Unfinished Business: Re-inventing 'modern' data tools

0:01All right, let's jump in. So, Colin, Jordan, let's start out by introducing yourselves. Can you share just a little bit about your background and how it led to what you're doing now?

0:10Uh, sure, I can go first. Uh, so my name is Colin. I'm the CEO at Omni. Uh, we've been at it about two years trying to rebuild the BI stack uh kind of from ground zero. So, the the way I got here was I was actually one of the first Looker customers at a company called Hotel Tonight. Uh, loved the product,

0:27asked if I could join the company. uh was there for about eight years doing a variety of things um including using the product a lot but also ran product briefly um and uh kind of postacquisition decided that there was still some interesting things for us to do in BI land and uh that's that's the team that's bringing you Omni now. So uh

0:46we're trying to take all the best things from a lot of past BI tools and bring them all into one place.

0:54And I'm Jordan Tagani um uh at at Motherduck. We got started just a little bit after the Omni folks. I remember uh seeing the Omni announcement and being like, "Hey, that seems like a kind of a neat neat idea. Um uh and you know, and neat idea to start a company as well. Um I you know started out as a as an

1:13operating systems guy and then a compiler guy and then a storage guy and then morphed into a database person. Um after a after all I helped um you know I joined Google because I was really interested in this sort of big data phenomenon and processing huge amounts of data. Um and they seem to be doing it better than everybody else. Um I helped

1:34create uh you know helped start Google BigQuery and um worked on the a lot of the core engineering for that. uh then led engineering for a while and product for a while and then after after a decade at at BigQuery um you know that just seemed like a long time to work on the same the same thing and as I as I

1:53told the um you know the VP there when I was leaving like you know it's just like I think it's going to be really successful but it's just not going to be fun anymore. Um you know kind of taking you know taking it in sort of squeezing customers for more money versus like trying to innovate. um didn't seem to be

2:11like the kind of thing that I I wanted to do. And so I jumped over to Single Store, which is, you know, a late at the time it was called an MESSQL latestage uh startup database company. Um they really needed to sort of like create a SAS uh version of it. And so I helped them helped them with that. I learned a

2:29ton about like, you know, how the real world works outside of outside of Google. And um and then I kind of you

2:37know realized that I like I wasn't excited about like taking something that somebody else had created and just sort of growing it like in order to be really passionate about it. I you know I wanted to work on you know something that was my own and see you know see how far I could uh I could take it. So um left

2:54left Single Store to uh to start to start Motherduck um and you know you know things have been going reasonably well ever since.

3:04Thank you both. Um, let's start by going back to the early 2010s. BigQuery had been released for GA. Looker had recently been founded. Can you talk a little bit about the market then and what made that opportunity so great for Looker and for BigQuery?

3:19Uh, I can I can jump in. So, um, I think there's a couple things. one is it was like the Hadoop era and you know so uh you know there was the all this uh excitement around around you know big data big data is coming um and uh but I

3:36think there was also kind of people were starting to recognize that there were sort of limitations to some of these some of these technologies and some of the like early big data technologies forces you to forced you to sort of change how you think in order to and how you operate in order to actually operate at at at that scale and there was sort

3:54of a second generation of tools that were coming sort of like you know you know BigQuery was one, Snowflake was one um they kind of let you operate in the old way um but over kind of the you know larger larger data sizes. So um you know I think there was sort of that was sort of one major major thing. The other one

4:15was just sort of the you know the cloud um cloud migration you know started really in earnest and people started using cloud for for actually real things and and you know one we had to convince people to actually store their data in the cloud once your data is in the cloud then you need to process it in the cloud

4:32and then that kind of like pulls you know pulls on the the the thread of that sweater uh and then you you know a lot more stuff starts to starts to to happen um happen in the in the cloud.

