Spatial Data and AI: The Next Frontier in Technological Innovation with Paul Copplestone
Explore the journey of Supabase CEO Paul Copplestone on Gradient Dissent Business Podcast, discussing AI, database challenges, and spatial data innovations with hosts Lavanya Shukla and Caryn Marooney.
Created on December 21|Last edited on December 21
Comment
🌟Dive into the remarkable journey of Paul Copplestone, CEO of Supabase in this episode of Gradient Dissent Business. Paul recounts his unique experiences, having started with a foundation in web development before venturing into innovative projects in agriculture, blending his rural roots with technological advancements.Throughout our conversation with Paul we uncover his coding and database management skills, and learn how his diverse background was instrumental in shaping his approach to building a thriving tech company.
This episode covers everything from challenges in the AI and database industries to the future of spatial data and embeddings. Join us as we explore the fascinating world of innovation, the importance of diverse perspectives, and the future of AI, data storage, and spatial data.
Thanks for listening to the Gradient Dissent Business podcast, with hosts Lavanya Shukla and Caryn Marooney, brought to you by Weights & Biases.
Timestamps:
- 0:00 What is supabase.com?
- 03:07 Exploration of exciting use cases
- 06:15 Challenges in the AI and Database Industry
- 12:30 Role of Multimodal Models
- 16:45 Innovations in Data Storage
- 22:10 Diverse Perspectives on Technology
- 29:20 The Importance of Intellectual Honesty
- 36:00 Building a Company and Navigating Challenges
- 42:30 The Impact of Global Experience
- 54:00 Integrating Spatial Data into Digital Systems
🎙 Get our podcasts on these platforms:
- Apple Podcasts: http://wandb.me/apple-podcasts
- Spotify: http://wandb.me/spotify
- Google: http://wandb.me/gd_google
- YouTube: http://wandb.me/youtube
Transcript Via Trint
Lavanya Shukla [00:00:00] Hello, my name is Lavanya Shukla, and this is my co-host, Kane Mooney. And you're listening to Great in the Business. Welcome to the show.
Caryn Marooney [00:00:10] Welcome, everybody. We're thrilled to be here with Paul. He's the co-founder and CEO of Supabase. What is Supabase? It's a opensource firebase alternative. For those of you that are even thinking of building, you can build in a weekend where you can scale to millions with Supabase. So welcome, Paul. And why don't you tell people what you do in your words? Because they'll be way better than me.
Paul Copplestone [00:00:37] Yeah. Thanks, Karen. And thanks for. Thanks for having me. That's a great intro, actually. You kind of know that we're at a platform for builders. We provide tools specifically for for builders who want to get started very fast. Most importantly, we provide the database layer, so we offer a PostgreSQL database. And then on top of that, we've got a couple of other tools an authentication system, a storage layer for large files, some edge functions for heavy compute, and then some auto generated APIs. And then most relevant for this podcast is we also have this vector offering powered by piggyback, which is an extension for PostgreSQL.
Lavanya Shukla [00:01:23] So one of the things that our audience is trying to wrap their heads around and say, we had the ML up stack and now we have the Elm up stack. I think we're close to getting the multimodal multimodal models out, so we're going to have the aiops tactics. What do you think that stat looks like and then how this would be added to it?
Paul Copplestone [00:01:42] Yeah. Well, so we're extremely popular for the builders. So, you know, for for reference, just to put some numbers around it, we launch around 11,000 databases every week. And so a lot of them are launching with maybe a frontend framework like next year or like a React type framework. And so it's largely like a JavaScript and database system. So I think that, you know, we've got a different view from probably what a lot of the ML engine is saying. But you know, actually the use cases are converging. So what started in super base with a lot of JavaScript is now actually moving a lot more towards Python uses and different use cases largely driven by Elm type use cases. But you know, what people are building is roughly the same reason we see a lot of these applications that are pretty similar today to everything else. A lot of them kind of trying to discover what is the new app that will be built. And at this stage, they kind of look like three different things. That's always going to be around, like chat with characters, chat with your dogs or some sort of generative character system.
Lavanya Shukla [00:03:07] In order to most like I guess you have this really interesting new into the apps that people are using using elements like this, the obvious ones that you mentioned. But have you seen anything that's like really exciting that people none of this are based on?
Caryn Marooney [00:03:21] And how much of them are not okay for work? I'm curious, just like when people are building, where does it start? Where does it go? I mean, you're seeing all the early stuff, so I'm intrigued.
Speaker 4 [00:03:32] Yeah, well, a lot of them are.
Paul Copplestone [00:03:37] Of course, self-aware, but the most interesting ones actually are not able to equip go one on our platform that scaled from 0 to 1 million users in ten days. So that's almost as fast as charity day. And and it's not not safe work you know chat with Garrett know exactly how it is.
Caryn Marooney [00:03:57] Let's talk vector databases and I know can you explain to people. How you think about multimodal and you know, in layman's terms how you can add video and other other modes into the database, like how do we think about that?
Paul Copplestone [00:04:17] So multimodal is definitely becoming a thing now, especially with Chai adding it. And so everyone's far more interested in what they can do and how they can build. We're seeing a lot more. So out of those databases that we're seeing where they're storing embeddings, I think like maybe last week or the week before, we're checking how many of them and what type of embeddings that using 98% of them are using. Editor by OpenAI. So almost everything is actually storing just tags. And then now there's a trend upwards or clip a lot of people starting to use clip to embed images largely. And then of course, one of the products that I said was storing large files. You, of course, want to type them, store them separately and then take the embeddings around it and store that in your database so that you can do similarity. These are quite interesting use cases for us, but I've got to say it's very nascent. I'm quite excited to see what people will build around it and what the different use cases are. We're seeing a lot of at the moment the similarity for images largely around not safe for work because you might want to segment what you think is a not safe or work image out of your embeddings database, maybe by a partition and remove that data so it doesn't slow down. So your indexes, your searches and you don't have to, of course, pay for that storage.
