non zero chance that this doesn't have as big of an impact as a lot of people think it is. You have to kind of just decide. And so our decision was like, no, this matters. And then therefore that's why I love what, be curious about how you guys kind of came up with your rubric, but we've been developing our own actually, thanks to seeing what you guys built.

Andrew Bialecki: 28:50

And I think then it's like actually pretty straightforward because then folks are like, okay, cool. Well, either, you talk about like folks that are proactive in their use of like integrating, whether it's LMs or other derived tools. Or I say it's like, hey, are you kind of like a no op? You say you wanna do it, but you're not actually doing anything. So you just have to decide, hey, this is the company we're gonna be.

Andrew Bialecki: 29:13

I think everybody, if you're just clear about that, they're into it. Then you have to a little bit like, hey, so what are some of the patterns? And like most things, you're gonna have the early adopters, you're gonna have some folks that are leaned into it, but they're just like sort of don't know where to aim And off you go. So we're well down that path now. And probably one of the most practical things we've done is we said like, hey, look, let's use this as a moment.

Andrew Bialecki: 29:33

There's probably some things we could have automated more of anyways, but let's use the case to re underwrite, in this regard of like, hey, what are the jobs we do? What are the kind of verbs, the actions we take? Like, let's just do an audit for each of us of what are the jobs we do in a given day, week, month, etcetera. And like, you tried to automate those or use more tech around them or like what elements, like what's possible today and what do think might be possible in the future? Let's run that exercise.

Andrew Bialecki: 29:56

And then the really hard, it's like, that's pretty straightforward. Then the really hard one is we then say like, okay, now I want you to run exercise like, so what are you gonna do with all that free time? And actually that's where we go a little bit back to sometimes it's doing more of, know, it's like, oh, I wanna build these next three things and now I can do it. But sometimes there's a little bit of like, I think of like the Brian Chesky Airbnb, like 11 star experience. Like what would be the massive plus up, So the way that you're This thing you're dying to do, but just felt like, oh, I'll never have time to get to it.

Andrew Bialecki: 30:26

That maybe you don't think about every day, but now you should. And then once you've got that, we try to chart a course. But yeah, so I mean, that's been pretty fun. I mean, like, and then you start, you know, you start to share the wins, you know, day over day, week over week, and to your point on, like, best practices. But, yeah, I'm curious.

Andrew Bialecki: 30:42

Have you guys thought about getting this more, like, in people's routines?

Wade Foster: 30:46

So there's, like, a very couple, like, tactical things that have worked well for us. The first was, like, hackathon. Just, hey, we're gonna pause the company for a week and just give people access to the tools and say, hey, go build stuff, and we'll do a demo day show and tell as part of that. And that sort of gave folks the space to actually go play around with it. And that was important because folks are often like they got their blinders on.

Wade Foster: 31:12

It's like, I got to deliver this thing for the road map for this month, or I got to hit this goal, or I got to close this deal, or what have you. And so just carving off space has been really valuable. And then we repeat those every six months or so because, to your point, the tech is a little bit novel in that in the past, I feel like if you tried software once and you decided you didn't like it, well, the software didn't really change much. It's fairly static. These LMs are different than that, where six months, the UI looks the same, but then you go work with it and you're like, woah, this thing is just better.

Wade Foster: 31:49

And so there's a little bit of retraining our own brains on how to work with these tools and recognize that they're more evolving than they are just like, here is your software. Do you like it or not? It's like, well, here it is at one year old. Wait until it's three, and then wait until it's five, and let's see how smart this thing is. And then, yeah, I think the last thing another thing we did that is just we have an all hands every week.

Wade Foster: 32:17

And so not every week, but probably once a month or so, just do show and tell. And that encourages folks to you know, we can show off, like, cool use cases. It, like, creates some social pressure where people, like, see other cool stuff, promotes knowledge sharing. So you get, like, a little bit of those things going on. I think the bigger opportunity, which we are we've had some success with, but, like, there's still a lot of areas where we're not, is how do you actually go reinvent team and organizational workflows?

Wade Foster: 32:49

You know, I think a lot of organizations have done a good job of putting AI into the hands of, like, the individuals. You know, you gotta chat GPT scripture string. Go figure out what you can go do. And so yeah. So, okay.

Wade Foster: 33:00

Great. I'm, you know, writing better emails or, like, my memo sound sharper or whatever, like, which is cool. Like, that that is helpful. But I think the real value is next, like, popping up a layer and be like, how do we, like, rethink how, like, lead generation works inside the company? Or how do we rethink how, like, employee onboarding and offboarding looks?

