How to 10X Data Work with Hex Agentic AI | Ep39

What is Hux?

If you're doing anything with data, Hux can massively accelerate your workflow. 80% of the interesting, impactful data work happens in Jupyter Notebooks and SQL queries and spreadsheets and share insights via like a screenshot of a chart, put into a PDF of a deck, sent in an email. And it was just this crazy way to like work. How do you actually 10x these workflows? Hux has some really incredible first-class, truly agentic features in it where you can just ask things with natural language and it will partner with you to get to the right answer. So I can say something like, who are our top performing reps broken down by, uh, customer segment. It's starting to think about it. Starts by actually searching for relevant projects within our workspace, searching for model data, looking across all the context we have in Hex. I can see that whole thinking trace, the SQL queries it's running to understand the data. So I can follow along and I can understand the logic, but I don't have to fully understand those queries myself because it'll actually explain what it's doing and explain what it finds every step of the way. It's actually using all the context in my Hex workspace, and under the hood there is the full Hex project that I can go in and see and look at as a notebook. Key findings up the top, table that summarizes it, segment leaders, and then interactive visualization.

What's your take on how AI will reshape data teams and workflows in general in the next few years?

We've already seen their role kind of shift from being bottlenecked to actually more pervasive in terms of partners across the business. The second big thing I think we're going to see a lot of is—

Don't forget to subscribe to Data Neighbor wherever you listen to podcasts and YouTube and drop a like and comment below. It keeps us bringing you great content like this.

With that, That, let's dive into our conversation with Barry McCardel.

Hey everybody, welcome back to another episode of the Data Neighbor podcast with Shravya, Hai, and Sean. Today we have Barry McCardel, CEO of Hex, who just raised, I believe, $70 million Series C recently. Is that right?

Yeah, a few months ago. Yeah.

Congratulations.

Welcome, Barry. Yeah, congrats and welcome. Thank you. Yeah, it's great to be here. I was so glad, you know, we met at an event we were doing and I had heard of the podcast before and I met you there and it was like all the stars aligned. So I'm really excited to be here.

That's so awesome. Yeah. For, for our audience who's wondering, Barry put up a really nice dinner and I just crashed the dinner party.

Definition of crashing was not quite met.

Yes, technically invited. Correct.

Yeah.

Cool. Awesome. Well, yeah, thanks for being on Barry. Just wanted to start off with what is HEX? And why did you start it?

Yeah, well, I started Hex, uh, to solve a personal problem. Actually, my co-founders and I weren't really actively trying to start a company. You talk to some people and they're like really just looking for a thing to start. We were a little bit different that we had, I personally had this problem. My co-founders had sort of seen it from afar working with data. You know, the, the typical thing then a few years ago and still now is you have these sort of modern data stacks. You've gotten all your data into the data warehouse. And then like you put dashboards on top of it, but the unfortunate part is that like 80% of the effort goes into like building and maintaining these dashboards, but it's like 80% of the interesting impactful data work happens outside of dashboards. And so we were doing all of our work, you know, in, in Jupyter Notebooks and SQL queries and spreadsheets. And, you know, you'd share insights via like a screenshot of a chart, put it into a PDF of a deck, sent in an email. And it was just this crazy way to like work. And it ultimately made our data team feel pretty low. ROI, and I started this journey being like, surely someone has built a better version of this. Like I was, I was kind of started as a, as a buyer, not a builder. My co-founders and I had met when we were at Palantir. So I kind of felt like I had a picture in my head of what like really great data software would look like and no one was really providing it. So, um, so yeah, so what, what we actually built, we, the, the sort of first origins of Hex were really, well, how can you take these workflows that are sort of don't naturally live in dashboards and bring them into the modern era. So let's bring these to the cloud instead of running locally. Let's make them collaborative. Let's really modernize these workflows. You know, the, the Jupyter workflow, with all due respect to the Jupyter team and the community, it's not great for a lot of things. Like if you wanna write a SQL query, still to this day, you are like triple ticking that SQL query in a Python cell. And so we built native SQL cells and chart cells and input parameter cells and makes it really easy to publish it as a data app, all these nice things. And what we saw was many thousands of people starting to use it to accelerate their data work over time, move a lot of things from BI, but mostly like doing all the stuff that didn't neatly fit into that. And then starting a couple of years ago, our main focus really became AI. And especially this year, I think it's like pretty much our sole focus as a company is how do you actually 10x these workflows? Like you can kind of look at Hex as like we took a lot of these workflows from like into the cloud and now how do we take them through the next shift and be really agentic? And so now Hex has some really incredible first-class agentic, truly agentic features in it where you can just ask things with natural language and it will partner with you to get to the right answer. And that works for technical users and non-technical users who want to be able to do self-serve on top of the same base of context. You know, I could obviously say a lot more and I'm happy to share more, but that's at a high level. Like you could think of Hex as just this, the, the, uh, this agentic workspace for, for analytics. Uh, and if you're doing anything with data, you know, serious work with data, you can be, you know, beyond writing a simple query or looking at a basic chart, Hex can massively accelerate your workflow.