Lavanya Shukla [00:05:54] But I think applications like that predictive modeling model is going to come. It's going to be the big thing very soon. How can companies think about other applications for Multimodal for themselves and how can we correct themselves and why would they want to use Superbe?
Paul Copplestone [00:06:10] Yeah. So I think there's kind of two ways to think about it. The generative aspect and then the rag type approach is very nascent in the rag use cases are very telling.
Lavanya Shukla [00:06:25] So what I guess just in case people don't know, they might be sure.
Paul Copplestone [00:06:29] So Rag is retrieval augmented generation and that's where you might find a relevant set, let's say for text. If you're going to generate an answer from your documentation, you'll have someone ask a question like, How do I use weights and biases? And then you take that question, you turn it into an embedding. Then you search your database and you might find the top results, say five results. Then you send all of those results and the question to a generative model and it will spit out an answer based on the results that you have. So it's very useful for anything that's in the private domain where an alum wouldn't have ingested it. So any private data. And so, yeah, when we're moving into multi modal, then what you might want to do is find relevant images. For example, let's say they want to generate a, a r image of a cow and a duck or something like that. You might want to actually find five images that kind of look like that so that it can do it more precisely than just the model has been trained on. So it might be relevant to your specific conduct that you want to generate.
Lavanya Shukla [00:07:53] Just select fine tuning. You're saying that.
Paul Copplestone [00:07:57] Just a bit like fine tuning. This is like another tool on the belt, essentially. You can either let the large language model do it or whichever model that you've got. You can add like sort of in Hansard with rank based on some data that you got, or you can then fine tune it based on probably a lot of interactions that you've had with your community. And we're seeing a lot more of this at the moment as well now where people are storing all of the questions that that people might ask, then they're seeing the ratings that might be generated so people can give a thumbs up, thumbs down. And then you use those results to fine tune later. So, yeah.
Lavanya Shukla [00:08:42] I can selfishly ask you like this continuously, like every time, every two months to see changes. But like, do you see people buying today? Or do you see people using drones to tune the models? And also, like just to give context, those of us who come from the machine learning space find it like the obvious answer is we got a fine tuned. We know how to paint you. We have the infrastructure. But most of the people doing this don't know how to train them on. You don't have the GPUs. Like, what are you seeing?
Paul Copplestone [00:09:10] Yeah, actually, in our space mostly people are just improving the prompt sorry vowels on the prompt if that. Actually, sir, they're very much just waiting for the models to to improve. And the way that I kind of see the lay of the land at the moment is kind of you've got several layers, right? So you've got the application, then you've got the models, then you've got the databases, then you've got those systems below. And as you go deeper and deeper, sort of the churn is lower. So applications are churning very fast. They're getting spun up. We've had one open source template, which is kind of Chat with Your Dogs, which has launched 5000 databases on super base. So a lot of people doing that use case even for that specific open source app. Then you've got the models which of course are improving at a rapid rate. Then you've got the databases, which all of them are improving, both traditional and specialized. And then of course, you've got in video down the bottom who's just winning everything. All about.
Lavanya Shukla [00:10:18] Like you didn't start out building a company to build better databases, but you were able to pivot to it pretty fast and recover out of the bleeding place. What is so special about how you think about products and how you think about like running your org? That makes me so adaptable to new be changing.
Paul Copplestone [00:10:35] Yeah, a lot of what we think about when we're building is just providing the sort of core primitives that any, in fact, any business might need or any developer might need. So the ones that we said like database or calls, storage, file storage, these ones are kind of key to everything that, you know, almost every business. So not a specific use case, but very general. And then on into each of these products, we build extensibility. And luckily with Postgres, it's got extensibility built in. So we're always thinking, you know, what will develop a need for the next ten years and how do we make it extensible so that no matter what the next wave is, this current wave is largely around embeddings. But ideally it can be kind of added very easily as a primitive on top and sort where we literally merged it within a week into our platform, built some docks around it, and within a month people were building building with DG backed up.
Lavanya Shukla [00:11:42] Yeah, it's interesting because it's always like you're building your Lego blocks or businesses and then they're very extensible. So you can both. So the only advice really how different startups can think about what their version of the Lego Block is. And then second question is here how do you think about hiring the team? Because legs, how could your team with this adaptability in a week it without you know everything can do that.
Paul Copplestone [00:12:06] Oh yeah well I guess to start from the second one. That one's a funny story. So I actually got this email from from an open source contributor and he said, oh, you know, I'm, I'm using super base and I really need this extension and here's a poll request. Can you merge it? And so I can start building with it. So I got on a call with them and and I actually didn't know about PGA Tour at the time. And we told her and then it seemed cool. So we merged. And let's say two days later he had kind of done something. And then the first use case was actually this chat with our docs. Now you see it everywhere. But I think we might have been one of the first to do this system where you ingest all of your dogs and then you can do this generative answer question and answer using Rag. And it was done by this contributor who then we ended up hiring. And so that's how our whole team functions. Actually, we're completely distributed. We employ a lot of open source maintainers when people are kind of too good to ignore. Of course, we want them to build within this super base all the time and be part of the team.
Lavanya Shukla [00:13:24] Oh, I see. You know, to go into this back, the millennials are the ones who are running the campaigns now, and they the easiest to have this be like vanilla tone. And now it's say you guys are like, oh, David Holmes and like other people are becoming more casual and more authentic. And it can obviously is it going after like a knowing how to message a company. So I'm curious what you guys think about like this shift in how especially let's just look at developer companies. How should you talk to your users? What should be the tone? What's appropriate?