Wade Foster: 33:17

Or how does our, like, IT help desk get totally automated? These systems that, like, are big, like, just time sucks inside of organizations. And, you know, stepping back and saying like, hey, there's, like, a totally new way to, like, do this stuff. And it takes it's not just the technology that has to get implemented. It's like you have to revamp the team around the technology as well too.

Wade Foster: 33:42

And so they're just a little harder to to get after. And so we've had a few successes around that. But those, yeah. They just take more effort, I think.

Andrew Bialecki: 33:49

Yeah. This you know, like all technology, I think, is we're kind of on this ever hopefully accelerating curve. And hopefully, it's, you know, net I I think it's generally it's a net positive to the world. But this idea of, like, being able to absorb more change, change management, don't know, kind of a quip internally that's like, we don't like change management. It sort of implies there's this big long process.

Andrew Bialecki: 34:12

It's gonna take a long time. And it's like, yeah, actually think a lot of people can just, they can roll the punches as long as things are like clear and they're willing. So you sort of just have to say like, yep, we're doing this. And like actually, hey, the timeline isn't months. It's like, you have like a day, you have a week, right?

Andrew Bialecki: 34:26

So figure it out and go. We've actually inside inside ClickUp of over the years, we try to embed this in ways that are like somewhat artificial. Like one simple one that we've done for a while or to bring back to our boss office is like literally just maybe we'll change their desks every three months. Right?

Wade Foster: 34:41

Oh yeah.

Andrew Bialecki: 34:41

Why? You meet other people, like you get a window seat or you get a different view of stuff, but whatever, at

Wade Foster: 34:49

least you just have to change.

Andrew Bialecki: 34:50

And it forces this kind of like, okay, change is gonna be a constant. Because yeah, to your point, there's gonna be a tax of learning because the tools aren't all there, but you sort of want to get the most experienced person. It's like learning to use under Google search and like, yeah, if you played with it a bunch and you know how to use some of the keywords, all this kind of stuff, you're just better as it got better. So I think the more you can do that with AI and just like, then make it fun. I think a lot of things that kills most organizations, it's just inertia, it's like the way you're doing things.

Andrew Bialecki: 35:19

So if you can say like, yeah, yeah, let's zero based budgeting thing, let's just totally rewrite it, re underwrite it. What does that look like? And like, hey, let's just go explore that a little bit. Think hackathons are great way to that. Hey, here we're gonna carve out some time to do that.

Andrew Bialecki: 35:32

That's the best you can do.

Wade Foster: 35:33

Yeah. Are there novel tactics that have really worked for you all on leveling up the internal usage of LLMs?

Andrew Bialecki: 35:41

So I'll say maybe a couple of things. One is, it reminds me of like the, I don't know, the Steve Ae, like Amazon memo about API. And like, I sort of have to play with them, like services, like figure out where they break. So a couple of things that the use case have been great. What we found is some are global, they cut across everything.

Andrew Bialecki: 36:00

And those are some of the more general purpose like LMs. But a lot of it is frankly, is very functional. So it's even like techniques on, hey, how to use your favorite coding assistant, pod, cursor? And there's like a lot of technique there that we've sort of developed. Then you can develop some tooling on top of that.

Andrew Bialecki: 36:18

The other thing that's interesting is we've been kind of grappling with is this idea of how much do you explore versus you start to standardize? The same way that we largely write all of our code and like there's maybe a little bit of C, there's a lot of Python, a little bit of Java, a bit of Rust and that's just what we use, right? So it's not everything. I think the same is true for a little bit. Think LM is like, okay, there's another way to code and languages and then these tools.

Andrew Bialecki: 36:43

Like how do you decide what to standardize on is a really interesting kind of problem. And then also like the data regimes are really interesting of like, hey, how do you think about privacy and how you wanna use it. So that's maybe less sexy stuff. But actually just even the process by which we're like, hey, here's how we go explore new tools and make that easy to do without putting a lot of red tape there. But also think through some of these things of like, I don't know that we want, you know, five variations on like the, you know, creative generation for like some of content we create or maybe we do because we're at that point where we don't really know what's, you know, what's good.

Andrew Bialecki: 37:20

I don't It was a little bit like how you evaluate technology and software. Like everybody has to do it kind of unwrap, you know, very rapidly because the stuff is literally changing, you know, whatever, month over month.

