Weights & Biases: Following the "Gradient of Admiration" to Million-Dollar Deals with Lukas Biewald

Host: Foreign. Hey everyone, thanks for joining us. For another founder led sales stories where founders who have successfully navigated their founder led selling journey share with those who are still in the middle of it. I'm Pete Kazanji, author of Founding Sales the Startup Sales Handbook and your host. Today we have Lucas Bwald, co founder of Weights and Biases, the AI developer platform that was recently acquired by Core Weaver. Lucas and his co founder Chris Van Pelt built weights and biases from a simple web hosting experiment tracker to a multimillion dollar platform used by AI engineers across industries like automotive, finance, healthcare and agriculture. I'm particularly excited to have Lucas here today because while everyone knows weights and biases in AI now, when they started out back in 2018, that wasn't necessarily the case and that required quite a bit of founder led sales to start. But plus there's not a ton of content available about developer sales motions and certainly not enough about AI research and uh, ML engineering sales motions. My biggest takeaways from our conversation are Lukas's insights on what he calls the gradient of admiration. Focusing on winning over the most respected engineers. First, their clever open source integration strategy that put their product exactly where developers needed it, and how they overcame pricing resistance when selling premium priced tools trust to traditionally price sensitive developers. You'll all hear about how they transformed from closing thousand dollars deals to landing a hundred thousand dollar contract with Fortune 500 companies, all while um, maintaining their focus on product quality over marketing hype. So with that, here's Lucas. Hey everyone, thanks for joining us. For another founder led sales stories where founders who have successfully navigated their founder led selling journey share with those who are still in the middle of it. We're here with Lukas Biewald, co founder and CEO of Weights and Biases, the AI developer platform and now a core weave company. I'm excited to have Lucas here today because while everyone knows weights and biases and AI now, when they started out back in 2018, that was not necessarily the case and so that required quite, uh, a bit of founder led sales to start. Secondarily, uh, there's not that much content available about developer sales motions and there's certainly not enough content about AI research search, ML engineering sales motion. So I'm eager to get some of the gems on the record here. Uh, Weights and Biases was acquired by coreweave a few months ago and ahead of that had a few dozen sellers. But I'm particularly excited to hear about what things were like super early. So Lucas, I'm really glad to have you here. How Are you doing?

Lukas Biewald: Honored to be here, Pete.

Host: Honored.

Lukas Biewald: Ha.

Host: Um, so. So for folks who maybe are less familiar, um, maybe you can, uh, share a little bit about what, uh, problem weights and biases helps solve and what the prototypical organization type is, and then who the humans are in the organizations, um, that care to solve that problem.

Lukas Biewald: Totally. Um, so we try to make AI developers, which is an increasingly broad term.

Host: Used to be more, you mean me?

Lukas Biewald: Vibe, Coders at this point, everyone. Um, but it kind of started with trying to make AI researchers successful. So the company actually wasn't started with a particular. It started with kind of like a Persona in mind. And, um, and we actually tried a couple different things, but the first thing that we did that really took off, um, was this thing called experiment tracking, which actually, whenever I talk to sellers, I always say, you know, this is kind of like salesforce for, um, your, your AI stuff, right? Because it's like you kind of have the central dashboard where you see like, you know, some charts and graphs and some individual examples of all the things that are happening, like all the, you know, models you're training, how they're doing, and then just like, how occasionally you want to go back in time and like, see what happened a couple months ago. You might want to look at that, or you might want to look at, you know, all the things that you've tried over the last couple months and see how things are trending. So, um, kind of like a central dashboard for all the things that are going on in your world of building and deploying AI models.

Host: Nice. And so, um, I kind of like that because I think oftentimes one of the things that founders can get, um, wrapped around the axle around is one they can think that there's a particular problem that needs to be solved. Which is, which is not terrible. Right? Like, you know, one of the things you want to say or folks say is, you know, fall in love with the problem versus, like, the proposed solution. The worst, of course, worst possible scenario can be like a hammer looking for a nail, where you're just like, hey, I have this really clever idea and it's totally right. Um, but I like how you framed that there where it's like, hey, here's this Persona. We know it's going to become more and more important. They're going to have needs. We have intimacy with that, ah, Persona, especially from your time at Crowdflower and figure, uh, eight. And we are best positioned to handle whatever those needs are using magical software. Um, I, I love that. And then, and then so early On. So it sounds like that was the first thing that was successful. What was like the average deal size for those, those transactions? Um, you know, starting out like, you know, it's funny, four years ago, five

Lukas Biewald: years ago, like median deal size, probably depending on who you ask, it's probably somewhere between 25 and 50k. And I think it's gone up a little bit. But honestly that's been like pretty consistent since the beginning. Like it's not like our first deals were there, but you know, probably like our fifth or sixth deal was like 100k. And you know, I would say it's always been tens of thousands of dollars typically. Um, yeah, yeah.

Host: And, and so maybe you can share with, I mean I hinted at it a little bit, but maybe you can share with folks what your background was before, um, starting weights and biases and generally like your um, you kind of like your professional background because I mean it wasn't as a CRO at a public company. How about that?

Lukas Biewald: Yeah. You know, so I wanted to be like an AI researcher. Turned down an offer from Google to work at Yahoo back in two, uh, thousand five. So I'm a genius. Yeah. And then I basically worked as like an ML engineer for a little while before starting Crowdflyer, which is like a data labeling company that changed the same to figure eight and, and then sold and then I started. Weights and biases.

Host: Yep. So prototypical uh, engineering background, but in this case like particularly with a uh, ML AI bent back before it was hip.

Lukas Biewald: Hell yeah.

Host: And, and so you know, you kind of mentioned this earlier, but you started out, the initial hypothesis was hey, this Persona is going to be important. Right. They're going to have uh, they're going to have needs. Uh, how did you kind of hone in on the um, specific kind of like problem case that ended up being that first use case that really hit um, like was it customer interviews or. How did you guys go about that?

Lukas Biewald: Well, I'll tell you the real story. It's a little complicated actually. Um, great. Either. So, um, actually it was towards the end of running um, figure eight actually my, my board had fired me, um, at that point, so I was like a little bit idle and um, and I actually was trying to m. Afford a house in uh, San Francisco. I was newly uh, married. I didn't have a lot of money and I was actually feeling, I feel pretty bad, you know, that I, I didn't um, have enough money for a down payment on even the most basic San, uh, Francisco house. I um. It's not small it's not small. Yeah. Yeah. I started teaching, um, AI classes on the side, um, to engineers to, um, make. Honestly, to, like, make extra money. It was like, you know, I love it. And, um, and that was interesting because it was actually kind of like, you know, in teaching people, you sort of see what happens when you get started with AI. Like, you know, you could build a model, but then it was a little bit hard to actually even see what it was doing. Um, and even to show your model to other people in the class was kind of hard in any tangible way. A lot of these models, you could build a face detection algorithm, but then it was actually a lot of work to even show. To show a picture of a face and see if it detects a face is actually a little bit of a different thing. Um, that was one. And then also I went to OpenAI, and I was like, hey, can I be an unpaid Internet? Um, and the company was just chaotic enough that they kind of let me hang out there, um, and work as an unpaid intern. Um, so.

Host: Great.

Lukas Biewald: And so that also was sort of a similar thing where I was kind of like, I was kind of doing stuff like I was trying to help them build a system to pick up objects. And so it's kind of like running into a lot of issues. I mean, there was no shortage of issues. I mean, the problem back then was not that people didn't think there were issues. They just thought the market was too small to back a company in it. So you just actually didn't have any companies trying to make tools. So it was a funny moment where there was plenty of, um, obvious, uh, pain points, but problems, um, no interest. Yeah, exactly. And so, um, it's funny. We started with one that, um, people really, really thought was niche, even the AI researchers thought was too niche. Um, um, which was like, I don't

Host: know, Lucas, I only have this problem once a quarter.

