The text below is a transcript of the audio from Episode 40 of Onward, "From Bootstrap to Billion-Dollar Business, with Tristan Handy CEO of dbt Labs".
---
Ben Miller: Tristan, welcome to Onward.
Tristan: Thanks for having me.
Ben Miller: I thought we'd start with your founding story because not every listener is going to be familiar with you and dbt as I am. And I love that you're from the suburbs of Baltimore. I'm in DC. I
Tristan: Oh, I didn't know that. Where do you live?
Ben Miller: live in downtown DC. I'm from DC.
Tristan: Yeah, I went to undergrad at college park. So have wandered all over the city. The founding story of dbt is really that I was at a prior company called RJ Metrics. And I had been in data roles for something like 15 years at that point. And at RJ Metrics, it was the first time I had worked in the technology part of the data industry.
Like I wasn't a practitioner, I was building the products and RJmetrix was a product that did BI, but it was formed pre cloud, or at least pre data in the cloud. So all of its understanding of the world was the ideas that BI tools were built with in a pre cloud environment. And then all of a sudden, we have this hockey stick curve and everything's going great and up and to the right and raising series A, series B.
And then all of a sudden, the hockey stick curve literally just flattens out. We lose all of our steam. Very quickly, and it turns out that is happening because there are a bunch of cloud native startups that are building on top of a platform called Amazon Redshift and Amazon Redshift is the first real cloud based data warehouse.
Now, you could talk about there's other products that have maybe pushed the envelope since 2013, but Amazon Redshift is the first time you could swipe a credit card and get like a high powered data warehouse in the cloud for free. It was like 160 bucks a month and then a bunch of ecosystem built on top of that.
And all of a sudden we just had a leaky boat and it was clear that we were not going to optimize our way through this. And so launched another product there. And then I wanted to get out and back into the practitioner space because all of these tools were solving these decade plus long problems that I had as a practitioner, I saw a lot of people trying to use them.
But doing really poorly. And so I wanted to see if I could prove out some ideas that I had that maybe we'll talk about over the course of the episode. I just hung out a shingle, created a company called Fishtan Analytics that was just a data consultancy that we helped series A, B, C venture funded startups implement cloud data.
And that's the beginning of the journey.
Ben Miller: Just a couple of things. BI means business intelligence. So with dbt, did you actually. Invent the original dbt open source code. Is that you, or is that something you've found and popularized?
Tristan: There are really three people that are involved in. The very first lines of code in dbt, the product thinking around it, I think probably mostly came from me. The first dbt commit ever in the repo was by a gentleman named Chris Merrick, who is now a co founder of a company called Omni, but he was the CTO of the company that we were at back then.
And. DBT was a little bit of a skunkworks project at the last company, RJ Metrics, that we were at. And then the third person is Drew Bannon, who became my co founder. And he wrote, I don't know, 95 percent plus of the lines of code in DBT for the first six months.
Ben Miller: We actually invested in Omni's recent,
Tristan: Oh, cool.
Ben Miller: my co founders friends with Chris went to, I think, I don't know if they went to college together, but so small world.
Tristan: He went to a little school called Princeton.
Ben Miller: I'll go to Princeton Yeah. So what I love about your story, so Fishtown's in Philadelphia. I've been there. Our company started in a similar ethos early in mid 2010 was like urban.
It was the revival of cities. I feel like your company actually came out of that culture too. I don't know if that's true. I just started watching it from a far up into your office, where your office is. I love that there's a shooting range across the street from your office.
Tristan: Have you been to the office before?
Ben Miller: Yeah, yeah, yeah.
Tristan: Oh, that's so great. So we got our first real office. We bounced around between a bunch of co working spaces for the first, I don't know, three years. And then we got real office space, enough space for like 25 people back in 2019 in a district in Philadelphia called Spring Arts and ninth and Spring Garden.
And it's in a building that was built in like 11 or something like that by one of the Philadelphia railroads. I can't remember which of the big old railroads it was, but it's like one of the ones on the Monopoly board. And over time the office had become a little downtrodden. And by the end there was a fire and people were having parties in there and whatever.
And so I took one of my coworkers. To tour the office and we were walking through bombed out shell of a building that had a lot of pretty explicit graffiti on the wall. It felt a little weird to sign a massive lease in a building that most of the art was currently phallic.
Ben Miller: And that's the kind of project that we used to fund and build. Growth in those neighborhoods was amazing.
Tristan: There's so much culture in the history of the building and the bones of it. I really love our office.
Ben Miller: Okay, but I feel like you're understated. So you go on. To raise hundreds of millions of dollars. You have hundreds of employees. I don't know what tech company valuation is billions of dollars. You've got to be one of the most successful entrepreneurs in Philadelphia. It's like out of a movie, you're like this normal guy who figures out something and then really builds one of the most important companies in the data ecosystem.