Wade Foster: 37:29

Yeah. I wanna come back to, like, this, you know, this change management org resiliency thing. One of the things I I I caught you say once is like, hey. I I'm hiring for, like, high slope individuals and, like, wanting these folks that can just, like, learn and, like, kinda go on the journey with you all. Has has, like I I I imagine the answer is still yes, but has AI changed who is able to do that or, like, made certain folks better than others?

Wade Foster: 38:01

Like, how do you sort of make that assertion? Because, yeah.

Andrew Bialecki: 38:04

I mean,

Wade Foster: 38:05

it's just interesting. Like, I have folks who, like, you know, I don't know, who've been here for a while who, like, pre AI have just been great, great, great, great, great. And now I'm sort of like, oh, like, are you gonna are you gonna make it? And, like, I prior would have said, like, yeah, you actually are going on a pretty high slope. And so I'm, yeah, I'm just kinda curious how you've sort of, like, changed, like, your assessment of talent.

Wade Foster: 38:24

Or if it hasn't changed and it's just, hey, the AI is just, like, a just the newest thing to to assess against.

Andrew Bialecki: 38:29

Yeah. Well, I think we you know, tech is you know, there there are very few moats in tech. It changes so fast. You know, one of my pet peeves is folks who do like LTV where they say, oh yeah, my customers never churned. They're gonna be here for twenty years.

Andrew Bialecki: 38:45

I'm like, there's no way in twenty years, and the technology is gonna be all changed. Maybe they'll be there because you're great at working with them, like, come on, that's a crazy assumption, right? So we just work in an area where it's like stuff. I mean, the fun part is it moves very fast. The hard part is it moves very fast.

Andrew Bialecki: 38:59

Slope, I've always had this like, the math background is like, yeah, it's like, look, it's a slope. It's not your like, whatever wider steps, And off you go. The fun part is actually what I found empirically is a lot of people that have a very like, they're very high in their skills right now. It tends to be because they're very high slope. I think that, anyways, so that tends to be true.

Andrew Bialecki: 39:20

So no, I don't think that's changed at all. Actually, I'll give you the thing that I tell myself and have told pretty much everybody at Klaviyo except for some of the interns we have this summer, is I said like look, if you grew up in the era of tech in the 2010s, I actually think you've got a big bias that's gonna work against you. Because you're used to working a certain way and you've probably had a fair amount of success doing that and you need to re underwrite everything about how you think, how you work. And this seems kind of obvious because I don't know, when we were coming of age, it's like, yeah, I don't know, I remember working on early web apps, or you're working with like early versions of Internet Explorer, like JavaScript barely works. There's alerts everywhere to try to even build these things.

Andrew Bialecki: 40:01

They're so hard. And if you're building, yeah, I'm gonna buy the thing or build the thing that you install on your Windows machine or your Mac, you're fine. Mean, you're like, yeah, that's just the way we do it, right?

Wade Foster: 40:11

You're comfortable.

Andrew Bialecki: 40:13

For us, you're like, well, come on, like the internet's gonna be a thing and we're gonna be new to this. And you're like, you're gonna run loose circles around everybody else because they're just like, oh, they're just stuck in their old ways. Well now like shoes are kind on the other foot, right? I mean, it's happened a little bit with mobile, but now it's really, I think happening again. And so I thought, said, look, I actually think I have more, well, I have a lot of comments to anybody that has a high degree of wanting to learn, but you've been successful.

Andrew Bialecki: 40:36

So you have this kind of inertia bias. By the way, learning is hard. It's just tiring. You know, it's like it's just a lot of effort. So like, you know, at some point you're maybe running it like gas in the tanks.

Andrew Bialecki: 40:45

You have to constantly refill that. I told, know, the folks that are interning, I'm like, I think you have a big competitive advantage over everybody at Klaviyo today because you don't have any of these biases of like the way things work. And that's what made, I think, you know, a lot of the Internet companies successful. And so now actually, I think a lot of it is like, there's kind of this race. Right?

Andrew Bialecki: 41:05

Like, if you're in school or you're, you know, just you're graduating school, then you're like, look, there may be some things, some like kind of fundamentals that are not gonna change that just need to learn, right? Like, know, how to work with customers, how to build great scalable software, all this kind of stuff, right? I think you still need a CS degree even if we're all using like, whatever, Claude, Cursor, etcetera. But you just need to upscale those skills, but like you're gonna have this great thing of like, you're just used to a different way. You don't have all these like baggage versus you and I, it's like, well, we have to like un rewire our brains to be like, okay, is the way I did it.