Lukas Biewald: Yeah.

Host: But it's a huge problem.

Lukas Biewald: Well, it's like. It just seems so basic. So it's basically like you just. You just want, like, an automatic graph of, like, how well your model's doing. So it's literally just like, um, you know, you just want to take. You know, your models are saying, like, hey, I'm doing this. Well, at each, like, kind of step, they look at a little more data, they get a little better, a little worse, and then they, you know, you just want. I just wanted a graph of, like, okay, is my model getting better or worse? You know? And, um, so it's like this little logger was Kind of the first, um, iteration. And the funny thing is, actually there existed a logger at the time that worked really well, but it had this flaw, which was that it wasn't hosted on the web. So you'd have to run on your laptop. So if you want to share it with your boss or your colleague, you'd have to take a screenshot of it. And so we basically made a simple hosted version of that so that you could just share the, um, that graph with your. With your colleagues.

Host: That is a awesome minimum, Minimum viable product. And like.

Lukas Biewald: Yeah, yeah.

Host: I mean, but come on. I think that that's where a lot of times, you know, like, Segment, supposedly the progenitor of the, uh, customer data platform Segment just started out.

Lukas Biewald: That is a true story. I remember when I first saw Segment, I was like, this is really useful. And this. There's no way this could be a company. So I actually, I use that. That, like, Segment when it was one.

Host: Yeah, yeah, just like, you know, event, event tracking, event logging, abstraction. Anyway, like.

Lukas Biewald: Yeah, totally.

Host: You'd be shocked. These. This is a very common thing that shows up in, in these conver. In these conversations. And I think the important thing is, is that, like, you're in an area that has substantial growth in it, and. And then you just have, you know, the option ball control on. On all the incremental, uh, use cases. Use cases there. Uh, I also think that when you described about being so close to the metal or you were doing it like. I was a teacher. Kind of a AI Ops manager. Right. If you will. Or like an AI manager of, like, a bunch of.

Lukas Biewald: Definitely. I was an ic.

Host: No, no, no. I mean, when you were teaching.

Lukas Biewald: Oh, oh, yeah, yeah.

Host: Or like you were an erstwhile manager of, of your. Of your, uh, like, of your pupils.

Lukas Biewald: Well, the amazing thing about teaching, actually is that you watch like a hundred people on board at once, and you watch it under extreme duress, where if, like, anybody fucks up, they're like, really upset, you know, because they paid like 30 bucks for your class and now they can't, like, do the class, you know, so you really, like, want them all to, like, onboard successfully so that. Actually, I think one of the big keys to weights and biases was that the onboarding was good in, like, lots of environments. And that was totally because we just had 100 people onboarding these cohorts where I was like, terrified that they wouldn't be successful.

Host: Phenomenal. But I think a, uh, constant theme is being extraordinarily close to the metal and seeing that, and then it gives you the opportunity to identify those narrow, white hot. Like, it might be narrow. It may not seem like a giant tamp, but boy howdy is a huge pain in the ass right here, right, right now. Um, great story on validating that hypothesis. So how did you get someone to pay for it? And who was the. Who are the first customers?

Lukas Biewald: Well, first we tried to get people to use it, and I mean, we could like, force people to use it in the class, but actually getting people to use it for work, I mean, I think with dev tools, it really is a challenge to get people to use it. Um, we literally would go to people and we'd beg them to try it, and then we would set up, um, weekly meetings. I mean, we set up like, weekly meetings with them just so we could, like, see them not use it. I remember this guy at Toyota's, guy Adrian, and he was just sort of like, I don't know if this is, like, useful, but I'm, like, really interested in, like, trying new stuff. And I was like, okay, like, let's set it up. And then we didn't get a step in the first meeting, but we had like a next meeting schedule. And then we like, you know, got it set up and then the next meeting, you know, he's like, okay, I haven't really used it. And we're like, well, like, why not? You know, and he has some, like, BS reasons. And it's like, okay, let's like, we'll keep setting you up. Let's see if we can, um, get you using it. So there are a couple. I mean, I actually think we probably tried a lot of people that sort of bounced out. Um, yeah, I mean, we got, we got a handful of people using it before we ask them for, um, for money. Yeah, yeah, yeah.

Host: I mean, in the kind of like this, the staircase stepwise, uh, progression of founder led kind, um, of selling. One of the things that I tell folks that they need to validate initially is one like, is there a problem to solve that people agree with? Because if you're solving a problem that nobody has, that's a great way to make sure. That's a great way to waste a lot of time and money. And then the second thing is, did the magic beans that you created actually deliver utility? And if people aren't going to use it, then it's highly unlikely that the magic beans are going to deliver utility because you need to create value before you capture it. Now, of course, there's like a loop there where if the pain is sufficient magnitude, then you can get people to pay for first skin in the game. Right. Then oftentimes can compel that, et cetera. Um, but of course, doing that still means that you have to validate that utility is being created, otherwise you're going to have a big pile of churn on your hands later on. And also, I think in, uh, dev tools, one of the big things there, too, is that engineers are fussy about software.

Lukas Biewald: Totally.

Host: So, um, if they don't, like, if you don't, if there's, you know, if there's rough edges or stupid shit, they're going to be highly intolerant of that. Um, which I think, you know, the approach that you're describing can kind of help with that. So that was. So the first thing was getting people to use it. And then we had some initial customers, and I think what you said there was that early on, a bunch of those customers actually were in the tens of, you know, tens of thousands of dollars. What. Do you remember who, like, the first couple, like, big, bigish customers, like actual big kid customers were.

Lukas Biewald: I remember actually the first big one was, um, Toyota. So we, I think we got a couple customers at like a thousand bucks or 2,000 bucks. But, yeah, like the guy from Toyota, that was great. I mean, because at first he was really skeptical and we kind of just kept, you know, like, plugging away at him and then he. He's like, finally using it, and then he gets, like, um, you know, some of his friends to use it. So we start to see, like, okay, there's like, more than, like, there's a couple people, you know, using it, um, you know, here at Toyota. And, um, so that was like the first real, like, negotiation we had. And I'm trying to remember it was a funny one where, like, you know, we charged people like, a few grand each time. We tried a couple different, you know, models just really having, like, no idea.

Host: Right.

Lukas Biewald: You know, what?

Host: Different. Different pricing models.

Lukas Biewald: Different pricing models. Yeah. Um, and, uh, I think the Toyota one we went in with, like, we said it's like 100k and then I think they just said yes. And then we kind of felt bad that, like, you know, we're like, fuck, we should have gone in, um, like, even higher. So that was just like. But I remember being, like, really scared to even, like, throw that, you know, that number out.

Host: 100K.

Lukas Biewald: Okay. Yeah, yeah, exactly. And you know, the funny thing is it, like, it's not like I sort of thought it for that. Like, oh, yeah, this is gonna be like, Huge. But actually no one ever, no one else was really ever willing to, Was never that generous. So I don't know why it like worked so well. That.

Host: What was the um, what was it like, seat based or what was the like did they have you know like 100 researchers or whatever? You're like 100k for a platform for 100 users or.

Lukas Biewald: Yeah, something like that. Yeah, yeah, yeah, like 100 users or whatever. But I don't think, I think back then we sort of like had a platform fee and a user fee and then we kind of realized people it was better just have a user fee because like people use it more than they. People would end up with more users than they thought. And so we kind of changed it to yeah.