It's central. It doesn't have, there's no real competitors. I think it's pretty wild, a
Tristan: Thanks. I don't like saying nice things about myself, so I'm glad that you said a bunch of them so that I can disagree with you. There's a lot of, Surprise in the journey, which the story of the classic startup, I think as everyone has submitted in their brains as like the social network and Facebook, and you set out to build a company that's worth a billion dollars and then work your asses off until you get there.
And maybe you die trying. That was very much not the story of Fishtown Analytics, now DBT Labs. I was really curious about a new set of technology. I wanted to roll up my sleeves and do the thing. And in the process, I really needed a particular tool. And I had a formed idea of The way that I wanted to build this tool, but I didn't anticipate monetizing it at all.
And dbt grew up as a productivity tool for our consulting business. And it turned out that it made us a very good consulting business. And that was great. We were more successful as a consulting business than I ever imagined. But as our clients got their hands on the product, they started using it themselves.
And then a little community formed around it. And that was fun because drew and I were. Off by ourselves, we didn't really have anyone that we were working with and so it was neat that other people were using this thing too and we could exchange best practices and a community arose around that and different companies would hold meetups and they would invite us to the meetups that were focused around our own product and the usage of dbt grew at a consistent month over month rate.
10 percent month over month for I think five and a half years and grew from nothing to now we're at 50, 000 companies using dbt in production. It's been a wild ride to be on.
Ben Miller: million data scientists or something like that, a million people,
Tristan: It's very hard to know the number of humans because dbt is open source. You don't have to log in, but if you say 50, 000 companies are using the product, then you can make some assumptions about how many humans are at a typical company. And I think you get to a million pretty quick.
Ben Miller: and we use it. And one of the things I like about it, and one of the things I like about the way you think is that I'm on the product side when I'm involved and my team can ridicule me when I talk about the engineering side, but it's transforming what we could do and we're building new products with it.
And what I like about it and why I think it abstracts out. The logic, you see say transformations, but the logic of what is happening, it makes it much cleaner. You can understand what anyone else is doing. One of the things I like spending time thinking about the tech industry over the last, whatever, decade plus.
So much of successful tech products are the right abstractions. Even, in a way, AI is able to see the right abstractions, the right patterns from this. What seems like a simple prediction algorithm. you did one, you said they invented this product, but then there's others you had some other instincts about the practices that needed to be followed, which I thought was really clean.
So I'll just say it and you can expand on it or disagree again. But you said that data analysts should adopt similar practices and tools as software developers. I don't know if you'd say it differently now.
Tristan: You know, the fundamental belief that I had when I watched data move to the cloud on a broad basis for the first time was that most people are used to using data in their jobs if they're some kind of knowledge worker, but the way that those people have used data historically has not been aligned around the idea that data systems are software systems.
It's more like The way that we used to use data was more akin to paper in paper based processes. You had an inbox and outbox and people put things on your desk and you looked at them and maybe you assigned them, maybe you stamped them, whatever. You're always like passing around these paper, right?
Ben Miller: Paper predates you, huh? Okay.
Tristan: I don't have professional career pre PC on the desk.
But when you think about how people used data pre cloud, it really was like passing around these data. Data artifacts, you would email around spreadsheets or TXT files or whatever. And fundamentally not that different than paper. It's just the same process, but digitize it. And that was really the first phase of how most.
Pre computer processes made it into the world of computers is let's just do what we were doing before, but do it on the computer. And I think the humans that are always furthest out in front of thinking from first principles of how you do things with computers is software engineers. Of course, they're always leading the way.
When you're in the cloud, when all the data is in the cloud, then all of a sudden. You stop pushing paper around to each other and you say, okay, I'm building an always on software system. Who knows how to do that? Okay. Software engineers know how to do that. Let's. Figure out what ideas we can steal from them.
And that was really the impetus behind almost all of dbt's product decisions. And there's like a million other smaller things, but that's the big idea.
Ben Miller: It's funny, so you don't know this, but Fundrise is building a new product, and I'm stealing your playbook trying to do that for real estate analysis,
Tristan: Oh, cool.
Ben Miller: which is exactly what you described, it's like real estate analysts push paper, they use PCs, but it's still a functionally a Word doc or a spreadsheet.
And they don't know anything about software engineering and all their practices follow from that old way of thinking, even though obviously it's digitized, nothing they do is in a cloud. It's all in an Excel spreadsheet. There's like a step change coming around that. That's why I've thought a lot about your business.
Cause it's a model. We're not just using the software, but there's another abstraction. Like there's this model of applying a more rigorous way of doing the work. Not just the support around the work, but the logic around even the words and data science are much more precise ingestion and transformation and the ETL and all these things that in non software world are just fuzzy, gathering data and what they're doing to it, the joining.