Got it. That's cool. And, uh, we're, we're definitely going to get into a demo, right, Barry?

Oh, I'd love to do that. Yeah, I love demos. Oh, great. The stuff, the stuff we've shipped in the last few weeks is the coolest stuff we've ever built. So I'll show it off to anyone who will watch it.

That's, that's awesome. Yeah, we'll definitely get in. That get in there. I'm curious, Barry, um, like, have you, in terms of like just motivation to keep going at this, uh, you know, like, yeah, at a startup and also like, yeah, this space in general.

Yeah.

What kind of keeps you going? Like it's a, it's a tough space to be in and it's a tough space to get adoption.

Yeah. Yes. It's a crowded space. It's also crowded because it's a huge market. You know, like it's like the, I think at some level our TAM is like everyone who touches data in any way in their daily work, which is kind of like everyone, like in white, you know, white collar professions or whatever you, however you want to categorize that. What keeps me going? Oh, well, first off, I love the problem. I love the space. I love the people. I love the people at Hex, but I also love data people. And I've thought about this a lot in the last year is like, there's been a lot of interesting questions, which we should probably talk about of like how data roles are changing. Like what makes a data person? It's kind of like a funny question these days. Cause for a while it was like, well, I know SQL. Or like, I know the Pandas syntax, you know, or like I've, I like know some stuff about how to do good data viz. And I think those things are just sort of like, you know, it's not obvious that like having a mastery of the Pandas syntax is like a thing to really, like when we interview data people at Hex, we're not really, we're not just like doing a lot of like the same like level of like coding tests we were doing for a while. So what makes data people data people? I think it's, they're problem solvers. I think data people typically love puzzles. They love, you know, board games. It's like they love like figuring things out and kind of solving little mysteries. And I think that there is a whole like way of being and a way you approach the world and the way you want to partner with other people to do that, that transcends like the technical stuff that LLMs can do really well now. And so I just love those people. Those are my people. I feel like I feel a lot of like affinity and, and fealty to these folks. And I think the number one thing that keeps me going is like when you build tools for people, And they use them and they're like, oh my God, this is great. And that's, that kind of response has sustained me through a few years now of building Hex, which, yeah, startups are hard, but we love this shit. We love building products for, to make people better, to make their lives better, make their workflows better. And as long as we're doing that and getting that feedback loop, I think we'll keep going. Nice.

That's cool. Did you, did you find product market fit immediately when you started?

Yes and no. Like within a few months, you know, we, we had worked together. My co-founders are really, really talented engineers. We were able to bring on a couple of people we had worked with before. So we had a good crew that was able to ship something pretty fast. We also had a lot of background experience in the space, pretty clear vision and a good network of people to like iterate with and like give feedback. So, so yeah, that the initial loop is pretty tight. I kind of think that we didn't really have PMF till about 2 years and 2 and a half years in maybe, which maybe sounds like a long time. That was when things really clicked as to like how the, some product philosophy stuff, some workflow stuff that really started to break through where people were like, oh no, this is like really, really changing my, my, my work. A thing that I think about a lot and I've thought about a lot in the last couple of years is I don't think PMF is static really ever because there's always this like technical frontier of what's possible tomorrow and people's expectations and you'll have competitors come and copy things you're doing or do new things and that definition of PMF will change. No more so than right now. I mean, like every day there is some crazy new thing that's being announced in terms of model capabilities or products or things that are pushing the frontier forward. And I, about a year and a half ago, but especially at the beginning of this year, started saying we don't have PMF. People are like, what are you talking about? We're at tens of millions of revenue. We've got like 1,500. What do you mean we don't have PMF? People were kind of incredulous internally that I'd say something like that. And I'm like, we may feel like we have it today. But it's illusory. That's the mindset we need because we can't, we can never feel like we've established PMF because the next day that's just going to shift further. And truly, like if we hadn't shipped, if we had stood still in our definition of what our product was a year ago to today, I don't think we'd be growing like we are. Certainly we might even be shrinking. I mean, we're growing very quickly now because we're all customers are like wanting that frontier. So there's kind of two answers to your question. Yes. But like, I really don't think of us as like having PMF as much as like a constant desire to keep pushing that frontier forward.

That's a good way to kind of put it because, you know, like especially with the technologies these days, right? Like Oracle could just pump up 40% when they were left for dead for a long time.

Yeah.

In one day, just because now like, hey, their investments are paying off and stuff. So it's everywhere.