4:46Yeah. And I mean I think on the BI side, it was actually just following the exact same two trends. And I think the the founders of Looker I think saw this more clearly than I did. I was on the customer side just using a Postgress or it was a MySQL replica. Uh and it just wasn't terribly performant for the types

5:02of analytics that we were trying to do and just the idea of like cloud columner databases was a very simple solve. So like our first database after that was Redshift and just the performance profile was completely different and once those types of cloud databases stood up that were somewhat elastic um like Redshift wasn't even truly elastic but you could turn you could tune

5:24performance at least um it the idea of

5:29doing BI extracting everything into the tool versus putting in cloud data warehouse just didn't really make sense anymore and in some ways Looker was sort of like right place, right time for that trend. Also, just sort of like the engineeringification of the data person that DBT has sort of continued was another thing that Looker wrote, but it

5:49was honestly just like a reinvention of how in database BI or I guess not a reinvention, but an invention of what in database BI actually looks like. Um, and Looker was really just building a product that no one else had really tried to build in in that kind of concept of in database BI. And so we were riding that trend and then

6:09obviously all the things around data modeling and centralization and things like that. Um we were just sort of bringing a a newish a modern perspective on a lot of very old concepts there. Um and and that was sort of the trend that we wrote. I mean I think the other interesting thing about Looker was like Lloyd will probably admit this is is

6:26like he really wanted to build an exploration tool like a a tool for people to explore data. Um, and on the customer side, I don't even think dashboards existed for a couple years at Looker. Like it was very much a pivot table experience. And I I kind of view one of my roles coming into Looker as sort of convincing them that we had to

6:43do the BI stuff. Uh, like even if we didn't want to, we had to build a capable dashboard because that was going to be a core of the product no matter what. Um, and so it was like sort of this reconciliation between all the fun interesting ideas that that we wanted to build and sort of like the market that

6:59actually pays the bills. Um, and you know, I think Looker executed really well also from just like a go to market point of view and all customer experience all that kind of stuff as well.

7:09You mentioned Redshift and DBT more broadly beyond BigQuery and Looker. How was the landscape changing at that time for data people for other data tools?

7:18So I think one of the interesting things is like the the cloud um it was this open playing field because there had been all these established you know established players and all these established kind of niches and all of a sudden you move to the cloud and those are like th those those players are not there anymore. And um a lot of the like

7:38a lot of the constraints actually that uh had sort of driven the way some of the technologies and tools were put together uh those constraints had been had been removed in the in the cloud or were different. when you have separation of storage and compute for example when you kind of have like this you know scalability everything you know

7:55everything is you know kind oforked everything is in is in the cloud um things just sort of look different and the opportunities look different and I think it gave the opportunity for different startups and different tools to to sort of grow you know ahead of time we were just talking about Fiverr and the great partnership with Fiverr and Omni and

8:15like you know connecting um you know BigQuery and Looker. Um, and just you know things things even things about like you know most data in data

8:28warehouses is like strongly rectangular like people you know have been building these sort of kimble models and these sort of star schemas and snowflake schemas forever and they've been sort of shoehorning things into that because those are what the tools could do. Um but you know real data is not always rectangular and sort of Looker was really good at being able to actually

8:47handle rectangular data. BigQuery because you know coming from inside Google and everything was protobuffers could handle sort of non- rectangular data and so it just sort of gave an opportunity to sort of build the tools that were needed versus the tools that were just sort of replicas or clones of the kinds of things that had you know

9:06had been built in the previous decades. Yeah. And I I think the other thing I would add that almost seems like trit to say now is um like SAS was becoming a

9:17thing. Um I I know it feels like everything is SAS now, but like 15 years ago you couldn't just stitch together a toolkit that easily. Like either you were buying like long-term perpetual licenses or desktop software or or the tools didn't fit as well together. And this was sort of the age where I could actually go to this to the store on the

9:38internet and buy like a data movement tool and a database and a BI tool and they just fit together and worked really nicely. Not to say that you couldn't do that before, but sort of like the consumerization of business software was another trend that essentially just enabled a data person now to stand up a stack on their own very trivially and

9:57actually put reporting into people's hands. Um, and it wasn't like served off a desktop or anything like that. And I think that consumerized a lot of the modern data stack um in a way that actually like really enabled organizations to move really really quickly. So like startups could just stand up a data stack that was that was

10:15pretty mature very very quickly

10:20and yet kind of being at the the nexus of a lot of all that was going on. You know within 10 years you both were looking to to do something new. So let's talk about that a little bit. What made you want to do something new?