Host: So why, why have a platform fee that you know, potentially creates friction there versus people just like, yeah, slowly kind

Lukas Biewald: of look at it as like just what is the cost of this? They weren't that tactical about, you know, so we just sort of had to like get the buyer to have like agree to it with a straight face and not feel like you're ripping them off. But then, you know. Yeah, so that, that's, that's where we ended up.

Host: What, what, what was that roughly per user?

Lukas Biewald: Well, we would shoot for like, um, we'd shoot for like, was it like kind of like four or five grand per user, uh, per year? And that was always like a real sticking point. Like I mean I think like, um, I think if we, who knows, you know, everyone else ended up pricing like us. But I think usage based pricing might have worked better like if we could send a message back in time. But it would require a lot of patience. Um, like, but yeah, we would just charge for user. It was like kind of simple. Um, you know, everyone kind of sort of knew how many users they would have so it, it would get through the, the problem was just that we were like higher than all the other things that people paid per user for. So they would be like totally kind of pissy about it.

Host: They'd be like what the hell is this? Right? Well, I mean I think this is a perennial challenge with dev tools is. And uh, I see this with folks that I work with all the time where like you know, a Salesforce seat is uh, A uh, Salesforce CRM seed is like 150 bucks per rep per month or whatever. So you know, nets out to anywhere between like 1500 or 2 grand a year or what have you. But then the problem with dev tools is that you have like GitHub, which is like I don't know, 50 cents a month or whatever the, whatever the hell it is per user. It's like wildly inexpensive which then creates this like floor. And so then people perceive that and they're like wait a minute, why are you like this? And et cetera. And it sounds like you guys encountered that. But the good news was, is that, you know, this is always a two edged sword with new categories, uh, is that, you know, the ML AI researchers like they didn't have an existing tool chain I don't think to index off of. So at least there wasn't like. Well the other thing that they use for the AI stuff is know, much cheaper than. I guess you became that problem is

Lukas Biewald: they all used GitHub, so they were all anchored on GitHub anyway, you know, so it's like, you know, it was always a, some griping was that like

Host: the, a number one objection was like why the hell are you guys like 10 times more expensive than, than GitHub? It's like, well because that's been around for 20 years and, and like, you know, and it's not all that complicated. Whereas this is, you know, actually how did you justify that?

Lukas Biewald: Well, the other thing, well you know, I'd be like, look, this is like a purpose built, you know, thing for a specific problem. But I mean if you look at, for weights and biases on Reddit stuff, the complaint that you'll find everywhere is just that, you know, people were mad how expensive we were and it's tough because like, you know, we had kind of a PLG model in the sense that like people would come in and use us and they would look at our pricing. And the tough thing about ML researchers is like, you know, a lot of them are just out of school, they're like incredibly price sensitive. Even though now they're like, you know, managing often like you know, hundreds of millions or more of, of like compute spend. But they would, you know, they would be really um. So we always struggle with like what pricing should we put on our website? Because they would come in incredibly price sensitive over time as there'd be more value. They would be much less price sensitive actually as higher you go up in an organization, there'd be like a much higher willingness to pay. But if we didn't get them using it like out of the gate then um, you know, we wouldn't be successful. So we put kind of like a low price on the website, up to the first 10 users and that. But then people would feel kind of tricked like when you have to go to like an enterprise plan above 10 users. But that's actually where you know, teams would be like getting a lot more value uh, totally out of the product.

Host: Yeah, it's like you essentially had like a non linear, you had like a bend in the, in the value curve right there. I think Jira did that, Jira does that. Way back in the day they flew it.

Lukas Biewald: That's why we, we had that happen with Jira and then we, we felt more okay with our pricing. But they're like what are you guys doing here?

Host: It's like we learn for the best.

Lukas Biewald: Yeah, yeah, yeah. It does feel wrong though. I mean I totally, you know, I understand where people are coming from but

Host: I think what you're describing right there is an extraordinarily important right like um, where what we're trying to do from a go to market standpoint is we're trying to deploy a technology into the market in um, a rapid, like as rapid as way as possible where we can capture value. And so sometimes there are kind of like, like non linear parts there where you're like okay, cool, I can sneak in here. And this is kind of reliant on like the physics of your, your go to market. Um, where if the person who's going to be adopting this thing especially in, in uh dev tools may be an end user who's going to be the one who's like trying to figure it out versus like a top down sell, then we need to make it inexpensive like low, low cost either time cost or you know uh, expense, budgetary cost in order for them to try this out. Which of course is like that's where open source go to market business models have come from etc. Etc. So um, anyway you have to get in where you fit in um, and, and not be and oftentimes this is why different go to markets have similar shapes. Because, because the buyer type, um, and the user type and also how the organization buys kind of have echoes across that. Um, anyway, don't beat yourself up for it. It worked out just fine. So how did like I think you were talking about this like early on you had a lot of these folks who were really price sensitive and you would start with kind of these end users. Did you iterate your ICP much over time? Although I suppose you started with like Toyota which is a very large organization. Did you event like how did you eventually hone in on exactly like who was a great fit customer for weights and bias?

Lukas Biewald: You know the funny thing about actually the, I mean I Think that one difference with dev tools and a lot of other things is like, I think dev tools tend to be less vertically focused and, and it's the. We never really were successful even doing vertical marketing. It's not like we were against it, but basically companies liked us to the extent that they had a lot of um, ML engineers, or you'd call them M AI engineers now. And that happened in a lot of different industries kind of at the same time. So actually what we would do in order to know if a company was likely to buy us is we actually got pretty sophisticated. I don't know if we got pretty sophisticated. That would be a stretch. But we spent a lot of effort. Sophisticated enough, sophisticated enough, more sophisticated than some um, like just trying to count how many um, ML engineers they had. And so like it's because like even like, okay, like one thing that's interesting, pharma, I still don't even understand this but like some pharma's companies really invested heavily in um, AI and had like huge teams of ML M engineers and were like great customers. And some pharma companies like never invested in it and are still not customers of ours. And I have no idea why that is. And so it's sort of like we're relevant to the pharma companies that decide that they want to have a drug discovery process like using AI and are completely not relevant to the ones that don't. And that's kind of true in other fields, although other fields tend to go more completely or not. Um, but yeah, one of the weird things about weights and biases is like we're so spread out over um, industries. It's kind of fun. I mean we're in like ag, tech, manufacturing, like insurance, like finance. I mean and it's kind of fun for me because I get to see all these like, you know, businesses. But it's kind of like if we wanted to do vertical marketing, it's like unclear where we would even, you know, begin. Yeah.

Host: But I think that what's, what's cool about that is the answer to that question is like we thrive wherever there are AI researchers or ML, ML engineers. And it just turns out that that can be in a variety of different places. And so just like getting in front of them is the way like getting in front, like identifying the organizations that have those folks present. Um, it's not going to be, you know, size of organization, it's not necessarily going to be industry, etc. Um, is kind of the, a number one thing. And I suppose like LinkedIn's probably pretty okay for that, I would imagine.

Lukas Biewald: Um, maybe like they all name themselves different. You have to like figure out like a taxonomic that they describe themselves. Yeah, but yeah, that's um, that's the main way we do it. And now we look at job postings too, which can be telling, but, you know, that's really what we do.