Paul Copplestone [00:13:59] Yeah, it is a great question. I don't know if we even really discussed like how time would be. Largely what we did was we just and to myself got online and started chatting to them about their problems. And, you know, if they were using it, we engage with them. And then, you know, if they had issues that are publicly raised, we would face them and we would consist of them. And talks are how we're going to fix it. So I think the term was just largely we want to help you build better. And that kind of has become the sort of psyche of our community as a result. We went really to focus an architect and any specific tone and just kind of spilled out the other side of like really trying to be helpful to individuals as much as anything.
Caryn Marooney [00:14:48] And I think of I knew you're asking because I've worked with, you know, Mark Zuckerberg for ten years and Marc Benioff for 13 years and other people not named Mark like Diane Green or Jeff Bezos for many years and read over at Netflix. And and I think a common mistake when you start a company is that you think there's some archetype or archetype you're supposed to be and then everybody just tries to look for lessons. And that's smart and that's good. But then when you try to be like anybody else, you just, you know, it's so everybody knows it's really painful for you and for everybody who has to watch. And so and it gets really boring really, really fast. So we think it just it takes a lot to put yourself out there and to be like, you know, initially when Marc Benioff launched Salesforce, people hated him because he's really a character. And it was like, Who's this guy making dolphin sounds and wearing Hawaiian shirts and telling us like, everything's going to change. So it's a whole I think mostly it's embrace the parts that are just you in and and then be really into your audience and who you're going to serve because that's all that matters. You know, all the other noise is noise. And you you can't focus on being liked. You got to focus on being useful. The one thing that you guys do really specifically, though, is this idea of first principles. I think of you as like a first principle company. And this for those of you that no super base, you're into launch weeks and they're very two different things. Can you talk about how you set principles? Because for those of you that don't know, Super base is a distributed company. You guys are all over the world. And I think it's really important to have this that idea first principles, especially when you're going to be distributed, which Divina and I want to talk more about, but first principles and then talk to us about launch week.
Paul Copplestone [00:16:58] Yeah, the principled approach is largely in alignment all and I think it's useful in any company, but especially around around distributor companies, because a lot of what we do is we, we leave the techies to kind of work on their own. So they're off for, for a week. They'll have one meeting a week and then they do their work. And so how do you trust that they're going to work on the right things? You just set up the context by which they can think about the product, think about whatever they're building in the right way. So we've got literally embedded in our docs the way that we think about the product. So for example, two of our principles. One is that everything must work in isolation. So each tool, the database and the off system, each one can be a standalone product. Then the second principle is that everything works as an integrated system. So even though that they can work in as standalone tools, when you pair them together, they kind of create magic. And it's a little bit like the Apple ecosystem. Like you can have a phone and you can have a book and you can have the AirPods, and when you connect them together, then you get this awesome synergy between the three three systems, but you can also use them individually and they're also quite nice. So, you know, a lot of these thinking just has to be written down on paper so that you can hand them over to say, the engineers or anyone else in the business and they can sync along and in the same pattern.
Lavanya Shukla [00:18:31] So this is fascinating to me. How are you building the API? Designing the apps like they do? They are compatible with each other as people build their own little tools that work in isolation. And what are the KPIs that you're measuring with and teams again?
Paul Copplestone [00:18:46] Yeah. So the kind of substrate for us I think is PostgreSQL, and that's the fundamental layer at the bottom. Everything else builds up from that and that's almost the API layer by which everything else can work. So when I say the whole system works standalone, actually it still needs they all still needs Postgres and large file storage systems don't need postgres then to interact between them. They can either use the schemas within the database and they can be separate, but then you can build, say foreign keys amongst them. Or if there's two systems, then we kind of think of two primitives that you have either an API or weblogs. And with those fundamentals, everything else can be joined together and, you know, interfaced quite cleanly.
Lavanya Shukla [00:19:39] And I do teams designing this together. Is it coming from top down? And also if you when you're starting a startup, you're like, oh my God, this is my baby. I don't know if I can trust my individual engineers to just go off and do what? So, like, how do you not go after that sense of control and just like, trust the people to build the right things?
Paul Copplestone [00:20:01] Well, the principles largely are the things that make sure you can trust, because it was like that at the start. For example, we had a guy join who's an incredible techie, and one of the first things I did with him was I didn't actually give him the principles. I said, come up with some principles yourself. And then I could see how he thinks about the product. And then I compared it to what I've got. And we found the Delta and actually some of the principles shifted based on his thinking. So we kind of converged and then by the time we got to the end of that exercise, we were just thinking completely aligned. And so I knew actually I never have to worry about product again. This guy knows how to build and I can trust him to make decisions even when I'm away. And the more you can do that across all of your teams. So at the moment, we're doing it inside with our head, of course, for sales. And how do we think about the principles for the sales team and how they should sell. And he's going to trickle those down to his team. Then he knows he can hand off a lot of what he's doing at the moment.
Lavanya Shukla [00:21:06] So I think if you could like tell us like, what are some of these principles for sales teams to help them crystallize?
Paul Copplestone [00:21:13] Yeah. So for example, at the moment we're extremely focused on our community and the developer persona. And you know, as you grow, people generally start to lose that focus on the indie developer. So for us, it's really important that, you know, you don't have we're going to bake that into our DNA. We do product like whereas essentially the developer comes in, they love the product and they'll bring it into the organization and sell it for us essentially is the idea. And if we do that right, we believe that we can be a really big business. If we lose sight of that, then I think you get this short term win where you quickly grow around enterprise and then your community tails off and flatlines and you can't actually go back at that point. As soon as you lose your community and you become irrelevant. So we want to bake this in. So part of the sales journey is teaching them about PLG. What are the principles like, for example, it's not sales, it's consulting. You don't have to tell them that you're going to solve all the problems. If you're not, you need to listen to their problems. Find out if we can fix it. If we can't fix it. Tell them that and tell the product team so that in the future maybe we can design some more primitives that might solve that problem later on.