Andrew Bialecki: 41:38

I literally when I sit down and code, I'm like, okay, I gotta like force myself to maybe start with a prompt versus just getting at it in the editor. And you gotta rewire and it's like, yeah, sure, there's some things I know, but like, and it's not obvious to me who's gonna win that race, right? And I think history has shown us oftentimes the folks that don't have that baggage that it's just, it's a really hard thing to overcome. So anyways, I think slope matters a lot. Curiosity matters a lot here, right?

Andrew Bialecki: 42:03

Very related. And I think the most curious teams are gonna win. I think you just, I don't know, we just say like, we just select for that. You just have to be curious about the world. You could be an A plus at your current job.

Andrew Bialecki: 42:14

I'm like, I just don't know that job's gonna be the same. It's gonna exist in five years, especially now with, you know, I don't know, whatever, some of

Wade Foster: 42:19

the turnover. So that that creates a like I I I would agree with sort of that assertion. Maybe an interesting question then is, how does the Klaviyo of today that maybe has all these advantage, you have all these customers, all this revenue, all this stuff, but you maybe have some biases from how the past works, compete with you know, if you were to found Klaviyo today and you have a blank slate, but you don't have any of the customers, don't any of that stuff. So, like, how would you think about, like, you know, in your shoes, like, oh, I'm gonna outpace this other one who maybe gets a clean clean slate?

Andrew Bialecki: 42:53

Yeah. Well, there's two things. So first from like a product perspective and how you work with customers perspective, we've actually spun up teams where I'm like, look, imagine you couldn't actually work with inside of our walls, right? You have access to our APIs. By the way, have access to any other company's APIs if you want them.

Andrew Bialecki: 43:11

Just how would you build the next generation? I'm very big believer for every product, there will be a version N plus one. The only question is like, what does it look like? And that takes a lot of judgment and taste and like being in that domain, but there will always be that N plus one. So now I think it's wide open.

Andrew Bialecki: 43:27

You have to wipe everything away. What would you build, right? So for us actually something that we're very interested in is to this kind of nouns versus verbs is like, I think we've spent a lot of time building the infrastructure to say, store everything you know about somebody, use that in a couple of different media formats, right? Email, text messaging, etcetera. And now we're expanding that, but you still have to do it.

Andrew Bialecki: 43:51

So you still need somebody to sit there and press the buttons. And we get feedback all the time from our customers. They're like, yeah, there's these things I wish I could build. Either, maybe I don't know how to, I have to watch some videos or it's just a lot of clicking. And you're like, well, what if we could just do it for you and you review the end result, right?

Andrew Bialecki: 44:07

You're like, well, that'd be amazing. Like, so when does that ship, right? So we're like, oh, actually we're shipping some that right now, right? You know, the last couple months, this year. You know, I think the same is true for Zapier.

Andrew Bialecki: 44:16

It's like, hey, there's probably these ideas that somebody just, I didn't have the time to explore that, play with it. That's a lot of work. And you're like, yeah, look, we can do that in a way that feels very safe. Right? So I think you have to just rethink it, you know, kind of from the outside in.

Andrew Bialecki: 44:29

And I actually think it's like for us, it's this layer of autonomy that somebody is gonna build. And like, obviously we're working on that. And then the infrastructure behind it, it's like, yeah, I I were gonna try to disrupt us, I'd be like, oh, well just treat the infrastructure as a commodity and assume someday you could go backfill that. So anyways, I think that's an interesting way to look at rebuilding the product. The other thing too is we have a saying internally is, hey, we're 1% done.

Andrew Bialecki: 44:57

I don't remember if like the old floppy disks or CDs, you'd install software. I'll be telling you like, look, if you're installing something on your computer or your phone or whatever, right? And it's like, those progress bars just started, right? I mean, you don't mind if you lose power or the internet flakes or whatever, it's just like, ah, whatever, wasn't that far along anyways, it doesn't matter, But when that bar gets to 50%, 60%, 90%, you start to be like, I don't want anything. Don't touch anything.

Andrew Bialecki: 45:22

Like we're on the finish line here, right? And I think that's like, that's a really, you don't wanna be in that position, I think in life ever. Cause it makes you very risk averse. You feel like you're trying to protect the thing. So we just have this point of view that you want, you just, you wanna work around folks and you wanna have your own personal ambitions be like, hey, what does a 100 X where I am look like?