Host: Yeah, I think that, that, um. Yes. Uh, back to the segment example. One of the things that uh, my buddy Raf was their early sales leader. And it was just really funny to see how organizations, like the only thing that mattered for them was organizations that had lots of events. And so you could have, right, you could have these organizations like Anheuser Busch that were doing as many events as like some mobile company. Right. Um, and like so their need was like similar, which is just predicated on like how many events they needed to, you know, to capture down. And that ended up being the high order bit which unfortunately, like, kind of hard to identify externally 100% of the time, but of course was like super important in discovery is like, all right, like, you know, how much, how many of these things you need to, you know, need to capture down. Um, so you had mentioned this about identifying, um, organizations that have, you know, uh, how many AI engineers or ML engineers they have in house. And then unfortunately that was, you know, there were lots of verticals. What were some kind of early customer, like what was the first customer acquisition channel that like really was great for you guys. Would you, would you say.

Lukas Biewald: Well, you mean like for, for new.

Host: Yeah, just like, like getting these folks, like getting attent, like getting these folks attention to be like, oh yeah, like I do have that problem. And weights and biases seems like a great way of solving that. Like how, how would you get in front of those?

Lukas Biewald: Yeah, yeah, I mean, so, so we, we had some tricks for getting, um, people using us. And so I think that was like, um, the biggest thing for us. So we got a lot of people using us like pretty early on. Um, and that was like, I mean we had a really strong kind of influencer marketing thing happen because people were really like passionate about our product. Um, so they would, they would actually like social media. Twitter was like a big one, but I think like an underappreciated one that actually kind of secretly works better was um, integrations with um, open source repos. And so, and that worked really well because it was like at the exact moment you need weights and biases, you would find us. So the weird thing about like starting with us is you Would like, like our on the people would come and they'd be like, okay, your like website flow sucks, you know, and it's. Well actually like nobody really goes through our website's flow. The, the real flow is like you're using some other project, like hugging face or something. And then you notice in the hugging face documentation that there's like a weights and bias integration. We make that really easy. You like instrument that then you're running it and now it's like, oh, you need to make weights and biases account in order to use this. So you're coming in out of your terminal actually typically at that time. And so it's actually like attribution is really hard but we did some tricks. But I think we know that basically the main way people found us was finding um, us in a third party repos documentation. And a lot of times it was funny because these are like open source repos so they're not like totally rational actors. Like sometimes you send them a PR and they're just like you, you know, you're a company. And sometimes they're like oh, like thank you. Like that's amazing. You know, I mean you kind of don't even know, you know, and then, and then it's like hard to tell like how um, you know, how many users you're really going to get out of somebody. So like we got like medium sophisticated at it. But I think the good thing was we kind of like kept at that um, that instrumentation to the point that then we kind of became the standard because it's like anything you're using you'd find a, an integration. So it made the product better, which is another reason I liked it. But actually that ended up being the main discovery mechanism for a long time. Um, and then in the end it was students actually. So we got a lot of student, um, adoption, actually ubiquitous, um, student adoption. Um, and that again took a while to kick in. But as the students kind of graduated and went into industry, um, that's what I think fuels the growth um, today. But it's always like, you know, we want companies to use us to some extent before we'd go in and try to um, sell them. Even though the buyer wasn't really ever the user. You know, we would look for usage and try to triangulate to the person who would buy us.

Host: Yeah, so I guess it's finding places where someone would identify a need for weights and biases. And largely where that was was in these open source repos. Um, and in that case what was the thing that you were doing a, ah, pr. It seems like you're sending them a PR for a doc update. What was the hyperlink to?

Lukas Biewald: Sorry. So what we'd want to do is we want them to make an integration where if they set some flag in their code, they would get some value out of weights and biases. So they just set a flag and now it's like on. And now it's logging to weights and biases. But then in order to see it, they have to click on a link because it's like a web app where they see the login that turned out to be useful for people because it's like even a couple lines of instrumentation will really slow down a developer. It's like there's cognitive overload. There's so many reasons that you wouldn't want to just go try a new library if you can get real value straight out of the gate. And if you're in a third party repo, you know what it's trying to do. So you can log the things that matter to whatever that repository wants to log off. And so you try to give people value really fast. But then it also serves as like a discovery mechanism because you're looking in the documentation of a library you're using, not us. And maybe you're thinking, I wish I could see a little more about what this library is doing. Oh, there's a visualization thing with weights and biases. If I set some environmental variable or there's some flag, whatever made sense for that repository, I can get that thing to turn on and start logging to weights and biases. That was the typical discovery mechanism.

Host: Yeah, so I think, um, it's uh, a variant of open source, kind of go to market here, where essentially it was leaving breadcrumbs in these libraries in a way that wasn't an obnoxious advertisement, but instead was high utility to the individual themselves. Like, hey, that I need some instrumentation. Oh, look at this, it's right here. And then also, um, presumably to the, you know, the person who is administering the uh, you know, the repository in question as well, maybe some of whom are, you know, full stallman and are like, get this shit out of here. Right. But maybe others are maybe a little bit more user centric and they're like, you know what, that actually does make sense for that, that hook to be, to be in there. Um, excellent. And then the, and then the student one was that, um, how did you like, did you guys do anything in particular to like promote that or did it just kind of like, because you had the low cost, like to your point about you were teaching and you identified the need and you had a low cost, you know, free version or, you know, non commercial version.

Lukas Biewald: We made it free for students. And I think also because I had teaching, I think I kind of probably just like snuck in a few features that made it good for teaching classes or just I had that, like. Yeah, that mental model was really like on my mind, you know, when we did it. So I think that caused teachers to use it. And then, you know, when teachers would use our tool in a class, then it's like, that's amazing because now all the students are like using it. You know what I mean? And then they would take it with them into, into companies.

Host: Yeah. I mean, that is how Texas instruments completely eviscerated HP's calculator, uh, business. Is that right? Oh yeah.

Lukas Biewald: It wasn't the, like, not having the reverse polish?

Host: Uh, well, I mean the reverse polish notation was like. I mean that wasn't easy. But all of us, I mean, you know, 40 years ago or 50 years ago, HP kind of shipped one of the first meaningful, like, you know, handheld calculators. Like, I think my dad still has his. It's not an LCD screen, it's like an LED screen. It's awesome.

Lukas Biewald: Yeah, yeah, my dad has one too.

Host: Yeah. And then like ti. Ah, just went in there.

Lukas Biewald: Yeah.

Host: I mean, I think this, the student go to market can be very effective. Apple has always done a phenomenal job of seeding that market. You have to be really as well.

Lukas Biewald: I mean, that's the thing. You have to. It takes like years, right, to, to really bear fruit, I think. Yeah.

Host: So my buddy, uh, Sam Bhagwat from um, Mastra, um, is kind of doing a variant of this right now where um, he wrote a book called Principles, uh, of AI Agents. And I swear to God they've handed out north of 100,000 of these books. They're not that big. They're like this big or what have you. Um, and they're kind of doing a similar, you know, thing with CS organizations as, as well, where I think probably lends itself in a, in an emerging category or like a new kind of like a new, new uh, category or like a new m. Like area of expertise. Probably lends itself, um, to that. Right. Versus probably less useful to show up after like the market has played out and been like, hey, students, here's uh, you know, seven minute ABS or six minute ABS version of relational database or whatever use ours versus something that doesn't actually have an answer quite yet. Um, okay, got it. So those are two good channels there. Um, what were some, ah, customer acquisition channels that never worked out that you tried and you're like, this does not work. Any of those.

Lukas Biewald: Well, SDRs, you know, I mean, I read your book, um, and I, I tried.

Host: Let's have a bunch of SDRs calling engineers. It'll be awesome.