10:31Um, I mean, yeah, I I'll take this first. Um, I mean, I I think I I've written about this a little bit, but like success creates an innovator's dilemma. And so, as you're building product and actually delivering something, you start having uh like a thing that your product does. And while like for Looker, it was like we set out

10:52to do modeled BI. Um, and we executed really really well on it. like we I I would say that we were sort of this backlash to the Tableau sort of like isolated desktop experience and it meant that we could deliver centralized BI in a way that was really governed and even though we talked about sort of bridging

11:09this world between like data breadlines so waiting on the data team and data chaos like truly we were a data bread lines company um like you needed the data team to do the things and uh from a personal point of view I think I actually gravitate a little bit more towards like the data chaos side of the

11:26world like I like to move really really quickly and sort of make a mess as everyone that's ever interacted with me knows. Um to the point where at Hotel Tonight I made every single user in the product an admin because I was like it's better for you to go fast and make mistakes than it is for you to wait on

11:40me and like we can go figure out the problems later. And this sort of core thesis just sort of kept persisting through the experience at Looker, which is like I love the model and centralized BI. Um, but I also love moving really quickly and like not modeling when I don't want to. Um, so like the the the

12:00kind of amusing example I gave is as we were getting acquired, I did a lot of the due diligence. I did a lot of it just writing SQL because like I couldn't be in the model. I lost permissions to model even I think as I was managing the data team at Looker and it just was like the the product wasn't completely able

12:15to compromise to what I personally wanted and I think like reflecting on Looker strategy we couldn't just go and undo the promise of centralized like governed BI that was what the product was and that was kind of what customers came to us for and it takes kind of a more wholesale strategic shift to do these sorts of things properly which is

12:36why you see kind more mature companies end up as sort of like a latch on like Tableau is desktop software. They have servers. A server is essentially just a network of desktop software. It it wasn't like a complete reinvention of the strategy. And so I always wanted this product that could actually sort of compromise in the right ways but still

12:53have a lot of the core of Looker. And uh that's essentially like that was the thesis from Omni the day after we left.

13:00And that is exactly the product that we built is something that you know appreciates the things about centralized BI but lets it decentralize and it just felt like big market there was this hole in it can we go do it um was the thesis like the last point I would say is I started a company previously I really

13:19really did not enjoy it um it felt really hard and I felt like very non-convicted every day so I told people for 10 years I would never go do this and sort of like the only reason that Omni gets to exist is because I I like actually deeply believe that someone needs a product that can be centralized

13:35and decentralized at the same time. So, it's just sort of like a manifest destiny of like gaps in the market and having been sitting there for a long time and then having people around me that, you know, we thought we could go do it together.

13:50Um, so yeah, I think the things that like kind of drove me to to want to start to to break out and start something new, um, there were a couple of sort of big events and they weren't actually um like they didn't seem big, but they were kind of they just sort of made me think think that we were kind of

14:08chasing the wrong thing. One was uh there was this Snowflake versus data bricks like spat on um you know over overbenchmarking and like everybody was you talking about oh are they are they cheating or how are they how are they doing it and I think the there was sort of like something that nobody was really paying attention to which was that the

14:27size of data that they were using for the benchmark was 30 30 terabytes. So 30 terabyte TPCH and basically means that the tables are you know tens of tens of terabytes and um and using using this is sort of like hey this is what the actual representative performance is going to be for for actual users and you know

14:46when I thought about the experience I had working in you know single store working at BigQuery that we re there were really nobody that was using data that that large like you know and we had at Google we had some of the largest customers customers in the world, you know, some of the largest data warehouses in the world. We had we had

15:05massive amounts of data, but the stuff that people were actually querying tended to be tended to be much much smaller than that. And and so I'm like, wait, so like basically these these basically two juggernauts of the industry are out there trying to trying to get bigger and bigger and bigger data versus like, you know, what users

15:25actually cared about. like I you know kind of thought back to like I did a bunch of analysis on you know data sizes at uh at you know when I was working in BigQuery and something like you know 90% of queries were subundred megabytes like tiny tiny tiny and I've I've heard you know people at other database companies

15:42like you know saying the same basically saying the same things is like almost all of our data warehouses at Snowflake or sub sub terabyte and and like all all of these other you know um people are using smaller smaller data and uh and so

15:59really like it seemed like okay well if everybody's chasing that other thing there's got to be opportunities at you know to build things that for the rest of you know basically for for the real customers the people that actually that actually exist versus that we think are going to exist in some in some future when you get to a

16:17billion customers and um and so you know

16:22I was like well you know the important things are like you know, usability, you know, and I think that's one of the things that actually made Bitquery really successful and Snowflake really successful um was sort of they were they were quite quite usable. On the on the other side, one of the things that I was seeing was, you

16:38know, we'd had a couple I was a single store, we had a couple of customers that were saying like, hey, I would love to run this in every car.