It's everywhere, but you see this even with products. I mean, there's so many interesting stories these days of products that come out of nowhere and catch lightning in a bottle and have PMF. And then kind of, there's already been AI companies that have gone fully parabolic, like full, they've gone the full parabola of like, you know, they're now like wound down or looking for homes or like clearly just don't have an independent future. And so it's very quick in some ways to get PMF. I think what's interesting is like, how do you sustain it? What are the modes? What compounds really over time? Having an idea that you can build quickly of like, oh wow, okay, like an agent can do this. We're going to put this piece together that you can ship really quick is great. There's a bunch of startups that have gotten out there and like kind of caught that lightning in a bottle, but that also suggests that it's not super defensible. Mm-hmm. Right? Like you had a good idea once and those things can get distribution quickly now because the internet exists. What actually compounds? So I think about that a lot. And I think there's my friend Dylan is a CEO of Figma talks about like having like your taste. Your product taste is like really your moat because it's, it's not just about the one hit of like, I had this one good idea. It's like sustaining that over time and figuring out how to bring these technologies together. It's still having, still the vibe coding stuff is really cool. We use Cursor a lot. Uh, we use Cloud Code a lot at Hex, but we really have to think about the system we are building. And LLMs can maybe be helpful, like rubber ducking that, but like, dude, you, you need engineers who really understand how to build a high-scale, reliable system. That is like a really tough thing to do over a sustained period of time. And so I, I think these things can happen overnight. What I focus on as CEO is like, how do we build a company and a team that can like sustain that rate of innovation? And, and it's not about like, like a one, one-hit wonder. You don't, you know, we want to, you don't want to be that one-hit wonder. Like you have to kind of reinvent yourself. And I think great technology companies are able to kind of reinvent themselves with each successive technology and be able to adopt it and reorganize around it quickly. That's what, that's what sustains. That over time.

You must have a lot of aha moments along the way then. It sounds like it should be kind of like a series of, I don't know, like lessons that helped you get to where you are today. What are those that you can share?

Yeah, sure. I mean, there's some of the initial ones, you know, early on we were, we were really, um, thinking hard about how to solve certain problems that were very endemic to, um, the workflows we were doing. If anyone's ever used like a notebook before, uh, like a Jupyter notebook, like there's all sorts of like issues with it on like kernel state and like reruns and reproducibility and performance. And we were the first people to really figure out that you could build a DAG via static analysis. So a directed graph for execution via static analysis of the code, inspecting the Python AST, looking at the SQL queries, and basically, um, have this, you know, as a user, you don't have to like manually construct it, but this sort of inferred DAG for execution behind the scenes, like that breakthrough, we had actually done a few different attempts to like, well, how do we solve that reproducibility problem? How do we want to think about execution and all of that? And like the end solution we came up with was very elegant. It's technically complex of like, you have to get some stuff really right, but like really it's very elegant and users really don't have to think about it. That was an aha moment. I think realizing that LLMs could be really useful for this stuff. I mean, maybe that doesn't seem so aha, like end of 2022, everyone was like super excited about LLMs, but figuring out how to apply it in our space. And I would just say there's been a lot of aha moments on that in the last 6 to 12 months, especially as we've been very focused on agents. What actually, Bill, makes for a great agentic analytics product? It's very, it's not obvious. And there's a lot of secrets in here of like, how do you bring these state-of-the-art agents together with the workflow that you have, with the context you have, with your understanding of the users? How do you think about which models are you using? How you're orchestrating these systems? How you're. Providing the way you're setting up your search systems. There's a lot in here that there's a lot of like small aha moments that I think have added up to like something that feels really magical for users. So yeah. Um, but you often have to bang your head against the wall to get there. Uh, there's some one-shot aha things I've had over time. I'm like, oh man, I'm, I'm such a, I'm so smart at product or whatever. But for every one of those, there's like 1,000 where it's like, you know, you're just banging your head against a problem. We've shipped stuff in Hex that I look at and I'm like, man, That is wrong. I don't know what's right, but that is wrong. And it like bothers us for a long time and you have to kind of iterate and figure it out. And like, that's how you build a great product over time. And again, going back to this thing, it's not hard to build a one-hit wonder thing these days. It's like, it's, it's, you need that sustained insight. You need the patience and the determination to like bang your head against a problem for a while before you figure it out.

Nice.

You mentioned like, yeah, product market fit being a moving goalpost and you kind of mentioned like potentially the role of like who a data person is changing. How have you seen kind of like your users, who your users are, change over time and how do you like stay ahead of that?