Lukas Biewald: Yeah, I mean it, um, yeah, we tried many different variations of SDRs, and honestly, uh, um, it was always unsuccessful. So yeah, that, that was like a disappointing thing. Cause I'm not, you know, I, I, I thought it might work, to be honest, because it was, it wasn't like they were calling the individual engineers. It was like the, the bosses. But. Yeah, yeah, so, so that one, um, never really worked, but it might have

Host: been who's the, who was the boss, the boss Persona. In that case, would it be like the VPE or cto or was it VP often?

Lukas Biewald: Yeah. So like, I think other things do work, you know, through that, that channel. I think the problem that we had, the really, the, the, the funny thing about our go to market was that like, no one except the individual contributors ever really got the value prop. So it was like we never, we were really never able to articulate the value prop in a way that anyone else like, really cared about.

Host: Uh, you have to use it to get it, bro.

Lukas Biewald: Yeah, I mean, we really tried, but boy, we tried because I do think if we could figure that out, it would have really unlocked a lot. But it just like, I mean, a lot of the advantages that we had were kind of hard to explain. It was sort of like, um, you know, our UI was a lot better than anything else. And people like, love that, you know, but it's like, it's hard to like, convince someone, you know, to buy something UI and any feature like comparison matrix like Amazon and um, you know, Google would like kill us, you know what I mean? So it was like, you know, we, I felt like we would almost like our ethos is like, okay, we're going to lose these like feature comparison things because we want to make like really good software. Um, but we kind of had to let the individual champions like, advocate for us and push for it. And you try to make these ROI calculators, but you make developers more efficient. No one cares. No one cares at all. So it was kind of more like the best thing we could say is, look, you're having trouble hiring AI engineers. Probably you want to use the software that they want to use. Just go ask Them what they want to use. Just ask them. That's my only ask is go ask them what they want.

Host: Yeah, and I think um, I mean sounds like linear's go to market like product strategy. Right? Um, I talk about this a lot with, because you know, oftentimes founders will, they'll erroneously like cargo cult somebody else's go to market. They're like PLG or die. And it's just like all right, well like let's actually think about that. Right? Like do you have a non complicated product? You have a complicated product? I have a very complicated product for a non technical audience. Oh, okay, let's not do PLG there. Right. Versus a non, uh, complicated product for a non technical audience. Like a notion or a Slack or a Zoom M or what have you. Um, the problem of course being that also those non technical audiences oftentimes at least the end users don't have a lot of juice in the organization. Whereas fussy, ah, scarce human capital that have opinions and also are a pain in the ass to hire and you don't want them to attrite. Well, it turns out that those folks oftentimes have more juice. Right. Um, and you see this with like, you know, the success of Figma in the face of uh, Adobe DX's and like I'm sure there's a bunch of product stuff, et cetera, et cetera, but at the end of the day it was like designers wanted to use Figma and they were just like, I don't give a shit about your Adobe Enterprise license. All my homies use figma and so I want to do that too. And it sounds like the same sort. Whereas we saw this in Go to market tech there are a lot of companies um, that thought that they were going to do PLG go to market. Like hey, um, a buddy of mine ran this company called Scratchpad that was this Ajax layer that lived on top of uh, Salesforce. And it was lovely because Salesforce is a pain in the ass. Every time you got to save something or like edit something, whole page Refresh takes like 15 minutes to reload or whatever. Um, and so they did this like AJAX thing that was like really great for Engine and it worked great. Sorry for salespeople, it worked great during ZIRP where people are like, oh, we gotta like, gotta make sure we can hire these account executives and we don't wanna piss them off. They might go to whoever, you know. Tiger global just dropped $50 million on et cetera. But then as soon as like Zurf ended those like those go to markets are like cool. Use Salesforce, you don't get the shiny front end on top of that. Right. Um, whereas you know, AI engineers, uh, designers, dev tools in general still probably have that, that like level of uh, of juice. Uh, don't, don't poke them. Right? Don't poke them. Let them use the thing. So what were, what were some go to market. So that's like a devrel thing. And what were some things you did aside from product to make it such that weights and biases was like the beverage of choice for the discerning, uh, AI engineer, for the individual using it? Yeah, yeah, aside from the product because like obviously you guys are phenomenal software craftsmen and so on and so forth. But outside of that, um, well, I'll

Lukas Biewald: tell you one thing I guess is a little bit product. But I feel like this is actually growth and I feel like sometimes product can be growth. I. One thing we did that like, I just, we can never do. I can never convince people to do now, but we did like a year wrapped um, thing like you know, Spotify does like year wrapped. So we would do like year wrapped of like how many um, like runs did you use and how many like GPUs did use and stuff. Actually it's funny, you know, we, we first, we launched one, um of the things we tried to do to get like centralized buyers to like us, the infrastructure teams is we, we would like show how much resources, you know, teams are using. Um, but actually the, the infrastructure people didn't really care. But the uh, users would like treat that as like a high scoreboard for themselves because they like wanted to be

Host: m. I'm going to steal. Look how many GPUs I used.

Lukas Biewald: Look at that. Seriously. They really like it. Yeah. And then, and then Toby like oh my God.

Host: It's a measuring contest.

Lukas Biewald: Although it's funny because I mean I just like, I love our users so it's hard to be mean about it because they're. And essentially the year wrapped thing is funny because it worked, worked really well on social media because everyone wanted to post it because it's kind of a humble brag. But then um, it's like, it's funny when I ask people for feedback on the product. That's like often where people's brains go. And I even met a couple where the, the like husband was like met his wife because he was kind of bragging that he was in the top 1% but then his wife was in the top 0.1% so she just like went, you know, over the top and then he was like, really?

Host: That's cute, honey. Yeah, go change the baby's diaper.

Lukas Biewald: It wasn't part of some strategy, but I think, I do think like some kind of delightful product features can really like, work in a fun way.

Host: But I think that's exactly what I think what you guys were, you realized who had the juice and then you were investing in that, both from a pure product standpoint and then also looking for opportunities where you could do things that would deliver delight and utility to the end user. But, you know, maybe in a way where they would like, share that with each other, you know, share that with other people. I mean, I guess like the earliest, earliest MVP of your product was that, um, that analytics use case that could, that was inherently viral because like, you could flick the URL to your manager and be like, here it is. Right? Yeah, like, you know, versus screenshots or what have you. Um, okay, um, one thing that oftentimes is challenging in these, uh, developer sales motions is, um, you have love at the IC level, but then you got to get up here to the leader level. And like you were saying earlier, if we showed up there and there wasn't usage here, it's not like these guys were going to come off the top rope and be like, everybody needs to use weights and biases. But if they were, you know, if we're already, if the organization was already filthy with weights and biases usage, then that would be a much more productive conversation if we can precipitate that conversation. So what were some of the mechanisms by which you would then precipitate that, that conversation with those directors or VPs, assuming that the organization already was, you know, lousy with, uh, with weights and biases.

Lukas Biewald: The reality, just for the audience to know, it's like, I don't think we were quite as like riches as like GitHub where we'd wait for like 30 using it and just like sit back and like wait for them, you know, to, you know, we, if we saw one person, you know, using it, we would try to, you know, to, to get meetings and we'd really try to go through that person because they usually, they liked us. So we would like ask them, you know, hey, who could we talk to about, you know, getting something, uh, in, in place here? And I mean, that was the main way. Like, we tried a lot of like, and we put out content for that person. Like, a lot of our marketing efforts were more targeted, you know, at the, at the boss and occasionally we, we'd get, we'd land on something that works well, like um, you know, it was a real power law. Like we got a couple things that they really liked always. Yeah.