16:46So like I want to run this this technology actually kind of running in sort of the dashboard and basically do some an analytics locally before we kind of push things up to the uh into into the cloud. And um it was it really

17:00wasn't designed to shrink down like that. And you know there's all these sort of microservices and pieces and like you know aggregators and leaves and like um it just couldn't shrink down.

17:10And so then I saw like this um this you know kind of upstart database uh called duct DB and they were um like they could

17:21run in 100 megabytes they could run in your browser and they were also really fast and really robust and it's like huh this is sort of interesting. So I wonder if I can sort of combine like this like hey well you don't really need massive amounts of data with like hey we can shrink this shrink this down and like

17:35somebody should build a serverless version of this and then I'm like hey maybe maybe that somebody should be should be me. Um so I you know um

17:45started uh started trying to trying to hack on that for for a while and um you know turned turned into mother duck. as you describe these learnings and observations. I have to ask, do you think your previous companies could make those adjustments and and address this themselves?

18:02Go for it. Um, you know, I think it's hard. I mean, there is a bit of an innovator innovator's dilemma. I remember when we first started mother duck there was you know apparently that set off some echoes at at Google about like you know little query you know building something you know something smaller but it's just you

18:18know if your whole like world is like is

18:22larger data and bigger bigger data I remember like I think Google has nine products that have over a billion customers it's a little bit harder for them to see that like hey there's all this other this other world world out there um And you know, I think it sort of takes somebody else to sort of poke them and say, "Hey, there's another

18:42there's another way of uh of of doing things." And they just, you know, you know, also is as time goes by like it gets harder and harder to move quickly and get stuff done. And and I think that um that also is, you know, can be can get in the way. And also just you know you know Google cloud is a huge business

19:02now and they don't seem to be making big bets on like new technologies and um so

19:10you know I I I don't see that coming from Google. I think perhaps you know other there's some other you know mega corp that could that could come up with something similar and we just have to hope that we're doing it faster and better.

19:23Yeah. Yeah. I the incentives at a startup are just so different than they are at a bigger company also. like you get to live with your customers early.

19:32Um like I think I almost forgot about sort of like how close we were to customers in the early days of Looker versus like the late days of Looker. Um like you just need to you need to be doing things that spread across every single customer. Like you can't do hero stuff for an individual. And I feel like

19:48starting from zero, you have to be doing more of that kind of stuff. both because like your product can't do everything that your user needs, but but also just because like you're young and you need to make them successful. So like you almost need a partnership with your customer that's like I need you to come along with me on this change that we're

20:05making together. Um which is I I think like probably the most fun thing about starting a company is like we have people that you know might want things that we don't do yet but they see enough of sort of the future in the product that we're building to come along with us. And I think it's very hard for big

20:21companies to have that trust with the customer because like with Looker, you know, we like we might be deployed in an app that has a million users on it or something like that. Like that customer is not buying Looker for the risk seeeking appetite of like development.

20:37They're they're buying to be solid. And so like we are we get to we get to take risk which is kind of cool because now all of our customers are like, "Hey, I want to be in the beta. like I I want the messy stuff in front of me and so it gives us actually space to go do those

20:51sorts of things. I think the other one is just like it's it's hard to allocate resources to bets as a big company. Um especially if the bet is like an attack on the core product philosophy and so like at Looker we talked over and over again about like oh can people just touch the SQL? Like we're writing SQL

21:09can people just touch the SQL? It's it's like it's so obvious. Um, but like once once you pull on the thread again, you're like, "Okay, wait. Every single thing that we does is dependent on this." Like I remember just trying to release SQL visualization in Looker.

21:23Like apparently very trivial. Well, it turns out all of Looker's visualization is based on expecting dimensions and measures. Like every single thing in the app. So we had to actually pretend that everything was a dimension or measure and build like a separate thing to try to turn raw SQL. This is just to like be able to take a two column query of time

21:45and count and put it on a chart. And it's just it's really hard if you're not thinking about these things early to actually stuff that underneath it. Um, which is like why it's also nice to just be able to blast out the tech stack, restart it, and be like, "Okay, what we're going to do is actually let people

21:59write SQL and model like what do we need to do underneath that to make it work?" Um so I I think it's very very very hard to do a pivot like that that is that is structural like I think at the margin um big companies find great strategies and like sort of pivot like Atlassian went to cloud from desktop on Jira it like

22:19they're probably still doing it 10 years later um so like it I mean Tableau exact

22:24same deal um so like I just I think they can evolve but they can't reinvent is probably the simplest way to sort of explain it.