And so that kind of language, the language is precise, the logic is precise, the practices are robust, and it's just missing from most of the financial industry.
Tristan: We've built a piece of technology. Certainly we've built a company, but I think the hardest thing that we've had to do was build a community of users. And. It's exactly for the reason that you're talking about. We essentially started with the idea that everybody needed to change the way that they were working.
And that is not a great way to start a technology business because most people that have been through this process can tell you that it's incredibly hard to get people to change the way that they, and honestly, it's one of the reasons why I think most people would not have started a technology company and built dbt.
And I actually talked to venture capitalists at the beginning of the process and they said, We're just not open to funding something. That's on user behavior change. We want you to build a product that works with users and not tells them to do things differently. If you're going to take this big group of users and say, guess what?
The way that you're doing things right now is probably not serving you as well as it could. Then there needs to be some way to take them through a process where they see the light on the other side and they deliver business outcomes or they get promoted or like whatever the thing is that makes them freaking fall in love with your idea that they will then turn around and infect the next person with it.
Ben Miller: Yeah, I'll come back to you when I have this beta product and we can talk about it. I didn't mean to tease it yet because I still have a couple more months of work here. But let me transition to the next part of your story because you just mentioned it, which is venture capital. Again, my view of you is you're this insider outsider because you're obviously superior to me.
You look like an insider because you're. In the center of this data ecosystem, lots of people who are peers and some of these peers now are Ali or whatever's got like a 50 billion company or something, but you're also clearly an outsider and not Silicon Valley person. You're not there. You're not like them in so many other ways.
and then you went and raised money in the peak of the bubble. You raised a lot of money. You've been through the process and I'm interested in your views of venture and tech, whichever, what do you want to do? You pick which one you want to do first. Cause there are observations about the tech industry.
Like tech industry has changed a lot. It changed, God, I want to say politically, but I'll just say just changed like the nature of it's changed a lot, it's matured, it's grown. It's not like a bunch of people in garages anymore.
Tristan: Yeah, it's the biggest industry in the country at this point. I think if you want to measure by percentage of the S& P 500. Let's do venture first. I have more well developed thoughts on venture and maybe I can spitball with you a little bit on tech, but I really don't like the fact that being a founder is a thing that we as a culture now both accept and revere.
It's now cool to be a founder. Why would I not want that? It's not that I don't want more founders. There are so many good things about having people be founders and the time when there were the most founders in The history of humanity was like when everyone was a single family farmer post feudalism, but pre industrialized capitalism, because everybody was responsible for their own outcomes and that they had a stake in everything.
And they had to make really good decisions that were tied with reality and blah, blah, blah. And there's so many things that are good about that. But the problem is when it seems like it's such a cultural stamp on your resume. Now, I founded a company. I raised some money. I. Got some early product market fit and then it didn't work and I spun it down and then I went to work at Google.
That's now like a badge of honor. I see a lot of people getting into this for the wrong reasons and without any particular need to change the world and without any particular experiences that would allow them to do something really big. And that, I think, is very dilutive to the ecosystem overall.
There's just a bunch of people that maybe should come back to this dream when they just feel this inevitable urge that they can't possibly resist. One of my favorite things that I've ever heard a founder say was when Jensen was recording, I think it was on Acquired, where they asked him like, would you do this all over again if you know what you know now?
And he was like, hell no, n never.
Ben Miller: Yeah, I don't know if I believed him when he said that. I was like, I don't know if I believe you. You're just salty. It's your brand.
Tristan: That's fair. This has been a wonderful journey that I've been on. It's been eight and a half years now. Things are continuing to go well. So it'll probably be a lot more, but I don't know if I would do it over again. I like really feel compelled. I feel like there are humans in the world who need what we're building.
And I think that there's real structure of the data industry that was bad for users. And I understand why it was the way that it was, but it wasn't good for the people actually in the business. And so I felt like I. Really needed to do this thing. And so I think you should feel that compelled to do the thing.
Cause it really hurts along the way. If you want to keep sticking with it, you got to be willing to go through the pain. And I think the VC often feeds into this cycle. We should revere people who start companies, not people who start startups. If you want to. Build a small business. Hell yeah, the world needs more small businesses.
I don't know that it needs more VC backed startups because the incentives are often like very misaligned between the founder who starts the thing and could really, success could look like a 2 million, 3 million business and outcome. You raise money and all of a sudden, It's either IPO or bus. And so there's real misaligned incentives between, I think, many people who could start a socially productive business and a satisfying business for them and the needs of the venture capital industry.
Anyway, I'm saying a lot. Let me pause there.