Host: And, and we get a lot. Do a lot of experiments.

Lukas Biewald: Yeah, exactly. And, and, and you know, so, so, you know, we'd make it easy for them to reach out and, and buy and um, you know, we. Yeah, I mean, so I don't think it was any, any magical thing. I don't think we're like masters at it. Like, you know, and it was funny because it's like sort of like we would just hope that they would come in like, not like hostile to us. You know what I mean? Like if they were like neutral, like that's like, you know, you were just

Host: hoping that the, the leader, the director or the VP wasn't, wasn't pissed that somebody was like erroneously or like gorilla using weights and biases.

Lukas Biewald: Yeah, that happened. All the, you know, people would standardize on one of the top down sold, you know, products and then there'd be like a gorilla thing and like, you know, we'd be um, you know, trying to wait for the right time, like trying to get like enough kind of purchase in the organization. Sometimes we could get like a, you know, like a manager to buy it in like a pocket. Especially like where sometimes they were like, you know, the people that know like LLMs are like really special and they can kind of do whatever they want. And so like, you know, they would get to buy us first and we'd sort of get the like little goes

Host: to the juice, right? It goes to the uh, it goes to the juice model of plg. Who's got the, who's got the most juice in the way? Like the weighted average, you know, juice.

Lukas Biewald: Like the gradient of admiration. I think this really worked for us

Host: where it was like, well, that's a great phrase.

Lukas Biewald: Yeah, it's a great phrase where you just. We try to go to the top of that. You know, like we had like, we kind of had like these celebrities. Like this guy Andre Karpathy is like a big deal in the AI world. And you know, he would be like, oh, like I love weights and biases and that would be like, honestly the best content. Yeah. Or like he'd always use us like in his open source projects and then like, um, yeah, I think it was stuff like that. Honestly.

Host: Take like screenshots of, of his tweets and then like, literally it's like the outbound email or it's like your first slide Is like, well, nice to meet you today, Bob. Oh, I'm sorry, did I have that Andrea Carpathy, uh, tweet just sitting on my desktop there? You've heard of him before, right?

Lukas Biewald: Yeah.

Host: Interesting. So it's almost like. Oh, I love that because it's kind of like the Ferrari model. Yeah, right. Where like Ferrari. Um, actually I was listening to the acquired, uh, podcast on Porsche and how, um, maybe less Ferrari, because, like, Porsche is more of a, um, like a, you know, uh, it's a luxury, but it's also like a very broad luxury. Like, you know, Safari only sells so many of their, you know, so many cars. But like, Porsche ships like a lot of cars, but they invest extraordinarily much in their racing. Right? Not because anyone. Like, they don't make money on that, of course. Like, it just sets money on fire, but then it, you know, percolates down to. Via the gradient of admiration.

Lukas Biewald: Yeah. Honestly, Pete, I think it's underrated in the developer tools because, like, that's a great point. If you go to like, um, you know, an executive at like a random company. I guess I don't want to name names because it's like, okay, a company's not tech forward. They.

Host: The CTO of Monsanto. You said it.

Lukas Biewald: Although Monsanto maybe more tech forward than you would think.

Host: But.

Lukas Biewald: But anyway, John Deere. Um, John Deere important with advice customer. I love you guys.

Host: National Cost Register.

Lukas Biewald: Do this. Okay. I don't know.

Host: It's actually so funny because, like, Monsanto has. Because Monsanto bought, um, Weather Bill or Climate Corp or whatever. So I'm sure they do insane ML stuff.

Lukas Biewald: And then Monsanto and John Deere have terrible examples. They have really good ML teams.

Host: Yeah, well, yeah, because, like. Because Deer bought a bunch of, um, uh, ag tech companies as well. Anyway. Sorry.

Lukas Biewald: Yeah, yeah, totally. Anyway, I. I can't even name one.

Host: But anyway, Barney Rebels Rock repository.

Lukas Biewald: Yeah. Yeah. They do have different issues than like, Google. And they know that, and so they don't. They're not impressed. Like, the VP of Engineering at. Barney's Rock Company is not impressed that, like, Google's using your tool.

Host: Andre Karpathy.

Lukas Biewald: But the, uh, the engineers don't care. Like, they want to go work for Google eventually. They. They want it. They don't want to use the baby tools. Like everybody. It's funny, back when we're starting, everyone's talking about the democratization of AI, and I was like, this is such a stupid message. Like, it's a good message for vcs. Maybe that like we're democratizing AI, but I was like, I don't think people want people's AI tool. I don't want to use like the really good AI tool.

Host: Yeah. Like, I'm a craftsman. Do you know how like expensive my knives are?

Lukas Biewald: Yeah.

Host: And my, like I have. Yeah. I'm a like, what? Ryobi, get out of here. I'm like the wall guy, right?

Lukas Biewald: Is the wall high end? I feel like the wall.

Host: I have no idea. I think there's like four at this point. And like they all, they all do the. It's the same mechanic as like Chevy versus Ford. But I think the point is well taken, which is like, I think the thing that I'm taking away from that is that if these people, if the ICs have juice and also want to tell themselves a story about um, you know, about excellence and um, that so not only should you invest and make the product high quality. Right. Um, but, but you should also make the branding and the go to market reflect that. Right.

Lukas Biewald: Versus I gotta say, I really actually. Okay, I just uh, I don't want to say it cynically because I actually think like, like it's like insane how little we spend on developer tools.

Host: It is crazy.

Lukas Biewald: It's just insane because it's like, it's

Host: like the GitHub thing, like makes my mics, like, I feel that if GitHub, like raise their, like raise their prices by two orders of magnitude, it would like help developer go to market like so much better because then people would actually be like, okay, cool.

Lukas Biewald: Well, I think they found a different path to monetization that you'll experience if you use other parts of their product. Um, that's fine. Yeah. And like, I think actually the core of acquisition kind of made sense because it's like, like so much more natural to pay for infrastructure, you know, versus developer tools. But it's really crazy to be cheap on developer tools. Uh, and like the fact that we had the money to make high quality tools, I feel like immensely like proud of that. We like helped all these different companies like actually like, you know, build things more efficiently. And I mean we talk about like 10x developer productivity gains, but if you took your ML engineers and made them 10% more efficient, we're like, yeah, the

Host: salary expenses, just bananas, right? Like this is like a component of, of pricing when, when founders are starting out, like, you know, maybe you index off of other stuff. There's of course, like, you know, like, how do you, how do you help make money? How do you help Save money, etc. But you know the, the salary expense oftentimes people get like, like uh, I don't know, time savings, et cetera, et cetera. But again if you have people who are extraordinarily expensive and you don't want them, you know, uh, it's like the jokes about you know, cicd like okay cool, I'm going to kick off the build and then like go home for the day at like 1pm or whatever. Like if you could not have that happening right, with your $300,000, $600,000, you know, seven figure a year human being, like that's probably worthwhile and Maybe shouldn't be $1,000 a year for their seat. It should be $4,000 a year for, for that seat. Yeah, I, I, and then, and then your go to market, it should reflect that From a branding standpoint. Yeah, again non cynically but from a branding standpoint etc. Etc. I think that could be very impactful. Now doing that with um, you know, doing that for kind of like lower cost uh, users or like software that services lower cost users, probably gonna be a little bit tougher, right? But like if you are in a, if you are close to a high amount of like economic intensity then probably leaning into that is not problematic.

Lukas Biewald: I would just say to like people listening it is hard because like I think we could not have picked a higher paid, like more demand Persona and it was still really for us. So.