Well, you know, our core audience was always people who thought of themselves as data people, like they had data in their job title. And I think there's an interesting question of how that evolves over time and all of that. And I'm going to set that aside for a moment actually, because especially just even just quite recently, we've always, we always had a steady pace of non-technical people coming into Hex, like early on. And non-technical, by the way, what I mean by that, really, that's a bad term. But what I mean to say is actually some of them are quite technical, like engineers, PMs, growth marketers, people who think in data and knew how to write SQL. In the case of an engineer, maybe it's like a full software developer, but you know, they don't identify themselves as a data person. Those people came into Hex for a long time. Maybe someone on the team adopted Hex. Data team brought it in and it was a really awesome, flexible way to do some analytics work. What we're seeing now, one of the best, if not the best solution out there for self-serve BI, and it does all these other things that, that, that with the natural language layer on top of it has just naturally, like amazingly accelerated the pace of people coming in. We even see this internally, by the way, like obviously everyone at Hex has a Hex seat, but the level of engagement. From people who previously weren't, didn't think of themselves as like Hex power users has been wild. It has completely changed like our relationship with our own product, our relationship with data. And it's been very, very thought-provoking for us of, of like what this looks like. And our data team, part of the reason I think they like working at Hex, something that's attractive is like, it's fun to work at a company where you're building the tools you use. But it's also really cool right now to kind of be like the guinea pigs in the R&D lab for What does data teams need to look like? Something I told, we just had a new head of data join. Her name is Katie Bauer. She's freaking awesome. One of the things I told her, I've known her for a long time, but one of the things I told her as we were talking about her coming to Hex was like, this is an opportunity to come in and like figure out what this job should even be. And you know, that might sound a little scary, but I tell our data team, you guys have job security. Don't worry. We need you anyway. Like, you know, come in and like, let's figure this out. And then let's go, let's go share what we're learning to everything. and that's really, that's a really like interesting moment and a cool moment to be at. Um, yeah.

That's awesome. So I imagine your entire data team uses Hex.

Well, yeah, I would sure hope so. They better. They better.

Yeah.

You monitor what else they use and it's like, it'd be really ironic if they sent me, yeah, if they sent me some like, hey Barry, can we get a budget to buy this other analytics tool? Um, I, yeah, no, no, they, they use Hex a ton and, and again, it's, it's a privilege and a burden. To be on the data team, you know, the privilege is you get to iterate and partner with the product team. You have a direct line and can just like literally be like, do this. And the burden is you're the guinea pig and the alpha tester for all the stuff that I was going to maybe break on you a little bit.

That's so awesome. That's cool. Now, now, now I want to see it. Are you, are you ready for a demo? Do you want to kind of show our audience?

Yeah, let's do it. Um, cool. Okay. Uh, so, uh, can you see my screen? Yep. Yep. All right. Um, this is Hex. We're here on the Hex homepage. Um, I'm actually going to start by demoing something that we just launched, uh, recently, um, which is called Thread. So I can just ask a natural language question. So I can say something like, who are our top performing reps broken, broken down by a customer segment? Um, this is some demo data, uh, that's like supposed to be like a sort of B2B company. It's not probably wholly dissimilar from Hex. So we'll go ahead and jump, jump in and, uh, I ask that question in natural language and I can see it's starting to think about it. So I can, it tells me the question back and interprets my question. I can see that it goes ahead and starts by actually searching for relevant projects within our workspace. So those are projects maybe that can answer my question that already exist, or there could be used as context. I can see it's searching for model data. So here's some semantic models that it's able to search and it's going through and looking for different things. Nice. It's looking for, uh, tables. So basically looking across all the context we have in Hex, and it's even able to look into some of these and see, uh, you know, context from existing projects that can inform its analysis. And so it's really nice that I can think that, see that whole thinking trace. It's now taking another turn here and looking through some warehouse data. So in this case, it's actually running queries against some of the tables. Now this product is really used by non-technical users who might not be writing SQL themselves. But it is really awesome that I can see the SQL queries it's running to understand the data. So I can follow along and I can understand the logic, but I don't have to fully understand those queries myself because it'll actually explain what it's doing and explain what it finds every step of the way. And so I can see it's actually now taking everything it did, understanding where all the data is coming from and going through to go ahead and, um, uh, sort of do an analysis with that. Uh, and so it'll come back and show me some results in a moment here. Now, one of the really cool parts about this, one of the things that makes Threads awesome is that it's not sort of just a standalone AI chat app, but it's actually using all the context in my Hex workspace, which is where our data team is likely already doing all of their work.

Right.

And it's using all the tools that are part of Hex. So when it's creating visualizations, we'll see in a moment, it is actually using our Explore cell. And under the hood, there's the full Hex project that I can go in and see and look at as a notebook and understand the execution of it. So, this is really cool. So, I can see it come back with key findings, right? So, here's some key findings off the top. I can see a little table that summarizes it. Dang. Should probably be better about column spacing. I can see segment leaders and then I can come and see some interactive visualizations it's put in here. And again, this is a like fully powered Hex visualization that we have elsewhere in our product. In fact, I can click this explore button and actually come in and look at it and slice and dice this data. However I want. So, I can see here the different columns that are available. I can change the visualization type, you know, maybe I want like a stacked column. I can go ahead and do that and customize this chart however I want. But, you know, it's living here as part of this natural language conversational analytics interface. And so, I can say, okay, what are some other interesting trends? Maybe I'm just sort of trends in our pipeline performance.