Ben Miller: Yeah. He said something really great at the beginning. I'll come back to, but the lack of alignment, you know, it as an operator and then it's inside the community, you hear stories that no one else is hearing, most people just don't understand. The significance of what you're saying. And then you said it's IPO or bust, which is crazy concept.
There's a thousand, 2000 unicorns now all worth more than a billion dollars. None of them are going public. Maybe like 10 of them will go public or something, maybe 20, but like most won't. So you said two things that I think it's hard for people to appreciate how significant. The meaning of that would be, can you expand on this sort of lack of alignment?
Like, what does that mean in practice?
Tristan: I bootstrapped Fishtown Analytics for three and a half years with my two co founders. I wrote a 10, 000 personal check into the business checking account and went from there. That business is really cool. If you can make money, you can have all the control in the world, you can decide how you want to shape work and your life and what kind of societal outcomes you want to create in the process, you can create your own culture, you can do all this stuff.
But the minute that you raise a dollar a VC, you all of a sudden have a fiduciary responsibility to these shareholders. And this fiduciary responsibility is really enshrined in corporate law and case law, where You must, in fact, make profit maximizing decisions when you have outside investors. And so, all of a sudden, you have a narrower space to operate in.
If you want to just take half your profits and donate them to the local cause that you care a lot about, that's a thing that you can do when you have a tightly held business. organization. When you take on VC, you just have fewer choices. But then the business model of VC is that most of the companies will fail and a few will return power law outcomes, quote unquote, return the fund.
And VCs are not looking for a middling outcome. They're looking for most of their companies to run into a wall and die. And then a few of them to get really big and founders outcomes would actually be better. If they could have a little bit more of a median look to them because humans actually only need so much cash to be happy.
Ben Miller: Yeah. Just to add some of my experience to that. It took me a long time to figure this out. Before I raised venture, I would go around talking to venture funds and I'd say, okay, let me just ask you a question. Okay. Real estate market is blowing and going. What should I do? Maybe I should do more spending, more go.
And I was like, the real estate market is collapsing. It's the opposite. What should I do? You should do more. You should go. But it's because their model is a few things that go, they want that to go to the most extreme possible. And so they're just always looking for extreme outcomes and advising to that goal.
And my experience is that actually most extreme outcomes don't come from extreme behavior. They come from a lot of luck, there's a lot of serendipity, a lot of unintuitive things happen. It's really hard to create extreme outcomes. I think Airbnb, they're unintuitive, they surprise. I think ChatsGBT, when November 22, they're sort of a shock.
And their only tool of a financier is money, so they use tools to drive extreme outcomes. I've heard you say this, but You need a lot of time to get extreme outcomes. Like it's the things that are strange and then you pile into it.
But if you're have hot money, you don't have time.
Tristan: The three and a half years that we bootstrapped DBT and its community were getting off the ground. And those three and a half years were just so invaluable because we had other problems and we were able to let it do its thing. And. That organic nature of growth, not outbound emails or AdWords or all of these different marketing channels.
You get a different outcome when you slowly refine the product to have such intense product market fit that people just love the thing and you're not trying to shove it into their hands. So yeah, I think that the most valuable resource you have is in the founding days of a company is time. And you should be very mindful of how you use it and you should not give it away.
And VC is one way to give it away because you'll be on a clock.
Ben Miller: You can try to speed up time with all the marketing money, but you don't have to try to speed up time by the whole thing. You can't birth a baby in one month with nine people. You start throwing people at it, and that doesn't work either. That doesn't work for us. More people doesn't mean more growth or better product.
It actually sometimes meant the opposite.
Tristan: I don't know how far you want to go down that rabbit hole, but yeah, I have a lot of scar tissue around just realizing how hard it is to build great product experiences across very large groups of people. The initial Version of dbt, the open source version of dbt got refined over many years and with very few people, but the commercial version of dbt came out of more of the venture funded model where we needed to build a cloud platform.
It required a bunch of humans to do that, and we had to bring it to life very quickly because a lot of people wanted it and were willing to sign up. And then all of a sudden you have this big team of engineers and designers and product people, and it's very hard to communicate. What was excellent about that original product and how do you translate that into the new product experiences?
And that's not to say, I think we hired amazing people, but they didn't all have 15 years of data experience. They hadn't consulted with dozens and dozens of different companies to help them build their analytics stack. I can't actually replicate the founding experience.
Ben Miller: And one of the things, the truism of nature is that you can tell somebody something that's true, but if they haven't experienced it, they don't believe you.
Tristan: It's like you're building this mental model in your head and it's a very thin lesson if somebody just tells you that something's true. But if you have experienced it yourself, you build this really dense mental model. Why is it true? And when is it true? And when is it not true? And all of this other stuff.