22:35Makes sense. Let's uh let's transition to talking about the present with Mother Duck and Omni. What are the big problems that your customers are asking you to solve today?

22:45You want to go? Sure. Um I mean it's it's the same problems that we got at Looker actually. Like I think the fun thing about being in the space is that the problem sets are very similar. Like we want people to make better decisions. We have data. we need to get into people's hands and we need to make every single type of user

23:03effective. Um so in that sense it's really about taking the company and just making it more efficient with data. Um like that is the actual underlying problem set. Thankfully

23:17um or that they need a database to store data like the nice thing about being in this big crowded space is we don't really need to reinvent the problem set as much. Um we just need to show people that we can solve it better. And so I kind of view a large part of the problem is like we need to build great product

23:32and we need to kind of deliver on this message between centralized and decentralized like make the data team really efficient at creating self-service make individuals really efficient on their own um to create their own self-service. A lot of what we need to do is also just on like the customer service side of the house. Like we need to partner with people and help

23:50them go figure out how to solve data problems. Um, so like we're doing a little bit of both of those things, but I would say from like a problem set standpoint, it looks very similar actually to Looker. Um, which I think has been very beneficial in sort of thinking about what our customer service orientation is or like how we partner

24:07with someone through a sales cycle. Um, it's just like we're using a slightly different set of tools on the top that let us do things hopefully better in many ways, but maybe worse in a couple ways also. um that like we we can go deliver that solution for them.

24:24I think for us the you know there's basically kind of two big problem areas.

24:28One is just sort of making it easier to understand your data. I think you know a lot of database companies tend to sort of take as their purview you know okay I get a SQL statement and then I return some results. Um and then but I think there's a there's a bit of broader question is like is like hey I have a a

24:46business problem that I want an answer to and I want to you know so the the problems that you you know formulation to answer is actually different than SQL query to to SQL results and I think that's you know one of the reasons we were focusing really on like on simplicity ease of use is like hey I've

25:05got this like gross CSV file that you know was generated from from somebody else and has a bunch of problems with it like um and I need to like uh understand that and join it against data in my in my warehouse like um you know we we want to make that kind of thing easy because like that ends up being the kind of

25:23thing that you end up spending your time on. It's like oh I got to ingest this like you know from this random you know random location and like you get in and then and then asking the question is is is easier. So that's sort of one you know one area and the second area is like there's just a lot of people

25:38building applications that um need to show act need to show data and you know kind of the existing data warehouse vendors are not really set up to kind of handle I've got 10,000 users and they all basically have non-over overlapping non-over overlapping data and I want low latency access to to that to that data.

25:58Um, and that's something that we can do really well because, you know, as a scale up system, you know, we can and they can scale down to zero. We can give each one of those end users uh a duct DB instance in the cloud. It turns off when you're not using it. It, you know, basically is three for the the tiny

26:15users and then you might have a couple of your users who are huge and we can scale those up to basically, you know, the size of the largest EC2 instance, which is roughly the equivalent of a Snowflake 3XL. So like you know that's you know a million dollars a year worth of worth of worth of snowflake that we

26:31can devote for each one of your users uh if you need it. And so that's um a big area that you know we're investing in is sort of and area where where people are asking us to help solve their problems in in these sort of building data applications.

26:47Yeah. one one of the things that you wrote recently that I loved um and like I you articulated something I think we've actually seen sort of working with duct DB because we we have it a little bit in our product as well is sort of like the end toendness of the customer problem. So like the example I give is

27:02we just released a feature uh like for for the Excel folks it's there's a function called text which essentially just lets you take a piece of data and then format it but it's formatted as a string. So you use it when you're writing like giant if statements that you're mixing between like strings and values and you need to format things as

27:19a string. And in practical terms, like in database land, there's just no reason to have customers do this. Like there's no way it's in the benchmark set because like why would you spit out a column as a mix of percentages and words? It like makes no sense. Um, but I think what's been kind of amazing is the way that

27:38DuckDB has sort of poked at the edge of sort of like what SQL is and what it's used for and how much closer it gets to like actual customer problems that people have, which is like I have this column and I want it to be a percentage and then say like was not a customer yet in this row. And it's like it's not a

27:57pure data problem. It's like a it's like an Excel style problem, but you're trying to create structure around it.