It's cool. It's not like searching and returning an existing workbook somewhere. It's learning from all those past workbooks and analyses and then creating this new chart specifically.

It's kind of, it's kind of doing both, right? Because it's, it's tapping into those existing ones. It is trying to direct me. It's saying like, you may find these related projects helpful, which is, this is like an existing project that exists in our workspace. In fact, if I open this up, interesting, I see that this is a published data app that someone on the data team probably took the time to come and publish. It's like a dashboard with like a lot of existing context. This is really sweet. And one of the next things we're launching is the ability to like chat off of a dashboard or app like this where I can invoke this same Threads interface and sort of use that as a jumping off point. So I mean, it was sort of almost like initial context. And so this is a good example of it going through and doing its own exploratory analysis against the data in the warehouse. Again, we also have semantic models within Hex, so I can actually tap into those as well. And one of the, again, one of the coolest things is if I hit save as project here, it's actually showing me the underlying notebook that, that undergirds that analysis. So again, it's not just a sort of random chat. There actually is an entire project here with queries, and that's the chart we saw earlier. I can look at the execution graph. So I've mentioned earlier that sort of reactive DAG. This is a pretty simple graph because it, you know, it's just sort of a, a bunch of queries that are running, but I can see there is a sort of dependency here from this query to the underlying visualization and the side query. And if I edited this query, it would automatically rerun those other two queries, but none of the other ones, because it understands the relationship between them. And then here in the notebook now, this is where a more technical user would probably find themselves. This is, you know, a really popular interface that we have a lot of people use. I can go ahead and keep working here with the Notebook Agent. So I can say something like, let's build a forecast model for performance over the next 12 months based on what you found. Cool. So I'll go ahead and do that. And I can see it's going to go ahead and save a version first, and then it's going to start thinking. So a lot of our users have like equated this to like kind of cursor for data. Like you take an existing paradigm and form factor like our notebooks here that, you know, are really nice and really great to use and are able to bring this sort of first-class agentic conversational interface into it in a first-class way. where you as the user are able to partner with it. And so I can actually see there's tools that it has access to here in the notebook. It's using the same framework as what we just saw in the sort of no-code UI, but it's using tools here like being able to look at one of the cells, right? So it actually, I can see the whole thinking trace here too, which is really cool. Like I can see what it thought about that cell. I can see it went ahead and like, uh, is actually installing some Python packages and starting to do some work for me here and building a forecast analysis. It's generating a viz. This one's using matplotlib, which is fine. And it is actually just going to keep going and I can sort of see what it's doing all the way down through this process and work with it to build out pretty much whatever I want with the data in my workspace. And then if at some point this is looking good and this is something I want to share, then I can go ahead and keep all those changes. I can jump over to the App Builder and I can actually drag and drop, configure a data app like to go ahead and share. So actually I'll remove that top summary.

Dang.

You know, here's some of the visuals. There's that initial visualization we had that showed that insight. Here's this forecast model. I could go back and add input parameters so people could play with different parameters heading into that forecast model. And I could go ahead and publish this so anyone else in the organization could use it. And then of course, by publishing it, that becomes a data app that lives in the workspace. That just makes the base of knowledge that the humans and the agents can tap into all the better. So we think this is, you know, a really exciting end-to-end workflow all the way from an initial question someone might ask Maybe they're sending that over to the data team to keep working on it, take it further in a notebook, go and do something more advanced like a forecast model. And then that in turn is like context that the rest of the organization can use. And we think of that as this, we call it internally the virtuous cycle, which is that whole loop from, you know, the data team exploring the new, the novel, and the gnarly. They're taking the outputs of that and using that, creating that as this sort of canonized context. And then everyone else able to self-serve off of that context to answer questions of their own. And then when they hit the wall, they're able to volley that over to the data team to get more help. So there's a bunch of other stuff I'd go into. I'd love to actually show you the stuff we're doing on semantic modeling, but I'll pause there and we can answer any questions or I'd be curious to hear what you guys think.

That's pretty cool. It's like, I mean, just like even that first part too of like, I'm going to go research all these existing dashboards and see how they query this thing. That's like very much a thing that data people do when they're just exploring a new part of the product or a new part of the business that they haven't queried before. It's like, That's my go-to, like, let me see what are like the established dashboards to— it's very boring to do that, honestly. So I like that this does it for you. And the publishing of the drag, like being able to drag and drop and then like publish a report at the end, I imagine you can then refresh that pretty easily. Unlike we were talking about before where it's like, yeah, a lot of my reports honestly are like, I'm taking a screenshot and I'm putting it in like Notion or Google slides or something like that. And then someone's like, can you refresh that? I'm like, oh, I have to see if I like totally screwed up my notebook since the last time I did that.