Ben Miller: Yeah. And it's hard to get that communicated through an organization, hard to do that remotely, through when people work from home, it's hard to do that when somebody comes with decades of other kinds of experience that's maybe contrary to it. Their interest may not be totally aligned with that because their job is to punch the button at the factory or whatever the thing is that they feel like is the thing that's going to get them rewarded.
And the product wants to be organic, right? Great products, there's something about them that is not created synthetically by a bunch of people with MBAs, Excel spreadsheets that analyze how best to do it.
Tristan: Without going too far afield here, I grew up in the era of the NES and the Super NES. And my favorite game that's been made in the last decade is a game called Stardew Valley. And it doesn't really matter that much. It's a game where you farm and you have plants and different farm animals and stuff like this.
And I play it with my kids. But the point is that many of the early NES games were built by extremely small teams. Just like a couple of people. And Stardew Valley was built end to end, one person, and I think it's such a beautiful experience. It all fits together really nicely. I have become totally turned off by the mainstream game industry, because I think these games are created by literally thousands of people, and they become very cookie cutter and not that interesting.
Ben Miller: And NES and Nintendo, just for people who are not our generation, maybe. But it's a challenge also because you're a little bigger than us. We're 250 people and you're, I think, 350 or 400 or maybe more now.
Tristan: We just crossed 500.
Ben Miller: Blown and grown, but it's hard to have scale and that product development. It's just like scale wants routinization and wants to, what you said, like factory driven processes.
Tristan: I think that we are extremely lucky that now that we've been at this for a little while, We have some pretty long tenured folks who have been through many of the iterations and learned alongside of the founding team. And my two co founders are still at the business, still building product. So we have a bigger and bigger group of people who really get it.
And I think that the only way to scale your ability to Create great products is you have to figure out how to retain a bunch of people who are the brains behind it all because certainly not every single person is going to have been through all of that, but you need to grow that core group.
Ben Miller: You're singing to the choir because our core founding team, we've been together now it's 2012. So what's that for 12 years? Every year we add a few people to that core. That's the core nuclear power that drives the submarine.
Tristan: Yeah, isn't it so much fun when one of those new folks, you realize that they know more about something than you do.
Ben Miller: Oh, it's the best. Then I don't have to, I don't have to go near it. And you're like constantly looking for the people to add to that nuclear core. And everybody in theory wants to be part of it. But I'm like, go to nuclear core. It's like hot and tense. So Not everybody appreciates it once they get in there.
That's not so fun. Before I ask you a question about tech industry, I just want to go back to something you said, you said something really thoughtful, maybe you've had more thoughts on it because it was really interesting. You were talking about this idea of founders and there's an era when everyone was a founder, post feudalism.
And pre industrial revolution, and that's basically when there was a country was founded, Founding Fathers. I was thinking about what you said and how that's so much of the ethos of what we want America to be, but most people, I can't remember exactly the description you just had of what it meant to be a founder.
Like you had agency and you had to be realistic and live and understand the whole thing so you could make good decisions. I have no idea where my food comes from. I don't know who grew it. The society is so far from that at the moment, but our country still has this ethos of, this is what we want to be.
I want to beat this world of pre industrial revolution founders.
Tristan: Okay, this is stuff that I have not actually said out loud to a lot of humans. It's thoughts that you generate when you are reading. And certainly apply across the board, obviously, at the same time period, the entire American South was running a very different economic model. And just want to acknowledge that.
But if you go somewhere, I don't know, in the Northeast or like Ohio or the places that German immigrants tended to go where, whatever, there was a lot of this family farm mentality and there were a lot of land grants. And if you. met the requirements of a land grant, which typically included clearing some amount of acreage and building a road or something like that, then the government would just give you land and you would go out there and plant yourself.
And maybe you would build a family and it was a hard life, but you had no one. Accept yourself and the people like immediately around you and that experience I think is very similar now It's higher stakes than because most founders can just go and get a job if they need to but the thought processes Would instill in you I think would be very similar to the thought processes of an early stage bootstrapping founder And I have sometimes having been through that experience.
It is sometimes hard to Truly understand the brain of somebody who hasn't been through that because I think this extreme self reliance makes your brain work a little bit differently and I find myself this fierce advocate of finding ways in your life, whatever they are, to express that extreme self reliance.
So I am really into backpacking. You go out in the middle of the wilderness, it's just you. I'm really into sailing. You sail out into the ocean, it's just you. All of these ways where you can figure out how to express complete control over your own environment.
Ben Miller: Let me go back to Earth for a minute. I feel like I'm interested in talking about that.
Tristan: Sorry, you took us there, I just followed your lead.
Ben Miller: I know I'm interested in it. Let me just go back to the data for a second, because I feel like A year ago or six months ago, everyone was talking about data, data is oil, AI needs data, AI needs algorithms, AI needs compute.