28:03And I I feel like the database wars are

28:07all about just like how fast you can return 50 terabytes of data and like group by it and slice it. And they're never about like I need to run some like weird median percentile function that can mix between like data types or something like that. Um or like can coersse between types in in like a way that's convenient for the user. Um, and

28:29I feel like those have been like a lot of the fun developments because in the BI layer, those are all of our problems, but like as the database sort of picks up some of them, we get to actually take advantage of them, which is which is really really cool.

28:43Yeah, exactly what we're trying to trying to solve. And I I um it's cool that you're you're seeing it as well.

28:51Um, let's talk a little bit about text to SQL. This has been a very popular topic of late. people are interested in replacing the analyst with with an LLM or there's all sorts of things in between. Can you talk about the approach that Omni and other doc have each taken to integrating the big innovations that are happening in the space?

29:12Go ahead. You can have it first. All right. Well, I think you know I think one of the things that you know like GPT4 came out and everybody's like oh well like you know it's going to put us all out of jobs and you know you look at you know GitHub copilot and like and you know engineers can be 40% you know

29:29faster or whatever. Um and but I think one of the things that you know has become really obvious is that um text to

29:39SQL on its own is is sort of a non it's a dead end. It's like um you know Ben Stansel wrote about it like the folks at number stationation like talk talk about it who you know partners partners of ours um like you really because there's so much business logic that's in the heads of it's in the heads of the

29:58analysts or it's in your semantic model somewhere um you know what does revenue mean? I've got two tables that actually look kind of similar. one might be the older version, one might be the new version or one has data from like some set of products and other like and if I want to ask questions about that um like

30:17it's it's you know it's not actually a a solvable problem to to be able to figure out the interpretation of that especially because you know one of the reasons you want to do analytics is because you want to make decisions about stuff and so if I'm going to make decisions like hey we need to build a factory in in Atlanta um you know

30:35because of like this, you know, thing you see in the market. If that's actually not happening and that's just, you know, an um an artifact of the data, like you're going to feel pretty pretty silly um at best. Um so, like I do think

30:51that um you know, text to SQL needs something else. And so, um we have not been necessarily really trying to solve the text to SQL problem. We've trying to solve the like how do we make an analyst you know have superpowers um and so you know things like GitHub copilot can give developers some superpowers um uh and you know as you're writing

31:14your SQL so we have like this you know fix it which basically says hey if you write a if you have an error in your in your SQL we can we can fix it for you um and the and you know jump right to the to the line that had the problem and and and correct it. Now, one of the cool

31:29things about that is it actually can change how you're doing your, you know, how you're writing your SQL because there's a lot of times where I'm like, you know, you're right, you're starting, you're starting a SQL like, okay, time stamp diff. Okay, I have to do like interval three day, is it days or day or are they in quotes or not in quotes? And

31:47what order order is the arguments in? Um, but you can actually just sort of like type it the way you think it should be instead of jumping out to the docs and then let it fix it for you. and it's like, oh, that's because it just keeps you in the flow a lot better. And I think there's a bunch of other kinds of

32:02things that we can do along those along those lines. Um, you know, broader speaking, like I do think that there is a need for kind of a semantic layer for for being able to answer these, you know, ask and answer these types of questions. And I I see like, you know, the some of the things you can do with

32:19AI are along the lines of like, hey, we're trying to make it easier to once you can pose a question, you know, you get a you get an answer. Um, and to sort of uplevel the um the the problem a bit.

32:32Um, but also like I think there's just a bunch of other pieces that are that are needed before that's, you know, before that's actually viable.

32:39Yeah. And I think I think Omni is is uh, you know, can can help with with some of that.

32:44Yeah. Yeah. I mean I we're in pretty violent agreement like I the the the first most important use case is definitely like the copilot style stuff.

32:52So like um we wrote one that writes Excel calculations and like I think as we weing dialect to match Excel one to one and sort of didn't fully realize that that was actually a perfect setup for an LLM.