Caryn Marooney [00:22:38] How do you structure your company in a way where the industry is moving fast? The tech is moving fast. You guys are a distributed company. You came from Y Combinator where you're used to like a certain amount of pressure and a certain amount of speed. How do you build your company to deal and like create its own pressure and respond to the changing technology? That's that's at a warp speed?
Paul Copplestone [00:23:03] Yeah. For us, that's our launch weeks. So, you know, we had this funny thing. We started in 2020, then we went into Y Combinator and it was the first fully remote batch because of Covid. We didn't actually Anthony were in Singapore at that time, so we had to do it at 2 a.m. every every day. But what happens during y C is that you get all this pressure leading up to demo day. And so we had like 100 days. And I remember I said to one of the engineers, build an interface that looks exactly like a table on top of Postgres and you've got 100 days to do it. And he smashed it like he just knew that this deadline was there and knew what he had to do. It didn't talk to him and just he would turn up every every week in order to have these awesome improvements. Then we got outside of Y Combinator and there's kind of this daunting future. Like nothing is ahead of you. You've just got this big open space which you into developing. And so we said to ourselves, All right, and three months from now we're just going to move from alpha to beta. And it doesn't matter what we ship. We just ship, ship, ship. And then on that day, we just changed it and we said we've moved from Alpha Beta with a kind of landing page. And we saw not only internally was everyone motivated towards that day, we saw this huge spike incline on the curve. So everyone started launching more databases, new databases. The community grew by 20%, I think, overnight. And we thought, oh, you know, that's actually quite interesting. We didn't we literally did not change anything on there or launch anything. We just changed this kind of positioning. And so we were like really excited by that. And we went away and thought on, you know, how can we do it even bigger? And so we thought, well, instead of changing one thing, why don't we ship one major feature every day for a week? And we set that goal sort of four months ahead. And then when we did that, the, you know, is the start of this kind of launch week activity. And every time we did it, we were just seeing the slope of the curve would change, the community would grow. Everyone get excited about what else they can build and mostly as well. And the community felt that they were listened to because they would actually come attend these launch week's activities and that's when they would see the things that we were shipping. You know, it's very hard to get your new features in front of the developers, even if they're asking for it. And so that was the way that we did it. And it's become really a part of how we run the company now. We are, I think, coming up to our 10th launch week and in a month or two and we aim for them every quarter. So it also helps with our company planning.
Lavanya Shukla [00:25:53] As you think it's going was uploaded the same thing the WWE DC and like it's nice to have the imposed self-imposed deadline. Can you get to see an insane amount of amazing companies? Are you seeing other examples of like, how can you stay motivated?
Caryn Marooney [00:26:10] I think that the deadlines are incredibly important because it's just too daunting. And there's. When do you call it done? When you're building because the people inside the company want to build a great product and they'll be builders and there be publishers and they'll be iterators and they all need to know when it's pencils down. And it's incredibly hard to decide when it's pencils down. So some companies like Salesforce have Dreamforce and other companies have these external events. Some companies develop more internal events. So a big deal inside Metta. Facebook, when I was there is that they have these internal all hands which they use is forcing functions for the internal groups to have deadlines that they're not quite ready to show the world. But when you're showing them to your peers, you care just as much as when you're showing them externally. So every quarter, Mark Zuckerberg would pick here's here's the things that I'm trying to drive. They may not be quite ready for prime time, but we need these are the deadlines. And I need you to have a working model to show the company on this date. And he would do that across different and it would be a quarterly thing, even though they would only have connect once a year or other ship dates. These are all hands really helped preface the ship dates. So it was it was another way. But forcing functions are incredibly helpful because if you don't have a date to work backwards from, you don't have a date to work backwards from.
Paul Copplestone [00:27:43] Super interesting from the Facebook perspective that he's kind of deciding as well the product as it grew. Did he continue to get involved on the product side?
Caryn Marooney [00:27:55] Hell yes.
Paul Copplestone [00:27:57] Hmm. And how how would he determine, like, what to what was going to be the priorities in the next quarter?
Caryn Marooney [00:28:06] There are always internal planning meetings where it was decided he thinks, in three, five and ten years. And so there were three year markers. And then you have a different cadence of how you see where you are in the three. And then it's even harder to plan the five and ten year, you know, how do you know where you are on those five and ten year journeys when you're measuring by month? But once you get into a company and your company starts to grow, people can easily get into the mindset of, am I going to make my bonus? And if you do yearly bonuses or you do half yearly bonuses, that ends up being the amount of time that people think about if they've achieved their goals. So how do you then break it into some really important things take longer than six months and they can take longer than a year. But how do you know that you're making progress on them and how do you keep it moving quickly? So that's when you set the priorities. But then you need to have some check check ins that show whether or not you think where you are on this roadmap. And it was really important to he was actively, actively engaged in all of this. You do have a leadership setting of priorities, and that's hard as your your portfolio sort of grows of what it is. And then you would also need to understand from you have different leaders of each group and they would say what's possible and then work would push you even harder. One quick story is when I first started at Facebook, it was one of the first efforts we used to do this user conference. The team was like, Mark is asking us to ship too many things. He wants too many things shipped. Go in there and tell him we only have 48 hours and we can only do ten things. So I went in his conference room. I wrote all the things on the board. I thought I was. So, you know.