And like now people talk about AI needs energy. I tried to do a presentation on what data was because it doesn't mean anything. People's minds actually, it's just so abstract. And I even think the data that the AI world's talking about is a different kind of data. Usually everything that software was dealing with was the tables, they call it structure, but it's a table.
It was like rows and columns. And essentially, AI actually is really bad at rows and columns, at least gen AI. And then I started thinking about punch cards. I don't know if you've read the IBM history biography. It's a really fun story. And the guy, Watson, he's a horrible human being.
Tristan: Selling computers to the Nazis.
Ben Miller: No, not just that.
There's just this whole life story. This crazy egomaniac, the selling, that's actually canard, like kind of not exactly true. He had a subsidiary in Germany that they had technology way before, and then they went off on their own and sold it. But he actually, there's an egomaniac. The Nazis actually offered to give him a loan.
This award, he went there and was awarded Order of the Eagle in Berlin by like Goebbels or Hitler or something. And he thought he was going to make world peace for them. He was going to be the one to go and settle this all out in 1938. That's how big an ego it was. He thought he would be the one, like no one else could do it but him.
And then he was just obviously looked really bad a short time later. But anyways, the point is that punch cards are electromechanical technology. Then you end up with mainframes and PCs and cloud. And so what we mean by data, the medium and the content, what we think of content changes when no one thought of television as content in 1750, or they only thought of it as a written word, like the actual definition of information.
It evolves with the technology. And I was thinking just about how much you saw it evolve from with cloud and modern data stack, and then how much it's likely to evolve again and affect your business and affect everything in a way data is just like all knowledge to think of data as knowledge and then recognize.
How significant way to appreciate what you're doing and what's happening in the sector. AI is able to understand knowledge in ways that we never saw computers understand it.
Tristan: I'm so glad that you used the word knowledge. I think that there are too many people that build enterprise technology by saying, Here are the types of technologies that are out there, and maybe you look at a Gartner Magic Quadrant or whatever, some like existing definition of a market, and that typically maps to a line of spend somewhere in someone's budget, and you say, ah, I know how to build that thing better than the last person who built it X number of years ago, and that's the like, build a better mousetrap.
I think that the more interesting. way to build technology is to, from the ground up, understand what problems somebody has and completely and totally ignore what the constructs are in the space. Just say, what's the best way of solving that problem? And in data, we have been obsessed for a long time.
Especially in the enterprise space with all of these very dusty terms, things like master data management and data transformation and data governance and all of these things where you're just like, I don't exactly know what that means. It's not English. It's just industry speak where you've read enough Gartner reports and you're just like, ah, now I understand what you're saying, but it all comes back to knowledge.
And the way that we frame the mission of dbt labs is to Enable data practitioners to create and disseminate knowledge. And that's it. And all of dbt's product surface area is oriented around either creating or disseminating that knowledge, but knowledge has to be linked together to other parts of knowledge, which is the dbt DAG, and it has to be easily accessible to many people, which is dbt Explorer and blah, blah, blah.
But you have to start from the perspective of what we're really doing here. Okay. is creating and disseminating knowledge. And you were talking about how forms of knowledge that we can address in computers have changed. And yes, historically, most knowledge that computers have dealt with has been in rows and columns.
You would not imagine an Excel spreadsheet where there was only one column and every row in that column was a picture. That's just not what we use spreadsheets for. But all of a sudden, a tremendous amount of Data that is being addressed by AI driven systems is kind of in that format. It's actually a bunch of image files in an S3 bucket somewhere, or it's a bunch of wave files in an S3 bucket somewhere.
And yes, we have to use different modalities of compute to address that, but we need all of this to come together to really understand the, the flows of knowledge inside of any given organization.
Ben Miller: The opposite of knowledge is noise. There's no signal. There's no meaning inside it, right? Just zeros and ones or static. How do you go from noise to knowledge is you have to have the relationships among the Components have to somehow stick together and mean something. So whether the letters form a word, or the pixels form an image, and trying to figure out how you take unstructured data and put meaning on top of it.
In a way, that's what dbt does for structured data. And so I've been curious how you're thinking about that because we're building an AI product, as I mentioned. And I started with actually structured data because I can understand it and then mostly what the industry, real estate, financial industry is, that's the engine or the gasoline of the industry.
I don't have a logical abstraction for the non structured data yet. There's no DBT for that.
Tristan: There are products that will allow you to do dbt type things for unstructured data. Unstructured data has so many different formats. I'm thinking about a product, I think it's called unstructured. io or something like that. I don't know that it would let you address arbitrary forms of unstructured data.
