Meet Hex, an AI-Powered Analytics Platform with Notebooks, Apps and more: featuring Rachel Herrera

0:00Rachel, how are you doing?

0:02I'm doing good, Tim.

0:04Nice to meet you.

0:05Awesome.

0:06Thank you so much for coming on the channel.

0:09It's probably worth me

0:11Just explaining how this conversation has come about.

0:13It's actually been a year in the making.

0:15You probably don't know this because I know Shah reached out to you probably a couple of weeks ago, but

0:19I've been a huge fan of Gabe Fomo who runs your social media content, your video content.

0:25I've always thought his videos are are fire and so I reached out to him like now a year and a half ago saying, hey, I'd I'd love to just chat

0:32Learn more a bit about hex, maybe you could walk through, you know, hex for the table audience.

0:36And so we went back and forth.

0:38Gabe's a busy person, I'm busy as well.

0:40We kind of let that drop.

0:41And then um

0:42Charles recently reached out to me and said, hey, we should actually do this.

0:46And this is something that I'm obviously pushing on recently.

0:48Yeah, you're here today to hopefully walk us through hex.

0:51So may maybe if I hand over to you to give a bit of an intro and

0:54Tell us who you are, what you do at Hex, and then yeah, we can get into it.

0:58Yeah, I love that.

0:59It's funny you should say that.

1:00I think Gabe's videos were also a big part of me really enjoying Hex because oh these people have a great

1:06sense of like how fun it is to work with data and they're just like cool with being a little cookie, which I love.

1:13But no my

1:14My career has been in data for the past ten, eleven years or so.

1:19Sorry.

1:20That's platform.

1:21Um

1:23No, yeah, so I I started out as a marketing analyst and moved into consulting, actually a fair amount of Tableau consulting, Power BI consulting.

1:32And then moved into the SaaS startup world with a company called Amplitude, where I did professional services there, standing up there analytics software for product analytics.

1:43And I just really love data.

1:44Like I don't ever want to not be in it, but I think the fun part of it is sitting in between the business and the technology.

1:52So

1:52figuring out how to teach the business to ask good questions of data, building good systems that let them do that.

1:59And I joined Hex.

2:01at the tail end of last year as their product evangelist, which that kind of a job has been like like on my vision board for a very long time because it I think it it combines all the things that I love to do, which is

2:13teach and talk and show, but then also stay very close to the doing.

2:17Because I don't think you could be a good evangelist if you're not actually a practitioner and like actually doing this work and can speak to it intelligently

2:25And so it forces me to do all those things, which is great.

2:28And yeah, and so I've been at Hex now for a little while, really focused on telling the story of what problems it solves and who it can help, but also with this AI world coming about

2:39How does that change the world of the analyst, of the data practitioner in general, of the business user trying to use data?

2:47And where does a tool like Hex fit in all that?

2:49And like how are we building towards sort of this future where

2:53the robots do a lot more for us and how do we sort of stay in control?

2:57How do we maintain trust?

2:59How do we maintain the guardrails?

3:01Security, compliance, all the things that like a business needs to care about when you're thinking about data questions

3:07Um, but get the freedom and the fun that I think a lot of the AI tooling lets us do, like just create unencumbered by writing SQL, writing Python, clicking a UI.

3:18So it's been a really fun, like fast paced five months.

3:22Like I I think in startups, like I I call it's like dog years.

3:26Like

3:27Could be at a company for a year, but in the startup world it's actually seven years.

3:30So I've basically been at Hex now for about three years.

3:33It's it is a super exciting job.

3:35I think we are both evangelists from different perspectives.

3:37I've obviously gone through sort of the content

3:39creationary, but there is a lot of synergy as you say in terms of um there's almost like a translation role, right?

3:45Like sort of bridging the gap between uh I guess in your case where the product sits and

3:50where people perceive it and actually just closing that gap um to to sort of make either specific use cases more real or just how the product works

3:57More real to people, people who are using it.

3:59So yeah, I feel like we have a lot of commonalities there.

4:02I've also been in the industry for 12 years, so around about the same amount of time.

4:06Uh so far I've been stuck in one place for a while.

4:10I really value your sort of journey in an interesting way because I it feels like you've definitely sampled different product philosophies along the way.

4:17uh amplitude you mentioned Francois who was previously uh on the channel he was chief product officer there for a brief moment he might have overlapped with you right a at that time.