Caryn Marooney [00:29:55] Organized and in charge. I handed him a pen and I said, Circle the three things that you want because we're running out of time. And he looked at the board. He's totally comfortable with silence. So he was really quiet for a while. And he took the pen and he added three things. And and I had to go back to the group and be like. Yeah. I didn't get an appraisal. I actually came up with more. They wanted to kill me, and they shipped all of it. So it was a really terrible lesson in terms of like always pushing and like I think when you believe when you've got the best people in the world and you believe in them, it is kind of amazing what they can do. But that was a real.
Speaker 4 [00:30:41] Moment of like, Yeah, I'm not managing this. It's more like he's pushing us. I get it. I get how this is going to go.
Paul Copplestone [00:30:50] That's amazing. I'm now going to do that to all of our teams. I'm going to.
Caryn Marooney [00:30:54] Be really the person who is taking that message back is not liked.
Speaker 4 [00:30:59] Thanks, Karen. Good job. Yeah, yeah, yeah.
Lavanya Shukla [00:31:04] I'd still like I don't know if I would how you do it. Positive. Say, just trust. I guess you have principles that you can ground your people in. But, like, if I had five people and it was just say, go and build whatever. I would have so much anxiety what they're going to come back with, you know, are you checking in with them? Ask for a plan before they start building?
Speaker 4 [00:31:25] Yeah.
Paul Copplestone [00:31:26] Yeah. So yeah, I guess as not completely hands off. I'm also, I would say a product CEO.
Speaker 4 [00:31:31] So yeah.
Paul Copplestone [00:31:33] I love the product and this is what I'm trying to sort of solve for developers, right? I want to build a product.
Caryn Marooney [00:31:40] You can.
Speaker 4 [00:31:40] Keep, they really find useful.
Caryn Marooney [00:31:42] You can keep being involved. You can.
Speaker 4 [00:31:45] Say, Yes, okay, yes, I.
Paul Copplestone [00:31:46] Can continue to get involved and focus most of my time on that. This is the thing that I think really differentiates our company. So I, for example, we're about to start planning for our next launch week, and I'm just going to write down a brainstorming list of what are the things that people are building. And then I'm going to say, what are the priorities? This one's going to be our 10th launch week. So it's kind of big where I'm got this theme that we want to make sure that people know that we're really ready, that it's like sort of enterprise ready, especially for 2024. We've already got enterprises building on top of us. How do we make sure that that is in the psyche? So a lot of the product features that we're going to ship are largely just emphasizing that. What are the type of things that both developers need but also enterprises need, and how do we position it in a way that's sort of exciting enough for for people to actually take take notice.
Lavanya Shukla [00:32:50] He said. The themes and then the people can come up with individual ways to exchange these and these different bucket.
Paul Copplestone [00:32:57] Yes. Or this one. So because we've got several different teams there, all you know, we've got teams and then we've got like these crosscutting initiatives. A really good example is like, let's say we've got an all star team and they are going to be on last one. They shipped multi-factor authentication and this, they know is table stakes for their if they want to build the best in the world. That's what we say to every team. What would the best old system in the world look like? Give us all of those features, stack, rank them and get to work and then they really just goes for and from top down on the stack, they try to identify what is the most important thing for all of the developers. And the way that they do it is largely where the principles come from. So we've got to make sure that they think about it in a way that can solve many problems at once, and that is well integrated with our systems. But they have a clear idea. If you hire an OS engineer, they know if you choose a also engineer with assault, he knows exactly the features that need to be shipped. And this is where actually it works well. In our model we now know better than either actually what features need to be built, and I just need to make sure that we're pushing towards the deadline.
Caryn Marooney [00:34:18] All right, Let's change the pace and do a lightning round. So, Paul, in this lightning round, we're going to ask you a couple of questions and you're going to answer the first things that come to mind. Don't overthink it. Ready?
Speaker 4 [00:34:31] Yeah. Yes.
Caryn Marooney [00:34:33] Great. Introvert or extrovert?
Paul Copplestone [00:34:37] Introvert.
Speaker 4 [00:34:39] Really?
Caryn Marooney [00:34:42] I guess that. Know? No, I don't know. I. I don't know if I would have guessed that. You seem okay. That's another.
Lavanya Shukla [00:34:50] You're doing great.
Caryn Marooney [00:34:51] Introverts. Yeah, I know. You can.
Speaker 4 [00:34:53] Hear you.
Lavanya Shukla [00:34:54] What is the first thing you ever built?
Speaker 4 [00:34:56] And how old were you?
Paul Copplestone [00:34:59] Oh, funnily enough, the first thing that I really remember kind of hacking was like, I remember opening a word document and notepad and then, like, playing with the, like, kind of code of word. And that's the first, like, really hacking, saying, I would have been quite young when I did that. But then like the first real thing because and this is a much of a lightning question. Sorry, but the first, like, real thing that I built actually was in New Zealand like focused on like an agricultural type use case and abattoirs. Yeah, we can get into that as well.
Speaker 4 [00:35:35] Get into that.
Paul Copplestone [00:35:36] Yeah. Okay.
Lavanya Shukla [00:35:38] If you could hack with anyone, dead or alive, who would you hack?
Paul Copplestone [00:35:43] Yeah. And dead would probably be Joe Armstrong, the co-creator of Line Alive would be John Carmack.
Caryn Marooney [00:35:51] Distributed or in-person?
Paul Copplestone [00:35:54] Distributed.
Caryn Marooney [00:35:56] I never knew the answer to that one.
Speaker 4 [00:35:58] And that's the.
Caryn Marooney [00:36:00] Craziest place you've ever met a customer or a potential employee? Or is there any, like, Bizarro Land?