Host: Yeah, still paying. Yeah, yeah. Um, when it came to CS like customer success, I think the interesting thing here is that you guys didn't, you know, it wasn't top down sold so you didn't like go through and like resell a bunch of people and like you got to use this because the CTO says you gotta um, but still like I imagine you have like patches of usage and then maybe you get an ELA and like you need to penetrate across the rest of the organization. What was particularly important for your success Motion early on.

Lukas Biewald: Yeah, it's funny, we had this thing where we would try to get like deeper into our customers like infrastructure. So like what happened is like people like integrated. Yeah, integrated. So people would like start by integrating us into their kind of toy stuff and that would be like researchers who liked us. But then if we could convince the infrastructure team that usually didn't like us to like bake us into like the core like infrastructure they were using. I mean usually there was a team called like the ML Ops team at the time. Now they're called like AIOps and usually they wanted to build like everything, you know, because that was sort of their purview. And usually like the researchers had like mixed feelings about them, but it was like they were kind of just wanting things to work. But then there was a Zemalops team that felt like a little competitive with us.

Host: Yeah, totally.

Lukas Biewald: We got like, I think again we just want the MLAPS team to not hate us. Like that was like, you know.

Host: Right.

Lukas Biewald: And, and so we would just try to like make our stuff um, where you didn't have to like use it all at once to get value out of it. So we tried to make it all like, you know, you could like like kind of buy into whatever you're uh, thinking.

Host: Use that part.

Lukas Biewald: Yeah, just you know, whatever. Like you know, if you have like your own versioning system, great. Cause like everyone for some reason built their own versioning system. So I was like all right, we have ours, but like we'll work with yours or like you know, whatever it was. And so. And that was actually all in the effort to try to get people to bake us into their. Because like our, our usage would kind of like be like, you know, it would go up users like we do like, like mini upsell. Mini upsell, mini upsell, mini upsell. And then there'd be this moment in the best accounts like John Deere. I can just think of as a good example. Although they'll probably kill me if I hear their name. But whenever it's been years of working with them, um, where they would jump to a massive install, um, and what that was was like getting hooked in. Yeah, getting hooked into be like a default. So when you start a new project now you're using um, weights and biases. So we would try to like we had different again I don't think we did that perfectly but we would really try to make that happen. And then a lot of the function of the CS team was to try to like encourage that to happen.

Host: Yeah, I think um, when you have kind of like lane and expand behaviors. Um. One Making it easy. There was another founder led sales story with Carl uh, Hanson from Tonic, who kind uh of talked about this about landing where you can get in and I think you mentioned that earlier about landing maybe in the fanciest group of AI engineers. Oh, that's the LLM folks over there. Um, but um, but then making it then easy for them to then buy more and then maybe not just easy for them to pull more but also like you know, maybe push a little bit there as, as well. And that but, um, oftentimes that looks like integration with, uh, additional tooling or what have you and not just being passive about that. That's where CS can actually be very helpful. And then I think the ability to kind of choose your own adventure, especially when you have, uh, uh, a part of the organization that potentially can be hostile to you. Um, like we had this with Atrium where we had a two legged sales motion. Um, we had sales operations and we had sales leadership. And usually we're pretty friendly with both. We just kind of had to talk out of both sides of our mouth with, with those folks. Like with Revos, you have to be like, oh, it's so frustrating. They don't appreciate your dashboard art, do they? Like, no, they don't. And then with sales leadership it'd be like, man, those nerds, they never deliver what you want, right? And they're like, yeah, it's the worst. Um, and so we would do a pretty good job of that. But, um, but the bi team always was just like super hostile because they're like, that's our job.

Lukas Biewald: Totally.

Host: That's our job. What are you doing? That's our job. And um, you know, and so, uh, I, uh, don't think we did an amazing job of this. It's actually was the, the results of this, you know, precipitated um, you know, some pretty painful churn situations where we hadn't been aware of that, as aware of that as we should have been. And I think probably the way of doing that would have been what you're describing, where it's like, hey, no, no, we're not. No, no, no. Your tableau castles in the air. They are lovely. You know, you support the cfo, you support the CEO, you support the board. This is this thing for these unwashed sales leaders, uh, over here who honestly, do you really want to interface with them? No. Right. And so it's just kind of like cool. This is not, you know, it's not going to touch your cheese. Right. It's okay. And uh, I don't think we did a phenomenal job of that. And we probably could have done much better to make sure that. Play nice. Nice. Especially in large organizations that where like you start having internal functions who are like, hey, I'm supposed to be doing, I'm supposed to be doing well.

Lukas Biewald: We had this hilarious dynamic throughout the history of the company. It's kind of like in the last year or two, I feel like this happens a lot less because our brand is bigger. But it was always like, oh yeah, like we'll buy you for now. But just so you know, we're, like, building. You know, we're building you, so we're gonna, like, take you out.

Host: And I like, sure, Jan. Be scared.

Lukas Biewald: And then I'm like, you know, this looks really good because it means they have a need. And, like, nobody ever. Nobody ever built us. There's too many other things to build, you know?

Host: Yeah, I think one of those things you can do is, um. I was talking with a team I coached the other day about. This is like, when you identify those sort of objections ahead of time, um, and you know that they're actually, like, defanged. You can actually kind of pull them forward and be like, you know, oftentimes in organizations that are sophisticated as yours, um, sometimes people are considering building this on their own, and I understand that. And honestly, sometimes people will do that, but in the short term, you don't want to eat that productivity loss on the part of your AI engineers. Uh, and so a lot of times people will start with weights and biases before moving on to something that's a little bit more purpose built. Yeah.

Lukas Biewald: Okay.

Host: See, we're not enemies. We're not enemies. Like, it's okay. You're just going to rent us, you know, for the short term here. I don't. I'm not trying to replace you, Bob. I'm not trying to replace you. Um, I love it. Well, um, last kind of part here I want to pivot to, actually. How many customers did you. So were you primarily doing the early sale?

Lukas Biewald: Was.

Host: Was Chris helping out on that? No.

Lukas Biewald: Well, so it's kind of like in the very early days, it was doing the sales, and Chris is kind of the sales engineer with sort of. That's kind of always how we've done it. Love it. Um, great. And then, um. Yeah, we get this guy Ari. Actually, I just gave him credit because he came on very early. He was my first sales hire at Crowdflower. Um, and then he came back and did it again for, um, you know, weights and biases. So he was kind of magic of that, too. And. And that kind of gave us a little bit of a dynamic where sometimes he could go in and suggest a price, and then the customer could get mad and come to me, and I'd be like, oh, he's crazy. Like,

Host: Mom. Um, how many, um, before Ari joined up, how many customers, uh, would you say that you had ended up closing, uh, ahead of that?

Lukas Biewald: 10. I. I would have done longer, except I just knew Ari. He's almost like a co founder.

Host: Yeah. It's great.

Lukas Biewald: Yeah.

Host: Yeah, I think, um, especially I, I talk about crowdflower so much. Um, when I'm talking to early, um, mainly I talk about. I talk about crowdflower and weights and biases to young founders a lot about the power of having extraordinary, like, subject matter expertise and how just like, being around the thing that matters is going to put you in the. Like, it's going to have you poised to strike. And, um, uh, all the time, like, I talk about it constantly, right? Because if you think about it like, you guys were so close to so many things. It's like so funny to watch everybody, like scale and surge and mercore and all these guys essentially doing what Crawler was doing with, uh, like Mechanical Turk stuff.

Lukas Biewald: You can imagine I have a slightly different, uh, reaction to that. But.