4:26I don't know if you joined right before you left or the timing of that because yeah

4:31Yeah, I think it was like after I had left.

4:33Right.

4:34Yeah, just after Rap.

4:35But I wouldn't love the chance to work with him because he's legend.

4:37So that would

4:38Great.

4:38Yeah.

4:39You've used Tablay, so you'll have definitely been impacted by I mean he was also a Power BI before that's it.

4:43I know you've got a bit of Power BI background as well

4:46So this is super exciting.

4:47I'm super passionate about Hex.

4:49Now let me be sort of completely transparent here.

4:52One of the things

4:53that I've always felt about Hex is when I come across Hex users, they love the product.

4:58They really like the way that it works and the workflow that it supports.

5:02And so

5:03Maybe this is a good bridge for us to do two things.

5:05Maybe give you a platform just to explain what Hex is and then we can jump into a demo and then after the demo we can uh you know just sort of have a sort of an analyst conversation about the tool and yeah, we see where we go from there

5:17Definitely.

5:18Yeah.

5:18I guess to sum it up in a sentence, like Hex is the best AI analytics platform.

5:24I think I can say that.

5:25since I'm evangelizing it and I work for the company and I I strongly believe it as well.

5:29And I think it it's true for a lot of different reasons.

5:31I think the first thing is we really care about the workflow, which is why I think a lot of our users

5:37are so happy.

5:37And it's what drew me to it even three years ago when I didn't work for the company.

5:42I I had former colleagues that had gone techs who I really respected.

5:47And so when I saw that they went there, I was like, ooh

5:49What's what are they working on?

5:51And I saw very early, like old versions of it and was I could see the potential because it feels very built by people that know and understand what we're trying to do.

6:02So you get so much delight from a lot of the, oh, it just works.

6:05It just does the thing that I would innately expect it to do.

6:08So I I'll kind of show you that in the demo.

6:10Yeah.

6:12always comes to mind for me is everything you make in hex is reusable.

6:15Yeah.

6:16And it's and there's not like a big heavy process to like ETL something.

6:21You just

6:22build a cell and you create an output of data and then you can do stuff with that output of data.

6:28And it really speaks to like the iterative nature and process of analytics and like the shaping of things.

6:34And then ultimately what we are under the hood is a notebooking tool, first and foremost.

6:41And the notebook has all of these different pieces and layers underneath it that lets you build

6:45a ton of different kinds of analysis using SQL or Python.

6:48You can write create charts.

6:50You can use Markdown.

6:52You can create call APIs and like even throw in like a chat GPT.

6:57response cell, like all this stuff, because it's all code-based under the and in those notebooks you can turn them into data apps that you would publish similar to like a dashboard.

7:09where we're very different because I feel like that sort of is yeah every other BI tool does much of the same thing.

7:15I think where we're different is we let you write Python in the same place as SQL, which is pretty cool and it it gives a lot of

7:21analysts who are also maybe data scientists or data scientists who also do analytics, or an analyst who wants to do more Python but doesn't want to download Jupyter and spin up a Jupyter server.

7:32You can do all that in one place.

7:34What I think makes Heck special is we've really leaned into AI as a core component of the platform.

7:42So we have agents on virtually every surface.

7:45So you have an agent that can help you create a notebook and do deep analysis.

7:50We've also got an agent that will help you do conversational self-serve through our feature called Threads.

7:55We've got an agent that'll help you make semantic models.

7:58And we've got an agent that actually reviews the work of the other agents and tells you where you could be getting better.

8:04And we have a whole like context studio that lets you

8:07curate and create context for your agents and manage conversations.

8:11So I'll show all this in the demo, but I guess to round up my point, it's not a we're just a one thing.

8:18Like I think Hex is really building towards being

8:21the unified platform for an analyst or it just really anybody in general that needs to get an answer with data and wants to leverage AI.

8:30to get that answer and making that workflow really delightful all the way through.

8:34Amazing.

8:34That's a really uh succinct description.

8:37I've sort of dabbled with some of those, let's say, bits of that description, but I think you've really sort of nicely encapsulated it.

8:43And I think

8:44Before we get into the demo itself, I like this discussion around workflows and I think maybe we'll discuss this after the demo.

8:50Like the surface of work is really, really important.

8:52Like where people actually

8:54Do their work to me seems to be the defining sort of underpinning of analytics platforms going forward.