Andrew Bialecki: 45:42

And always have the mentality that you're going after that. And then it's like, yeah, then everything doesn't feel so precious, right? So you're not so protective, risk averse, right? And obviously you wanna do, at our scale, you have to focus on reliability, it matters because like it matters to our customers. But you wanna say, like, yeah.

Andrew Bialecki: 45:58

But what is what does it look like if this was a 100 times better? And just that just forces a different way of thinking.

Wade Foster: 46:06

So we've mostly been talking about, like, the entire internal transformation around AI, but I do wanna shift a little bit to talk about the customer side of this house because you do have thousands of customers, small businesses that are adopting Klaviyo for AI. Where where do you today see the biggest opportunity that most small businesses are sleeping on? What are the use cases that your best customers are doing that you're like, hey, small business out there, you should be implementing AI in this, this, and this way, and you'll immediately just have better results for whatever it is that you're doing.

Andrew Bialecki: 46:39

Yeah. So there's two things for any, I think, business of any size. Certainly, I mean, it's very relevant for businesses that for entrepreneurs that are doing nights and weekends, smaller businesses where, like, I just feel strapped for time. I could figure this out, but just abstract for time. But also scales to a lot of the much larger businesses.

Andrew Bialecki: 46:55

We work with big CPG companies, all the iconic consumer brands. So two things to talk about. One, if you are a business that's like one of these like one to large end, you know, you have a lot of customers, consumers, users, you should be providing, I think in the very near future, you're gonna provide every single one of them with their kind of own, their own concierge, their own personal like, you know, agent, but like let's call it concierge. And I know it's a bit of a dated term, but you think about that, you know, I remember in the, you know, nineties, early two thousands, you go to hotel and you're like, where should I eat? Or like, you know, how do I get to this place?

Andrew Bialecki: 47:27

And you go to this person and they just know the answer. And they're not just there for supports, they're also there to like give you recommendations. Maybe, hey, I see that you're, you know, maybe you're, I know you're looking at this and maybe you're looking for Italian place, can I make some recommendations for you? Like they're sort of proactive as well. Every business is gonna have the ability to offer that to every single one of their customers.

Andrew Bialecki: 47:47

And it's quite the like mindset shift because, you know, we're so used to it. It's like, well, come on. I mean, no person, I can't scale that. Like, what is that? What is that?

Andrew Bialecki: 47:56

Are they offering? What services they offer? Right? What questions can they answer? What's their personality?

Andrew Bialecki: 48:01

So I think this is just coming and it's gonna be very multimodal, not just in the sense of like text and images, but literally it's like, it's gonna be available. Sure, we're using this to power messaging through Klaviyo, but also it's gonna be on your website, it's gonna be your mobile app. By the way, I think in the future it's like it's gonna be like in your ear, you know, say like the analogy, one of the examples we use is like, hey, when are like the national park system gonna give you like your own guide so when you go visit a state park, you're not with a map, you're with a person who's gonna like talk you through and be your like your trail guide for the day. And it's just like, yeah, because you pop in your butt and boom, off you go, right? So one, I think this is coming.

Andrew Bialecki: 48:38

And we're very hard at work at building this, both like this kind of like central intelligence, as well as the like, okay, how do we get this into all the various form factors? And actually this is where we're trying to list a lot of other companies we like to partner up with that are building some of the end, hey, where would you inject this intelligence if you had it? So that's one. The second thing that I think a lot of businesses should lean into, I think it's a little bit easier now with LMs, is giving their business and brand some personality. Know, a lot of us are just naturally, we're not like the life of the party.

Andrew Bialecki: 49:10

We don't wanna stand out. And that's like a very, we're little more, we're biased, we're introverted. The reality is, is like people, for the businesses and the brands that they interact with, they actually kind of want them to have like, be like one of your friends that's like, yeah, they're a little different, right? You know, it's obviously we encourage, you people to be like, yeah, you like you should be yourself, but then a company I think too should be itself. And sometimes I think especially as companies, you know, either they don't have time to do it or they think they're supposed to act a certain way.

Andrew Bialecki: 49:39

One of my pet peeves is I think there's a lot of people that they get a job and maybe they're in their 20s or something. I feel like they've like watched The Office or something. They have a way they think work is supposed to be and they pretend that that's the way it is. And it's like, where did you learn that? Like not in school or anything, and I don't think you actually had a job, you think that's the way you're supposed to act, but it's like overly formal and kind of, it's just not very interesting.