But the same magic trick twice, right? You're like, well, let me find that query I ran and where's that CSV? And yeah, it's kind of crazy. And so this just brings it all together and makes it really easy and wraps it around with agents. One of the coolest parts, the coolest parts for me, the thing that it kind of surprised us. How much we love this is that it's not just trying to answer a question, it's trying to like help you think about it, right? So it's like lead source gold mines, right? It's showing me this chart, it's, it's showing me like, okay, there's some like really interesting things here around cross-sell opportunities. It's going through and looking at seasonality and I can pick up any of these things with it and keep talking with it and asking about it. And you know, it's not just like, here's the answer. And I think that's what a lot of people got wrong with the initial wave of AI data tools. It was like about like, oh, like, How many opportunities did we have last quarter? And it just brings you back a single one-dimensional answer. What's much more exciting for me is the level of depth this can go to on really digging in and like trying to help you really understand something, whether you're technical or non-technical. That's, that's, um, that's a line I'm very excited to keep pushing on.

Yeah, it's like insights instead of just findings.

That's exactly right.

Yeah, I mean, we have leveling guides for, you know, what's an IC 1, 2, 3, 4, 5. This is like a senior. This is like a senior level here.

Yeah. Well, it's interesting you say that cuz it goes back to that data roles question, right? Like, does this replace a data analyst? Well, maybe, maybe not. I mean, it replaces a lot of the things the data analyst might have been doing more as like point analyses, but it also makes them better. Like there's a lot of questions in the organization. A lot of things your data people wanna be looking into. And ultimately what really matters is what's behind this. And so, you know, I'd love to jump over and actually show you, you know, I, if I click into the data tab here, I can see. That we have all these tables that our data team has actually gone through and added categories or statuses to. And actually see we have some data models here, like these semantic models here in our workspace. So this is a, you know, this is a demo of using Snowflake semantic views. I can go back and see I've got these other semantic projects here in my workspace. Okay, this is a good example. So I can see that this one is actually imported from Qube, which is an external semantic model vendor. There's others that we can bring in from Snowflake here. So Snowflake has semantic views defined in their product. We can also sync semantic models from dbt. So a lot of people use that, but now we actually have in Hex the ability to author semantic models as well. Cool. So I'd open up this other semantic model that this one's actually authored in Hex, and I can come here and edit it and actually, um, work with an agent to edit this. So I see this is the underlying semantic definitions in YAML. I can see I've got a few files here. And I have a similar sort of agent interface over here, and I can actually even just do something pretty open-ended. I can be like, what are some ideas of things to add to my model? Like, that's not a thing you might normally think, but like, you know, actually we've even internally found it to be really interesting and useful to be like, okay, like, yeah, what should I do next to enrich the semantic model more? And a bunch of the things we're working on now is actually sort of upstream of that in that virtuous cycle loop. That I was talking about, which is like, well, what are the questions people are asking? Where are they getting good answers? And where they're not, can we bring a experience to bear for them that is actually able to help the sort of curator users go and like enrich their workspace's context to be able to create better answers? And so I see here, like, I've got some ideas here. I can say like, let's try one of those ideas. I don't just, obviously a demo is probably not something you're going to do for real, but what we'll see here is it's actually going to be able to go and edit these semantic models, either create new models. Edit the current ones, and very similar to what we saw in Threads with the notebook, I can kind of partner with it to edit and sort of update my underlying models here. So it's thinking. Yeah, I can see it went ahead and added a couple of new blocks to this model, right? Doesn't look like it's done yet. It's still going. And so it actually just went ahead and validated the model for me. So it ran a successful validation with no errors. It might be summarizing the result for me right now, and I'll let it just keep thinking here. Cool. And if this looks good, I could actually go and publish a new version of this model out. I can see You know, here's the diff. I can see which projects are actually referencing it. So actually nothing's referencing it. So this is a safe and easy one to go ahead and publish. So I can say update and go ahead and publish it. And now I've just updated my semantic model.

So cool.

And if I went back to Threads or went back to the Notebook agent, it would now be able to use those updated definitions in its answer.

Sorry, analytics engineers.

No, no, it just makes the job. I think it makes the job better. Like our internal analytics engineers love this the most.

Dude, no, this is sick. Hey, what does that references thing do?

Is that like a lineage thing?

Yeah, it shows you, um, anything downstream. It shows you the downstream projects that actually use it, uh, there. So, okay, this is the— this is our B2B SaaS semantic model that actually should have been demoing off of. Um, so I can come in and edit that. And if I, um, like just do a kind of no-op, uh, edit here, um, when I come over to references, I should be able to see that there are a bunch of projects using this model and see like the downstream dependencies. And then obviously you can imagine. The next step for us is help you better understand what your, uh, changes kind of impact your changes will have on those projects and, um, be able to go through and, you know, resolve if it's going to cause any issues. And you can imagine a bunch of ways in which we could have agents be part of that workflow too, as an example, like, hey, there's these 10 downstream projects that are likely to break with this change. Do you want me to suggest fixes for them and stuff like that?