9:00Before you came to the platform to do your work and have the work routed through the platform, but it almost feels like as you said

9:06You know, giving analysts the options to do SQL or Python in the same place, but then not have to refactor that work to do the to do other bits is actually um a really important thing.

9:16I have many more questions, but maybe this is a good jump off point to to see a demo.

9:21Yeah, let's jump in.

9:23So jumped into the hex platform.

9:26When you land on the homepage, you get greeted with your kind of classic prompt bar.

9:31But I want to speak a little bit about like how data arrives here and what that process looks like because we are essentially connecting to a cloud data warehouse.

9:40And so if you've got data sitting in Snowflake, Databricks.

9:43really I I I struggle to think of a kind of s mainstream data storage or data warehousing tool that we don't connect with.

9:52You would just pass your credentials and then we're gonna query it just like any other tool.

9:57So there's no live versus

10:00Downloaded connection, if that makes sense.

10:02Like there's no extract as like a concept here.

10:06It's very like we're touching the data live from the warehouse.

10:09And then at the notebook level is where we start to filter out that data, bring some things in memory to speed things up, but we're not having to do a trade-off of I've got to bring all the data down before I can do any sort of transformation to it and trade off on

10:24how fresh is it versus how fast is the query.

10:28And this is just a demo environment, but you can see here we we can mix and match as well.

10:33So we can have a Snowflake data connection, a big query, a ClickHouse, a Postgres.

10:38And data connection is like the parent to any sort of sub-analysis that you would do in hex.

10:45So start with the data connection and then build from there.

10:47But you can also mix data connections inside of a notebook.

10:50So

10:51Not to say that that is one-to-one, but a bit of code can be written against one data connection.

10:57And what's also nice is

10:59Hex really treats like metadata as a first class citizen when it comes to connecting to your data warehouses.

11:05So if you've got things like

11:07DBT decorating all of your columns and tables with definitions and descriptions.

11:14We can sync all of that down into hex

11:16And we will display that and like it's also something that our agent can read, read and review.

11:22So we've got like a table description here, and then I can see like the different column descriptions as well.

11:27And then here's where you can also start to think about like, if I'm bringing data in, how do I want it to be viewed by that agent?

11:35And how do I want it to be leveraged by everyone in the organization?

11:38So I could

11:39include this for AI or I can exclude it as well.

11:42So we give you a lot of fine-tuned control on the data before you even kind of get to

11:49Curating context and creating all of the different assets and artifacts that need to exist.

11:55Which is nice.

11:58So that's the boring data bit.

12:00I like to just touch on that because I think it's like really important to understand how we interact with your data sets versus maybe an old world where you might have had to like extract or

12:11prep before you can even do anything.

12:13Yeah, so let's talk about some of the workflows in here.

12:16Cause like I said at the top, I think that's one thing that we do exceptionally well.

12:20And let's say I'm like an analyst and I need to come in and do a deep dive analytics project on something

12:28One of the data sets we have here is like a B2B SaaS data set.

12:32So maybe it's on sales performance or sales forecasting.

12:35So I'm gonna click project.

12:37And in Hex, a project is analogous to like the main workspace for where you would do the deep data work in notebooks, and then you could also turn those notebooks into what we call data apps.

12:48And so in the notebook area, I've got this sort of blank canvas and I can do a bunch of things.

12:53I can add SQL query and in there I can start to

12:57handwrite my sequel or it can be very modern and ask the notebook agent to actually go and and start building things for me.

13:05And

13:06the tau of hex, the underpinning of all this is we think that everything is a cell.

13:13And so e every time I create something like this, or I add something for text, these are all cells.

13:20And now each of these cells produce outputs and they could be a data frame, they could be a pivot data frame, they could be a chart.

13:29And then all of this analysis becomes chained together.

13:33So they become dependent on one another.

13:35And you really get that feeling of, okay, I started with this really big table and I've reduced it down.

13:41And then in another cell, I can go and do maybe a window function across it to do some sort of aggregate.

13:47And then I can take that and use it in a chart and create it.

13:50But then I could also take that output

13:52and do some Python against it as well.

13:54So you've got a lot of like malleability of stuff because you're not stuck having to

14:00pre-prep all of the data, you get to do a lot of that fun stuff in the same place that you're also visualizing it and telling that story.

14:09Yeah.

14:10So um here's where you oh go ahead.

14:15It's really good.