Paul Copplestone [00:36:12] Oh, yeah. Actually, there's a funny employee story, yet another open source one. But I was email. I've been using this tool called Post Quest for many years and it's actually part of the super base stack. And I actually emailed one of the maintainers when we're starting to to just see if he's okay, that we use it within super base. I always ask first and he said yes. And he also said, Oh, by the way, I'm looking for it for a job. And this is this guy in power here who, of course we we employed. And then suddenly enough, because we're fully distributed, he had actually never left Peru. So the first ever time he got a passport and left Peru was to come to Munich for our first company offsite to meet the whole team. So I guess that's the strangest employee story, but also one that I'm the most proud of because, you know, it really speaks to our open source philosophy.
Lavanya Shukla [00:37:18] That's so Oh, that's my hiring blast. Me too. And that's why I love working with developers and literally on Twitter and looking at what our users are building and the people who are the best thing. So I'm always like, Hey, to any job, you know? And that's the kind of people you want to hire because they love your company they're excited about.
Speaker 4 [00:37:36] But yeah.
Lavanya Shukla [00:37:38] We got to get into your life story too in the okay to what questions and we will get into your life story.
Paul Copplestone [00:37:43] This is a very light thing. I'm sorry. I'm Irish.
Speaker 4 [00:37:46] Okay. Miles shorter and it.
Lavanya Shukla [00:37:48] Yeah. What do you do? Get your shit once. You get this. You're here with your board member. Like, what do you do.
Speaker 4 [00:38:00] That is.
Paul Copplestone [00:38:00] Concise?
Lavanya Shukla [00:38:03] What do you say in your internal voice that you don't out loud?
Paul Copplestone [00:38:08] Am I inside voice.
Lavanya Shukla [00:38:09] Inside your head. Like, would you say in your head that you don't see that?
Paul Copplestone [00:38:14] Uh.
Caryn Marooney [00:38:15] Man, get me out of the do. Why did I say yeah?
Speaker 4 [00:38:21] Karen's amazing, you know. Shot board member ever.
Paul Copplestone [00:38:26] I never tell her that. So now's mouse the moment. Yeah, that's a good question. Now. No, I think I know.
Speaker 4 [00:38:35] Maybe you like.
Lavanya Shukla [00:38:36] It thinking still. You're not saying it out loud. What is it?
Speaker 4 [00:38:40] Oh.
Paul Copplestone [00:38:42] That's the joy of being the CEO. You can just say whatever you want, whenever you want. Like, I just interrupt as soon as I have a thought.
Speaker 4 [00:38:50] No, I don't know.
Paul Copplestone [00:38:52] Like largely, you know, I try to make sure that, you know, if I've got any, like, real concerns, I voice them immediately. That's actually one of the most important things, I think, for a distributor company. Otherwise, like, there's, you know, there's already lag. You've got to cut it down.
Caryn Marooney [00:39:09] So far. Where did you grow up? And what parts of that? Growing up. Helped. You know, you could start a company or maybe you believe you could start a company.
Paul Copplestone [00:39:23] Yeah, I grew up in New Zealand and moved around a lot within New Zealand and for a brief and three years I lived in Thailand actually as a young kid, but then spent most of my childhood growing up in the South island of New Zealand. I went to a boarding school because I lived on a farm. And I think the thing about growing a company like building a company, my dad is an entrepreneur who, funnily enough works and and super base now, but it didn't actually film that. Or in fact, the film itself was strange working. At one point I worked in Accenture and I really did not enjoy it. So I think I've always been more drawn to doing, you know, start ups.
Lavanya Shukla [00:40:14] And what when did your passion for building start? What was the first thing you built that was meaningful and how did you transition into it?
Paul Copplestone [00:40:23] Yeah. So I started doing websites for people, but then actually I got this weird job, so I was like this pork processing company in New Zealand. Once again, as very TV, it's agriculture.
Speaker 4 [00:40:38] Where you.
Lavanya Shukla [00:40:39] Can touch, but.
Paul Copplestone [00:40:40] They, they have these abattoirs which I would have to go into and then I would literally put in like.
Lavanya Shukla [00:40:48] One is that sorry for an.
Speaker 4 [00:40:50] Abattoir.
Paul Copplestone [00:40:50] Is like where they how do I say that.
Caryn Marooney [00:40:55] I don't know what that is.
Paul Copplestone [00:40:57] Oh, you don't know. Okay. It's like where they like. It's a slaughterhouse.
Speaker 4 [00:41:06] You know, with a vegetarian.
Caryn Marooney [00:41:08] This is.
Speaker 4 [00:41:08] Perfect. Yeah. Okay. So is a.
Paul Copplestone [00:41:11] Very is a very strange job. But because New Zealand is like all farm land, basically everything. I had to go into these. Well, I didn't have to, but this was the job. I went into these abattoirs and I would install and program these like scales and these touch screens, and then I'd build this like automated system where, you know, I'd be automating these kind of like abattoir lines. And I just went through the entire business actually, and sort of building software around the processors. And it was everything from hardware to databases to to the software and then to faces. Yeah, I got to do it all because I was the only programmer.
Lavanya Shukla [00:41:55] And really teaching yourself how to build all of these like all about databases and sending you code and all that.
Paul Copplestone [00:42:02] Yeah, that was own self-taught. And funnily enough, the first, the first thing that I used was this one called Cold Fusion. You would never get a user anymore. Is it Adobe Adobe Software? Yeah. And it was quite honestly, it was very cool because it allowed you to do real time systems, which funnily enough, is what we build now. It's one of the key features of super basis that you can listen to your database in real time. And so it's kind of gone full circle and that that's how I started my career, building these like real time database systems for farmers. Now I build them for developers and there's this really funny usage patterns. So what farmers would do is they would send them their pigs. This was a pig abattoir, though. Send them to this this processing plant once a month and then they would just log in to this website for one day, watch as their sort of pigs like went across the line. And then and then they would log in for another one month. So as a completely overengineered, they don't even need it in real time. But there's one one hour of of of usage. But it was very good for my my learning.