Host: Yeah, no, but, like, I mean, look, being early is. Is frequently indistinguishable from being wrong. Right? Like, I know this, have had many, uh, situations like this as well, but if you are like in the mix and then I think you characterized it so well early on where it's like, we are going to be the most expert at this Persona and whatever their needs are and like, they're going to pull us in the right direction and, and it's going to be good. Right? Um, and. And I think the, the human capital component of that is, is non trivial as well, where you guys had about, like, you knew exactly what your roles were. Like, you know, Lucas the seller and you know, Chris is the sales engineer. And then it's like, okay, cool, we have repeatability on this. We figured out that this experimentation thing, um, really seems like the early kind of like thin edge of the wedge, like, to the wedge product. All right, Ari, tapping you in, right? Um, and I think that's, you know, that's, that's fantastic. Um, as compared to, um, where it's a little bit more risky. You're like, no, no, no, I really gotta, I gotta tighten down this sales motion. I gotta get this. I gotta Repeat this like 20 or 30 times. So it's idiot proof because I don't know who my first seller is gonna be versus maybe like, you know, actually I know the idiot I'm gonna hire and I know that. I know his frailties. So it's actually okay. He can, he can take this half baked thing and run with it. I think that can be very powerful.

Lukas Biewald: You're not an idiot. You're a miracle worker. Just if you're listening to this are, you're the man. All right?

Host: You're the man, uh, wonderful. Well, you know, if there was a, if there was a parting shot that you had to offer to someone who's in your shoes, like at this point, I guess it's like five years ago or so, uh, you know, you guys were at that, you know, that very early, kind of like first 50 customers or what have you, um, what would you kind of offer to folks, especially folks who are doing developer sales, motion, maybe like AI focused stuff, what have you. What would be, you know, a thing that you would offer folks there?

Lukas Biewald: I guess maybe one contrarian view that I have that I feel kind of strongly about that maybe is helpful is like, I think if you pick, you know, if you pick a consistent, like, Persona and you care about them and you're like serving a few of them, well, I think you're likely to be successful if you stick at it. And, uh, I mean, I say this like my first company, you know, we went like 12 years and it was unclear that it was going to be successful. So it's like you might not be successful. I kind of feel like right now things are changing, you know, so fast and there's so much like, opportunity to like, serve people well and like, value there. I do think there's like a little bit of like, a craziness with like, you know, kind of companies, you know, going from like 1 million to like 100 million ARR overnight or something, where I think people maybe bounce around too much and get a little too discouraged when you really have to, like, push someone to like, use your product and buy your product. And I think it's okay if like, you know, you really have to, like, you know, push people for like a good long time. And I, I don't know, I feel like people kind of give up too soon, kind of pushing, pushing their product on people. Like, I don't know, I. I see that in like, companies that I invest in and you know, advise sometimes, like, I kind of wish they would just, um, you know, I don't know, push a little harder and, you know, make the product better because it's like, that's like the value. It's like when your product's like, really good. Like, of course, when your products new, it's bad, it should be bad, you know, and it's like, you got to keep, like, you know, keep pushing, keep pushing. And then, um, I think you probably, if you push long enough, it'll get good. And then, you know, like, I mean, software's so amazing. Like, every segment is like, valuable. If you can win it now.

Host: So. Yeah, totally. I mean, and I, and I think that looking for, I think what you're kind of describing a little bit is like founder, uh, market fit or like founder, like user market fit where like if you one have passion or like, you know, intellectual curiosity about that topic or like, which in you guys case, you're like the kings of, I mean early on, data labeling and so on and so forth. So you're like, all right, we know those things, right? I, I think one, making sure that you're, it's something that you're excited about is, is a good thing because it's going to be a long haul. And then I think the other thing as well is just being honest about the fact that is there a lot of um, commercial intensity there in you guys case? There's a lot of commercial intensity around AI engineers and ML engineers and what have you. And commercial intensity of course can be a small number of high intensity human beings or business processes or it could be a lower intensity, but there's a lot of that. Right. You know, uh, Stein from uh, Lassie was on here, you know, and like they do this, you know, AI automation, uh, AI, like building automation stuff for like dentists. And it turns out that like, you know, the individual human who would be doing that, um, probably costs like, you know, $80,000 a year or what have you. But like, it's in the flow of like millions of dollars of billings for the, the dentist, the dentist's office. And it turns out that there's like a bajillion of these, of these dentist offices, right? So like when they were identifying that, they're like, okay, cool, we can be excited about this. Tons of like commercial intensity in here. We're going to go full like ethnographic research and like sit in their lap and actually do the coding behavior and so on, like the billing, coding behavior and so on and so forth to really nail the hell out of this problem. Because we know that there's a lot of commercial intensity here that we can, that we can, you know, take a slice of for, for ourselves.

Lukas Biewald: I, um, guess maybe this goes against like you're like, I don't know, you kind of go to market orientation, which I respect. But I do think like, you know, when we started there were actually a lot of competitors, like a lot of different, like every take you could imagine on our market, even though it was small, even though it was nascent. But I kind of feel like people like got distracted by like making their product actually like good. Like, I mean, I Saw products that were kind of, like, kind of nicer, like, kind of nicer onboarding. Like, you know, products with way more kind of go to market energy, you know, behind them. Like, I kind of saw, like, every variation. I think, like, we had good vision, but the vision is maybe similar to a lot of other, you know, companies in our space. But I think the way we, like, out executed the competition. From my perspective, I wonder if my co founders agree this, but my perspective is like, we actually kind of just listened to what customers wanted and gave it to them relentlessly in a way that it didn't seem like our competitors, um, were doing at the time. And I'm kind of. I think that often could be, you know, the winning strategy, because it's not like we came out of the gate with the best practice. And I even remember we had a customer was just like, I want to buy your product, but, like, the UI is so garbage that I, like, can't even, like, take you seriously, you know. And then I was like, okay, like you. I master. Our customers will prioritize it, you know,

Host: but, hey, hey, boys. Guess what? The next Sprint is on.

Lukas Biewald: Yeah. Yeah, exactly. But, yeah, I don't know. Like, I, um. I don't know. I. I guess, um, maybe. I don't know what I'm saying, but I do think, like, just sort of like that, that kind of relentless iteration, you know, with customers trying to, like, make it a little better. A little better. I think it's kind of underrated. Like, I wish I had done that more with my first, um, company also. Like, I think that would have, like, kind of served us better.

Host: Yeah. Like, stay close to the metal and, and react to the pull.

Lukas Biewald: Yeah.

Host: As it, as it shows up. Which I think oftentimes, unfortunately, as, as founders, we want to abstract and look for, like, okay, like, what's the meta pattern here? Right? Like, oh, I don't want to get stuck in a local maximum here. And so it's like, man, just go up. Yeah, right? Which I. And I think that the YC guys do a pretty good job of talking about. The common failure modes are like, don't worry, dude. Your common failure mode is not that you end up at a local maximum. The common failure mode is that you ship something that people don't want, or you're not responsive to your customers, or you spent so much time making this abstraction when all the customer wanted was for you to hand them the pepper. And like, you spent so much time making an abstraction for passing an arbitrary condiment that, uh, they decided to say you should. They churned. They churned and they went to someone who would give them fucking pepper. Right?

Lukas Biewald: Yeah, exactly.

Host: I love it. Uh, well, Lucas, thank you so much for taking the time to hang out with me. Uh, I'm sure this will be super helpful to folks who are engaged in, uh, uh, developer sales motions, especially of an AI Variety. And, uh, I hope to see you soon.

Lukas Biewald: Yeah, thanks, Pete.