Yeah. So cool.

That's awesome. Yeah.

Yeah, we're pretty excited about it. I think, uh, I think it's the coolest stuff we've ever built. And, um, it's like a line we're just going to keep compounding on. And like I said earlier, just sort of more philosophically, like it's not just about one big breakthrough you have in the course of building a product. It's multiple and they ideally they stack and compound. And so the only reason we can build a lot of this cool stuff is because we built stuff before. Like the reason Threads works so well in part is because people already do a lot of their analysis in Hex and it can tap into that context. And we have things like rich visualizations already built out. Part of the reason our notebook agent experience is so awesome is because we already had a great notebook that the agent can basically use as context and context and like tools to be able to have a really awesome workflow. And so these are, these pieces, again, they combine and they compound in a really, really cool way. So I'm really excited about that.

Hey, on that note, when you're like onboarding a new customer, is there some period where you're like, get to this much usage or leverage of Hex and it's really going to start shining? I mean, I know Threads is new, but do you have thoughts around that?

Yeah. Traditionally, the way people would adopt Hex was very much like started with the data team. So you'd have like these people who are mostly using the sort of notebook interface come in, use it. And really early on, especially like that, we were really just kind of— they were our people. As I mentioned earlier, we saw this expansion over time. But one of the awesome things is it was just really easy to get started with. You could just start, connect Hex to your data warehouse, just start writing queries and you're off to the races, right? There wasn't some long implementation period. I think what's interesting now is before you set a bunch of people loose with Threads, you actually do want to take the time to curate some context and do things. With net new customers, we're finding ourselves embedding much more closely with them, helping to build a lot of that out. Uh, get that foundation established. But with existing customers, I think it's really cool because they actually already have a lot of stuff in their workspace. And yeah, if they want to turn on Threads for that next set of users and really empower conversational self-serve, we want to make that really super easy. And there's a bunch of things we're excited about there. Like I just showed the modeling workbench. You can imagine on a lot of things we could do around recommendations and suggestions, like how to get, hey, based on the stuff you're already doing in your workspace, here's some really neat things. Here's some great things you could go do and configure for context for for people, what types of questions do you want people to be able to ask? Oh, okay. In that case, you should consider building this kind of semantic model and like, we can bring agents for that whole experience.

Wow. That's cool. So you're, you're, uh, you're almost embedding data strategies as part of your future rollout.

I was thinking the same thing. I was like, it's going to start doing quarterly planning for the data team.

No. Well, no, I mean, that's their job. That's right. Like, that's what they're— the great data team is like, you know, thinking about how they're setting the foundation so that they and the people they're supporting can go in and, you know, ask and answer questions really well. We want to help kickstart them on that, but ideally our product should be able to create a nice flywheel so that we're supporting them doing that going forward. We're making their lives really easy and they're doing it.

Nice. That's so cool. Um, hey Barry, you, you, you have, you've, you've said this a little bit, data people are problem solvers. So if we zoom out a little bit, What's your take on how AI will reshape data teams and workflows in general in the next few years?

Well, there's a few things I'm thinking about a lot. I do think the curatorial librarian, and I don't mean librarian like pushing books around on a cart, I mean like organizing knowledge, like that facet of the data profession will be, you know, even amplified much more. We're even seeing this internally. Like we have a lot of appreciation for our data team already, as you can imagine, being a data company, sells to data people. But even now, like, you know, we're using the crap out of Threads and the Notebook Agent internally. The adoption's off the, off the charts. It's not making us appreciate our data team less. It's actually making us appreciate them more because we're reaching out, partnering with them, going deeper on analysis, understanding, is this the right way to look at this? Can we get some additional context added? They're, they're able to use these tools too, to go deeper on the types of problems they're trying to solve in the business and the company. But it is really interesting in that I think it, it's already, we've already seen their role kind of shift from being like so bottlenecked to actually almost being more pervasive in terms of partners across the business, which is really cool. The second big thing I think we're going to see a lot of is, um, as the kind of dismal tide of quick questions ebbs, that like this sort of daily interrupt of like, hey, can you pull this number is something that an agent can be the frontline for. There's a couple of things I'm really interested in. One is how can the data team be able to spend their time on, on that deeper dive work? And can we want to make sure we're not only building the tools to enable the self-serve side, but keep doing a great job on the deep dive stuff. Cause we think there's a lot there and a lot we can do for those technical users to help them really, really understand things, to help them set up experiments, really research stuff, take action. There's a really rich set of stuff. I think that we can be. Helping them do to, to enhance that side of their job. I also think a lot about how they can be observing and managing the answers everyone else in the organization is getting. And so I mentioned earlier some thoughts we've been having around, we could sort of think of as like data observability for like almost inside observability. And can they be like getting involved in partnering with people on the right way to think about something, the types of questions they should be asking. And I think if we can give them the tools to understand those patterns where people are consuming data internally and. Be helping manage that at a meta level. I think that's really interesting as well. So there's a bunch of facets of the job. I don't think any of these are necessarily new, but I think they are all going to be like greatly amplified by AI as other parts of the job, like doing a lot of drudgery syntax, YAML indenting get attenuated. Like you're going to see that shift and exactly where it is in 3 to 5 years. I don't think anyone really knows. I do know that people are going to be consuming more data than ever. And you're going to need data people there to help make sure that's happening accurately and well and efficiently. Nice.