Andrew Bialecki: 49:58

So anyways, I think a lot of brands want to have more personality and inject that into all these different experiences. And we've obviously seen a lot of folks that differentiate on that, but it's hard to do. And I think that's another thing that's just gonna be very possible. I think, I don't know, you and I probably have a thing of like, every customer we talk to, they're gonna know that they talk to like Wade or Andrew. Like they're gonna be like, oh, I'm gonna remember that conversation.

Andrew Bialecki: 50:22

It's like, yeah, why can't that exist, you know, times, I don't know, a thousand or a million or shoot, the nice part about a company is there's some shared set of, you know, I don't know, values or attributes that they have, but there's also, there's a lot of, you know, different approaches. And like, why don't we scale like the best amongst those or let people match up with like the I think this is gonna be a lot more doable. And if you're gonna People should want that.

Wade Foster: 50:44

Yeah. I think you're onto something here. Like, I feel like you're already starting to see some of these really large content creators get really good at that, where they have an audience. There's people feel a certain affinity towards them. They like them for x, y, z reason.

Wade Foster: 51:01

And now these folks are going out and making chatbots of themselves and clones of themselves. And now it's like, oh, I can kinda talk to this person a little bit. It's not quite the same, but it it it sorta has the same impact where it's like, oh, I I know this person because I listen to them every week or I watch their, you know, their reels or whatever on Instagram. And so I say, I feel like I I feel like I know them. I feel like I'm friends with them.

Wade Foster: 51:23

And if you're able to, like, actually make that feel like more of a one to one connection, I think that's that's pretty awesome. And to your point, like, most major corporations simply just haven't had the tools to do that. It's like they they sort of feel like big bureaucracies rather than, you know, your local corner store. And if you can make tools that make your you know, a big company feel like your local corner store, like, that that's pretty awesome.

Andrew Bialecki: 51:45

I'll give you one example. They they just made me think of the thing that we've done internally. These are like, they're kind of not very sexy tools, but like one of the things we've built a bunch of is a couple of applications that will like help people give feedback to each other. And so, okay, whatever. There's a lot of that's like a blank canvas problem.

Andrew Bialecki: 52:02

So actually some things we've done is like, oh, can we do like voice to text transcription? Because when you do voice, you sort of just have to say it, right? And it's a little easier than like trying to type the perfect sense or whatever. So that's fine. So you do some text or video or voice to text transcription.

Andrew Bialecki: 52:15

But one of the features we built into some of those tools is actually having it sort of edit down your transcript, but give it a personality. And it's fascinating because I actually one of my tweaks about the kind of the default LLMs is probably makes sense, but like they do a good job of managing up, I think. Sort of, they kind of tell you what you want. They're not very harsh with feedback. And so we've got a lot of folks like, no, just give it to me raw, like tell it to me straight.

Wade Foster: 52:43

Yeah, yeah.

Andrew Bialecki: 52:44

What's funny is we've literally in this tool built this little dropdown that's like, how do you want it, right? Like, if you want drill sergeant level, butt kicker or do want to like, hey, just I'm not in a place for that. Just be a little bit nicer about it. And it's fascinating actually watching what sort of, you know, personality type people choose, right? When it's like, Hey, take this thing, you know, wait, it has to be back for me.

Andrew Bialecki: 53:06

Okay. But translate it into this, you know, falling persona.

Wade Foster: 53:10

That's awesome. I actually wanted to build this, the last performance review cycle, because I was like, oh, I should make, like, characters where it's like, you want your performance feedback from, like, Bobby Knight? Okay. Like, you're gonna get it this way. Versus like, do you want it from, yeah, I don't know, Leslie Knope?

Wade Foster: 53:22

You're gonna get it a different way. Yes. And it's but it's the same content. It's just delivered different stylistically. I always thought that I was like, I thought that's, like, a pretty helpful thing based on just who are you as a person.

Andrew Bialecki: 53:34

Yeah. So

Wade Foster: 53:35

anyway It's been good because

Andrew Bialecki: 53:36

it just encourages more frequent stuff. And, you know, like, it's, serious I mean, stuff, like at least people have, get to have a little bit of fun with it. And then yeah, anyways, so it's good.

Wade Foster: 53:43

I love it. Well, we're running up on time. I could have spent another hour jamming with you on this stuff, Andrew. There's so much, happening and stories to share and stuff to learn, but appreciate you joining us here on, agents of scale. And, until next time, folks.