Lavanya Shukla [00:43:15] I think it's I think almost like you were back to build this company and like everything in your life like led you back to building but there's a, you.
Speaker 4 [00:43:23] Know a pig.
Caryn Marooney [00:43:24] Farm.
Lavanya Shukla [00:43:25] Who.
Speaker 4 [00:43:26] Yeah.
Paul Copplestone [00:43:26] From, from a farm to a to a start up. Funnily enough after that, you know, because I was using databases a lot, I was thinking of, you know, the way that we're building with databases isn't very effective. And so I kind of pitched I didn't know this concept of startups, they don't really exist in New Zealand. So I found this millionaire and I kind of sent him an email and pitched him this idea of like an interface on top of a database that would make it very easy to build. And he outright rejected me. But of course, that's once again, that was 15 years ago and here I am today essentially building the same thing.
Caryn Marooney [00:44:08] That's amazing. And I think being starting anything, you have to get used to a lot of rejection, a lot of things that you care about that nobody else cares about. A lot of we were joking with Chris Van Pelt that when you're talking to people, you actually just see the top of their heads because they're still typing. You know, it's one of those things where you never get to It's hard to get feedback from folks. Mm hmm. One thing that you and I have talked about is this idea of are you in a tunnel or a cave? So when things are tough, it's sometimes hard to know if it's a tunnel or a cave where a tunnel is temporary and a cave is you've hit a dead end. And you should maybe rethink about this. Like, how do you think about tunnel in caves when you're building and you're building a company? And any advice for folks?
Speaker 4 [00:45:02] Hmm. No.
Paul Copplestone [00:45:04] Well, first of all, you should probably always assume you're in a tunnel. In a cave. You know, always be skeptical about what you're building. We've got actually one of our core, like, onboarding traits. And what we try to hire for is intellectual honesty. And largely surfaces itself a lot in distributed companies because your alone programing, you can get really stuck in your idea. And a week goes by and you've got this sunk cost fallacy, right? So you just keep doing it. So we on in the onboarding call, we tell engineers that, you know, intellectual honesty is super important for for mitigating this. And then a lot of the processes that we do are designed to mitigate it. So we do our offices and our company requests for comments. And a lot of what that process is geared up to do is, first of all, ask what the problem is, and then it asks them to come up with not just a solution, but multiple solutions to the problem. And it largely stemmed from this idea that Charlie Munger has this concept where he's got a very good quote. I don't know exactly, but he says, like the mind is like, you know, a human egg. As soon as it's like, got an idea, it's like it's like impregnated with this idea and you cannot let it go. And what you want to do is understand that that's going to happen and that you need to design ways that you can actually release this idea. So coming up with at least three ways to solve a problem is a very good, good start. If you just come up with one, you truly believe that it is the only way.
Lavanya Shukla [00:46:50] It's a silly thing to mean everything that you talk, but you're very thoughtful in how you build the company and like how you structured. You have processes for everything. How do you hire people that will also like, do this when you're not in the room? What characteristics are you looking for? And then how do you indoctrinate is the wrong word. Teach them. Inculcate your culture in that.
Speaker 4 [00:47:13] Yeah, I think that I.
Paul Copplestone [00:47:16] Don't know if you can we do say that super basis process driven once again in the onboarding call and we call it out as an onboarding trade. And we tell people that, you know, this is one of the traits that we look for and what we want you to embody. I think a lot of it just comes down to, you know, what the say founders are doing a lot with writing down processes and principles. And we're saying it very often that a number of times. And actually, funnily enough, it goes back to your point, Karen, like on a tunnel in a occasion, we've got this concept of Kaizen, which means like very incremental changes comes from the Toyota part production system. We've got a Slack channel just on Kaizen, where anyone can suggest a change, not just on the product but across the whole business. And we talk about Kaizen when when doing shipping is just one of the most frequent words that said. And you know, if you do these things often enough, everyone will of course adapt. So I think it comes from the early DNA spreads out, and then you've got the the core members of the team kind of really soak that into their own DNA and then they take it into the rest of the org.
Caryn Marooney [00:48:29] One thing that you and Lavinia have in common, if you spend a lot of time outside the United States, neither one of you were from the U.S.. Was that true love on you? Where were you born?
Lavanya Shukla [00:48:40] India.
Caryn Marooney [00:48:41] Okay. So neither one of you are.
Speaker 4 [00:48:42] Well, we're in India.
Lavanya Shukla [00:48:46] And near Delhi, an hour out.
Caryn Marooney [00:48:50] And so much of Silicon Valley thought is that the world is around Silicon Valley. And do you feel that being from outside the U.S. is a feature or a bug for both of you and and you both spend a lot of time outside of the U.S.? Not only are you from outside the U.S., you spend time outside the U.S. So how do you how do you think about that? A lot of times people in Silicon Valley are like, that's a bug. It's like, is that a is that a feature or a bug?
Lavanya Shukla [00:49:20] Paul.
Paul Copplestone [00:49:21] You can still live on air. I'm very interested in what your answer to.
Lavanya Shukla [00:49:24] This is so interesting too. I started coding when it's of my mental model of the world was a Silicon Valley software like it. And that was the thing. And then I moved to the US when I was 15 and like I was there for 15 years and that's the only mental model that I knew most of my friends, the programmers, it was like non-diverse at all. And then during the pandemic, I moved to India for a little bit and then Europe. And it was interesting because they when you're in the US, you think that's the only country that exists, That's the best country. That's how we do things. And locally and even within the US, it's like Silicon Valley. This is how we do things. But then suddenly I was like in London and like I was looking at like, Oh, they're innovating. Like the stuff that they're doing in the financial industry thought much more advanced than the US. And then we're like, Wait, it that is not maybe it's not the only way, you know. So I think I started picking up different mental models from different places that I lived in. And now I don't just have one mental model and one I know my mental one is not the only one, so I should be adapting. And then even occur to me in second make. I have all these different option for mental models, which I think have made me a better thinker. And also it has made my life richer personally, because I think I would go to all these different places in Paris and London, which I would have done. Like I was just in like my own little apartment throwing, like tacky party in Venice. It.