33:06Like it turns out the internet's pretty good at writing Excel calcs. So the LLM is actually pretty good at writing Excel calcs. Um, and like I think the the the first thing that we learned after that though is that the error use case was just as important as sort of like the correct use cases for it because it'll

33:24it would write functions but it was wrong sometimes and like you needed to be able to touch it to follow it up. And so like I I think a lot of the full experience loop is really really important when you're talking the LLM in like what is how does it fail? And I think the hard part about like blackbox

33:41SQL is that on the one hand it's just like unbelievable magic that can write anything that's coherent at all. Like it can give you numbers, but like the output of a question just being a number is just not good enough because like you're never just asking one question like you're immediately going to slice it and then like you start needing these

34:00UI controls and sort of like follow-up experience. So like the completeness of the experience really really matters to how these things get used. So I my view is like they'll be tucked in the products all over the place but in much smaller use cases that actually tie back to the full user experience of the app.

34:16Um and like the retraining point is a great one. Like I I think when these things are used really well, you almost have to retrain how you do things because they they essentially like create forgiveness around doing stuff wrong. Um, and like even if all they did was fix sort of like error handling or like search inside a little component of

34:38the app, they can be just enormously valuable. Like our most common support questions are just like, how do I do a thing that the app can already do? And they can start solving a lot of those problems for sure in like very tactical ways.

34:53Last question. U data is and always has been a pretty crowded space. What makes you think that your companies can go the distance?

35:04Um, I I think we have I think we have a unique um take on on the data world that I think is aligned with the direction that the world is moving, you know, with um scale up actually works. You don't have to build these scale out systems.

35:19Like uh user experience is is super important. Um and you know duct DB is

35:27sort of taking taking the world by storm and we have a pretty special relationship with the with with DuckDB and um and we've also sort of you know we're veterans of of kind of seeing you know we have people from data bricks snowflake bigquery um uh you know AWS databases like so we have kind of like a

35:46lot of combined experience with seeing how it's done how it's done elsewhere and what you know can be done can be done better. Um and you know I think as a startup we have you know can can be pretty nimble and um you know and I think uh you know hopefully we'll uh um you know hope hopefully we'll succeed

36:05and and um uh because it is a it is a crowded uh a crowded market but um I do think the world kind of uh you know it's it's a huge market and it's growing and it's not a winner take all market and so um I think that sort of leaves a bunch a bunch of opportunities.

36:25Yeah. I mean, our answers are obviously very similar, but I like I I think you win on a few different axes. Like one is that the product needs to be great. And I think that we have a sort of slightly unique take in a very crowded market and it's kind of fun to be like we're in this really big market and we're two

36:41years old and we haven't built that much, but we're already pretty different than everything that exists. So, it's like a fun starting point to actually find space in the market. And I think the advantage that we have there is just that we've been in here for a really long time and we understand the sort of gaps not not only the gaps but the good

36:59things that exist too. And so like I think you need this sort of gentle balance between uh like doubting everything that exists but also like deeply appreciating the stuff that exists. And I feel like we're kind of at this interesting point where you know we got to build the last one. Uh so we we like we know some of the good things in

37:19there, but we have sort of like a healthy uh like lack of trust in some of the ideas also that we can go test and push on. And so I think that's part of it. I mean like what I I think that one of our advantages, this is going to sound really silly, is just that like we

37:33probably care more than all the other companies that are out there right now. Uh and like customer experience matters and we want to deliver a better one than everyone else. And so like that's probably not a sustainable advantage for like the next 50 years, but I like I will fully lean into just trying harder than everyone out there to deliver a

37:53better experience for our customers, too. Um, but like and then the last one is like we got to we're going to have to keep fighting to to do it. So like we're going to have to kind of prove it progressively um because we need to keep doing those things. But I I think that we've been given this really nice

38:08advantage which is like people that have done it before, a problem that we understand, a bunch of people on the customer side that sort of like trust us and are close to us. And so it's kind of like our opportunity to just kind of keep doing this stuff and and and learn and go.

38:25Thank you both for sharing all your thoughts. Um thank you also to our live audience on LinkedIn. um this if you join late or if you want to share it with co-workers, we will be uh keeping the recording up so you'll be able to access it via this link. And yeah, hope everyone has a wonderful day. Thank you.