The world that you're describing is, I remember seeing like a pyramid of, I forgot what it was, but like maybe like the type of work a data person would be doing. And then you have like the very bottom layer, like just doing foundational stuff. And at the very top is like, oh, I'm predicting the future. I'm doing crazy stuff to help guide the business. So it almost sounds like Barry, you're trying to invert invert that, uh, that pyramid over there.

We're, we're working on it. And I think it is funny how much of a love-hate relationship people have, like data people have with certain things. Like you hate that you're bothered all day by data questions, but what happens if they go away? You know, like you hate, you hate doing a bunch of this sort of drudgery backend work with the, that is sort of endemic, you know, this sort of data plumbing and data munging that's endemic to trying to do a data analysis. But what happens if a robot's gonna do all that now? It's not surprising that there's this identity crisis. This, right? Change leads to uncertainty. Uncertainty leads to anxiety. And one of our goals with Hex is to help answer those questions. Like, what, what, where can you be still doing your best work and actually be even more fulfilled as a team? And again, I, I don't think the thing that ever defined good data people was like code syntax. It was all these other things. Yeah. And none of, a lot of those other things are not gonna go away. And I'm really, really excited to see where that goes and maybe play our own small part in defining that.

That's awesome. That's cool. That's, that's, that's great. I feel like, uh, that, that probably is a good, good way to end our podcast episode here. Would you agree, Sean?

Yeah, no, this is so cool.

I appreciate you guys having me on.

Yeah, no, it's, uh, yeah, thanks. Thanks for coming on. I mean, it was great seeing this and I'm sure our audience would, uh, would take a ton away from this. Yeah. Uh, any, any sort of final piece of advice or kind of thing that we haven't gone through that you would like to bring up?

The thing I'm just talking to a lot of people about these days is, um, I've actually told some people recently, very surprisingly, that they shouldn't, they shouldn't buy Hex, which sounds very funny, right? But the reason I told them that is because I thought that their expectation, that they're not set up in their organization in a way that's realistic for what, like, we're seeing a lot of things now where there's a, there's a tremendous demand from business users to be able to like chat with my data. And a lot of people just looking for a tool they can throw at them to like satiate that. And I think these setups are not always going to be set up for success. I think you as a data team have to be really, really invested and not just like giving them a thing, but actually like wanting to really partner with them and wanting a system that brings both sides of people to the table. Like, oh, there's a big belief of ours is like, you don't want a standalone chatbot thing for the business team. You don't want a separate tool then for the data team to be using. And so. I really think that if you're on a data team or you want a data team that has that level of partnership and connection, like we'd love to, you know, you should check out, come visit our website at hex.tech and get a demo and all of that. But if not, like Hex won't solve that problem for you. And we do see teams sometimes buy Hex thinking it'll solve all their problems, but there are a lot of the problems and the ways that they have to solve things is actually much more people-oriented. That's something I've been thinking a lot about. And I'd encourage everyone to think about before they just reach for a new shiny tool. Thinking it'll solve all their problems. Yeah.

I think that goes for any sort of like, uh, wishful thinking of self-serve, one-click solution. Like, uh, it's not really how it works.

We're one more tool contract away from data bliss, right?

Like, I mean, I'm a tool vendor.

Like, I should be telling you my incentive is like, no, you are one tool away. Just one. You are one term committed, full free enterprise contract with Hex. Away from solving all your problems. But I think that's actually— 10-year commitment. And usually people are just going to be disappointed. What I want is a lot more customers who are finding that Access is like incredible for their teams and their workflows. And that takes not just the technology, it takes like people and process and organizational stuff. That's a huge part of the job.

Yeah. Organizational and personnel problems are the hardest to solve.

Yeah.

But once you solve it, it's, uh, it's just a tools problem, right?

That's right.

You see how Barry came in there at the end and then was like, make sure to hire more data scientists. He said he's, he's keeping our jobs in there. Don't worry. He's, he's given us some job security there.

That's right. That's right.

Appreciate it.

I do what I can. Well, this is awesome, guys. Thank you for having me on and I'll maybe come on, maybe come on another time with more updates down the road. Love it.

I know. Anytime, anytime. And to our audience, thank you so much for listening and we'll see you next time. Please don't forget to like and subscribe and drop any comments or questions you have for Barry. See you guys.