Paul Copplestone [00:50:50] Very cool.
Speaker 4 [00:50:53] Paul. At that.
Paul Copplestone [00:50:56] Yeah, it's a good question, sir. If I wasn't running a tech company specifically my type of tech company, which is the developer tooling. Then I think it couldn't have been as successful as it is without a the period of Covid. And B. Has to be Deftones. So I think, like, if there was anything else, like, for example, if I was starting an AI specific company now, it would have to be in Silicon Valley. That's just really one of the most important things. For now, I think like the we've got enough traction that outside of outside of the US and looking in WI, for example, when I visit the US, we already have a lot of customers in Silicon Valley and strategically I can go and saw a really good example was there was the AI engineer conference the other day we came in, I met a lot of our partners, you know, the folks from Langtang and Lama Index and Swigs from who was organizing it. And this is very effective use of my time. But actually I parked up at Six's in his house for a couple of weeks and it was super exciting because I'm going every night to another AI conference and a meet up and one night I went to three different meet ups and I look back on the things that were most effective. And it was really the AI Engineer conference. That one summit was extremely effective. And then the hackathons, you know, had diminishing returns. So I think it's a bit of a feature for our current stage of the business that we're extremely focused. We don't have to be too distracted and then we can go in very strategically when it's when it's important.
Caryn Marooney [00:52:48] Can you both give me your prediction for what's going to happen in your industry in three years? Everything's worked. And I know the video. I'm putting you on the spot. And Paul, I'm putting you on the spot. So think about in the world of databases or vector databases or however, in Lavonia, when you think about everything that's being built in a AI three year prediction with everything going well
Lavanya Shukla [00:53:19] What the most excited about is the new kinds of companies that are going to emerge, because that's what these multimodal multimodal models make possible. So, for example, everything you're going to have an actual in a piece so you can like literally just say, Hey, I want to build a new home in my house. I'm here. Show me some ideas. These are the ones they like. Now make it happen. Great. So if that's the problem, certainly it will be Pinterest. Who boasts that? Or like a home renovation company who boasts that or like. Like, you know, whoever Costco or like whoever says like the materials or maybe it's a whole different type of company who just comes in and owns that whole journey to make really excited when like our customer, how people are asking or what they want becomes different. And they the nature of the ask becomes so expansive. What are the kind of companies that come up to serve that?
Paul Copplestone [00:54:14] Nice from outside. Yeah, I think specifically on the air smell type of future. Then the thing that if you are willing to bet that it's going to continue, then you can almost guarantee that everything needs to be turned into an embedding. Whether it's a photo or video or some text, it needs to exist both in its raw content and as an embedding of some sort. And this is a huge, huge amount of data that then needs to be converted. So I think I think about things from a database point of view. If you want it to be useful in real time, you need it to be accessible to an operational database, say like like PostgreSQL. And you also need it to be extremely like large in terms of all storage. So I think what you'll see is that there'll be a lot of the operational databases will start thinking more about how they can store huge amounts of data, either by shading, partitioning, storing it in S3 And that's kind of the future where we're also driving towards. So it's pretty exciting. I think over the next three years, don't be this spill out effects of this move whereby big data, not just embeddings, but anything to do with an analytical data OlAp use cases will start being baked into more OLTP use cases as well.
Lavanya Shukla [00:55:51] Excited for every NFT ideal, like everything is going to be stored in embeddings and like, how did it come out? We think that would be so cool.
Paul Copplestone [00:56:01] Just I don't know if it's extremely or relevant to this audience, but as well, like we maybe it is so or so I might as well say it. You know, a lot of was, say, Apple bringing out or even, you know, Facebook. I think they just recently bought out some some cameras and a lot of that my computer wants to do and stop pushing into these sort of news cases, then not only is everything going to be turned into an embedding, but a lot of the real world, the spatial, everything spatial, is also going to be stored objects where they live in physical space. So this is a super interesting paradigm as well, one that that I think is an important future. I don't know how far away it is, but there's this convergence between raw data, spatial data and embeddings that I think is going to produce a lot of like really, really interesting businesses.
Caryn Marooney [00:56:58] I do think a lot of the absolutely work, a lot of the air work that Apple and and Metta are doing are going to help map a lot of this. And then it'll be interesting to see what the side effect is on that, like what that means to how we can tag and understand spatial information in a way that we can't today. And then a fun fact. Paul is on his phone, so this is all through the camera on his new iPhone just to show how it works. Ah, because I think it's really great quality.
Paul Copplestone [00:57:34] Thank you.
Paul Copplestone [00:57:42] I actually was talking to a streamer the other day at the Y. He made up and I asked them, you know what, What Tamara, do I need to buy if I want to get into streaming? And he said, actually, if you are not a professional, you shouldn't buy an iPhone 15. That's actually probably the best that you can do. So it's.
Caryn Marooney [00:58:04] Paul, thank you so much for coming and spending the time with us for sharing about how you think about building a company. And I'm really excited for the future and of what's going to be possible. And thank you so much.
Paul Copplestone [00:58:23] Thank you for having me.
Lavanya Shukla [00:58:26] Thank you for listening to that episode of Creative Dissent. If you like what you saw, don't forget to subscribe to us and we'll see you in the next episode.
Add a comment
Iterate on AI agents and models faster. Try Weights & Biases today.