Anyway, we are not really doing anything in the unstructured space today. And. That is partially because the transformations that happen inside of a transformer, like a large language model, are in some ways replacing for unstructured data, the transformations that you might do for structured data in dbt.
It's a really interesting thing to try to figure out what is going on inside of a transformer. And, oftentimes, if you have a image classifier that's saying, is this a cat or not? You can actually map individual neurons to, ah, this group of neurons is trying to identify a whisker or an eyebrow or a cat mouth, kind of evolve those ability to identify those parts of a cat.
And if there's enough of them, then they say, I think it's a cat. Maybe you don't actually need a dbt type thing in that space as much. I don't know. That may or may not be true, but one of the things that is currently being talked about in our space, is it possible to train a large language model, get business intelligence out of structured data, without doing all the work people currently do to try to figure out the answers to these questions?
And currently the answer is absolutely no. AI cannot replicate human analysts. I don't know how long that will persist.
Ben Miller: I heard on your podcast, the guy who put events streaming through a transformer. And I was like, I went red, hit right about it. I was trying to figure out, could I put other kinds of data into a transformer? And would it be good at recognizing the patterns of meaning? But if you have to train on an output, right, a good outcome feels like you'd almost had a transformer transformers in a way,
Tristan: It would be super weird to try to train from your data set via human feedback a revenue classifier. Was this your revenue last month or not? I don't know that the concept maps super well to structured data.
Ben Miller: yeah, you need the first generation AI to get to a place where it can start to do things. Yeah. I'm sure you're going to start adding features.
Tristan: Our bet is that if we build the platform correctly, you can take all the discrete tasks that humans are doing and figure out how to get AI to do those discrete tasks one by one with human oversight. End. Not that you just completely rip up all of the scaffolding that we've built to make current data pipelines.
You use it, and you have AI, for example, transform these three data sets into this other data set, and write the SQL to do that, and then write the tests to do that, and then write the metrics on top of that, and have a human check it at every step of the way. Make it more efficient, but not assume that there's going to be some magic behind the curtain.
Ben Miller: Yeah, what worries me about that, and that's just as a metaphor for I think most of the stuff people are using AI for, which is you're taking a discrete task and having AI do that discrete task, is that the paradigm shift from PC to cloud or whatever you want to call it, it created a modern data stack.
And the paradigm shift before that created the previous ecosystem. And the beginning of any ecosystem, beginning of any paradigm shift, rather, is that you start, and you probably know all this, but like they originally, when they created electricity, they just plugged in the existing discrete tasks because they used to have it all powered by a single like windmill.
Tristan: It wasn't particularly useful.
Ben Miller: Wasn't that much better than they were, we organized the entire plan or electricity, then they got the benefit. For example, I would say when I used to, when I first got up my car, I learned to drive, used to drive the bus routes. That's the only way I knew how to get anywhere.
Tristan: Oh, that's funny.
Ben Miller: So I just drove the way the bus used to take me to school or something.
So you don't understand the new paradigm and what can be done differently at first. And you just replicate the old, as you said, old paper way of thinking before there were cloud. And so I feel like we're in that stage where AI is going to create a new paradigm. The moment it's in this phase where we haven't changed the way we think, we're thinking the way we used to think, and AI is helping us do it discreetly better.
Tristan: Yeah, you're totally right. I don't want to make an argument that's a long term position that we should stay in or that the market should stay in. But we've stayed in that version of data for, I don't know, 20 years. And so these intermediate periods can be long and productive. And the question is, what is the underlying technology ready for?
So I was just talking to a CEO of a publicly traded software company and. They were sharing how they're offsetting a lot of support ticket volume to AI. Their ticket volume is growing, but they're able to increase the percentage that they're moving to AI such that their support team is flat. And they expect it to continue to stay flat for a little while.
And that really improves their margin. And what they were saying was that. AI is turning out to be, like, pretty good at that use case right now. Okay, great. That's saying that AI for a language based task and a discrete task that has a very clear body of knowledge it has to operate from, like the documentation of a software product, it can do it.
But we're just not yet at the place where it's completely and totally changing entire complex areas of knowledge work. We're just not quite there.
Ben Miller: So our fund, we have a fund, we invested in dbt, but we also invested in open AI, Databricks, Anthropic. I feel really confident that the next couple generations are going to be a lot better. GPT 5, GPT whatever. I don't know.
Tristan: Do investors get to know how soon the next model is coming?
Ben Miller: No. If I look at our fund strategy, we had invest in AI, invest in picks and shovels, and picks and shovels are a lot of these inputs like dbt and Databricks. We're a user, we use all these products, building products with OpenAI and Omni. We adopted Omni for our BI, we shifted from Tableau. It just seems like a really good bet that the hockey puck's gonna be there.
The maximal view of five years from now or seven years from now, I think that's impossible to guess on it.
Tristan: Totally. And I have seen Sam speak and Sam Altman speak in small group settings. And he's really made a brand for himself in a way that is really funny. He constantly trash talks. It's just like, ah, this is the shittiest version of an open AI model that you will ever see. You get it. Things are moving very quickly and so everything current is going to be garbage very soon.
But I've never heard him predict anything more than 18 months out. Because he's just So respectful of the idea that progress is happening rapidly and exponentially in a way that the scaling rules don't produce linear outcomes like behaviors happen that we actually would not have anticipated. And so we just don't exactly know what anything's going to look like in more than 18 months, which makes it very hard for people like you and me who are trying to build products and trying to anticipate what the future is going to look like.
And it's a period of very high change.
Ben Miller: Yeah, and I think that the tech industry is probably most exposed to it. At the end of the day, the less technological your business is, like probably the less it's going to impact you. And the more knowledge driven, the more software driven, I think it's like the more for service areas for AI, it's not going to be building buildings anytime soon for my business.
Tristan: I bet you Purdue is still gonna make chickens.
Ben Miller: Yeah, and for us 18 months, it takes 18 months to build anything really to build a good. So the cycle time, the fact that NVIDIA and OpenAI are operating on this, some large magnitude faster than normal hasn't yet affected the economy or my business yet, but you can see it coming.
Tristan: Totally. There's this weird intervening period where one of two things will be true. One is The entire world will change and we'll rebuild so many things that we think we know today. The other one is that we will have spent way too much CapEx on GPUs. And it'll be like dark fiber in 2001, uh, where it doesn't get used for a decade until we find productive uses for it.
Ben Miller: I have a presentation on board all team members and I have these operating principles and one of my operating principles is about how most change happens nonlinearly. The world times linear type experience, everything incrementally, but everything big changes nonlinearly. But every time we project the future, we always project a linear future because it's going to be more or less the same over the next 10 minutes.
And, if you think about a memory or memory between an hour ago and five hours ago, and I always say people, my examples are always like the pandemic. We went through the pandemic with total nonlinear break for me. 9, 10, 9, 11, I get obsessed with nonlinear breaks because they're just so much more significant.
Building a business, our business, we've had like four or five nonlinear breakthroughs in the business. And then just 10 X our business and everything else was great. It's growing whatever, 50 percent a year or something, but then you have a nonlinear break and a 10 X is it. And then you're just like, it's hard for me to care about the normal stuff because usually the nonlinear stuff blows up the code, you throw the code out.
You have to retrain your customers. You do all these things that are painful. I'm watching AI. The idea that it's like not going to be nonlinear break to me, just defies every single. thing I've experienced as a founder they actually happen way more frequently than people believe. I shorted the market February 2020 because people couldn't believe there'd be a pandemic.
It's just too hard to imagine. It's just too hard to imagine AI. I don't think we forecast it well.
Tristan: I could not agree more. If you live in Philadelphia, you've probably heard that Ben Franklin left whatever money he had left at the end of his life to some charity. He left out this money and he invested it somewhere and you think, gosh, this compounded for 200 years and this must be so much money now.
And the answer was no, it turned into 20 million dollars, which is a lot, but you would think if one of the founding fathers of the country
Ben Miller: He was wealthy. He was a wealthy guy back then.
Tristan: It was growing at like 2 or 3 percent a year. I guess the non linearity matters, but also it just requires a big effect to generate big numbers. It was very striking to me when you said you're growing at 50 percent year over year.
And yeah, it's almost hard to care about that. What you really need is these 10 X's. And I think sometimes it's really just very hard for folks who haven't started a company before to understand just how big of a change it is. They need to make in the world, any of it to register on a scale that is useful.
You can just put in a lot of work and without multiple orders of magnitude of improvement, you won't actually get anywhere. That's going to be better than just going to work for another job.
Ben Miller: It almost sounds crazy when I say 50 percent is not good enough, but I have a slide when I say it and it's at 28 investors a month. At the end of a year, we had 75 investors a month. It's 50%. Growth is great. And then we did this nonlinear break and we got a thousand a month. And so you need both. The incremental day to day process, the mundane, and then you also need the non linears.
Anyways, we've now passed an hour. I'm enjoying the conversation. Definitely have a couple of things I have to go offline to talk about. So I'm interested in, in following up a couple of questions. I'll tell you at least one story I have from investing in some of these companies. But anyways, it's been awesome, Tristan.
I really appreciate it.
Tristan: Yeah. Thank you. This has been a lot of fun. You run a very interesting podcast. We went all over the place. And before we even got on, just so that listeners realize this, you were talking about a book by Thucydides. I don't know what your reading list includes, but it is broad and I'm jealous.
Ben Miller: Thanks again. It was great. Onward.