14:16I the modularity of what you're talking about, I think, is something that people are more interested in now than they were before.

14:23Just because I think

14:25Maybe AI has changed where that line of comfort is for some people, if that makes sense.

14:31Five years ago I would never have thought that this would be a workflow that I would entertain typically, but I think in today's landscape

14:37People are definitely more curious, especially with the agency support that you've got there.

14:41I think that's I'm not sure it's like a perfect fit for everyone, but I think there's a lot more people now for whom that line has shifted and this would be something they'd be a lot more interested in.

14:49Yeah

14:50Yeah.

14:51Especially if they spend a good bit of time maybe writing SQL in one place and then copy pasting it into some visualization software.

15:01turning that into the version of it for that particular piece of software so that it can make the chart that they want or God forbid they build it in Excel, which I Excel, she's c she's still got a place in this world.

15:14But I think this sort of brings a lot of all of that into one place, especially with the SQL to Python translation.

15:21Also you could even upload CSV files and rake SQL against a CSV file, which is pretty

15:28And it's loading that into Snowflake or Databricks for you, I assume.

15:31So if you put that CSV file in there, it's also available in the future if you want to do something else.

15:37In this particular project.

15:39So the project is running in its own

15:42Kind of environment.

15:43And so as you upload things into it, it becomes aware of those things in the context of that project.

15:50I see.

15:51But this could be cool, let's say you have a document in Google Drive that gets updated by some process.

15:57You could share that in here and then it becomes part of your data set.

16:00Maybe it's a mapping file or it's a a forecast.

16:04Spreadsheet that's managed by a different team that you don't touch but you still need as part of your analysis.

16:09Yeah.

16:09And just a brief question, because I know my audience has people who still use things like network drives that.

16:15Is those sort of cloud connections the better approach in that context?

16:19Or do you have yeah, do you support things like network drives that might sit within an enterprise setting?

16:24I'd have to check on exactly what that piece looks like.

16:28Uh no for a fact Google Drive.

16:30I think the Microsoft Suite's a little bit difficult, like one drive I'm not as sure about, but

16:36I can go check.

16:37Yeah.

16:37No worries.

16:40Something's top of mind at the moment in another piece of work I'm doing.

16:42So I thought I'd ask.

16:44But no worries.

16:44I'm disrupting your flow.

16:45I'll let you carry on.

16:46Oh no, you're good.

16:47So let me just type a real simple And we've also got like

16:51like smart typing here so I could just do from and then in in the right hand corner here is where I've like

17:01I've selected my data connection.

17:04And so if I wanted to, for example, actually look at the data that's available to me, I can open the data browser again from the same surface that I'm working.

17:12And I can see the different schemas and databases and also the tables.

17:16So let's say the dim customers is what I would want to do here.

17:21I can add it.

17:22Very quickly I could just copy the table name and paste it here so I don't have to get fancy with it.

17:28Or if I'm starting to type things out, it should get smart and kind of give me that like autocomplete autofill.

17:34Although it does quote, which is geared

17:36And so if I do a quick little command shift D, I run my cell.

17:42So that's essentially running that query.

17:44And then now I have to do that.

17:45There's your table.

17:46Nice.

17:46And I can start to put filters on this.

17:48Maybe it's like where handle date is brighter than some date.

17:55And then I get filtered down and then I have that as a data frame that I can go do stuff to.

17:59So maybe the first step in my analysis was give me all the customers who are renewing in the next

18:05quarter and then forecast their renewal likelihood or understand where we need to spend more time nurturing them so that they have a more successful renewal or upsell.

18:17Now I could hand type all of that.

18:19I can build all of these different cells, or I could just tell the notebook agent that's what I want to do.

18:24And I could give it a pretty sophisticated prompt.

18:27What's it?

18:28I I want to

18:30build a I'm like such a slow type when people are watching me so we can just cut this port out.

18:35I spit it up.

18:39Small

18:42Now just get a quarter.

18:46Maybe it's like breakdown by customer segments and accounts.

18:53Executive.

18:55And so just with that, now the notebook agent, which has all of the tools that I have.

19:02So it has the ability to write SQL, it can write Python, it can create charts.

19:06It can format charts, it can do all these things.

19:09It's gonna do its thinking, it's gonna plan it out.

19:12It's gonna go read all the context that's in the workspace.

19:16So as you can see, like we have this concept of guide.

19:19that gives it the context around how we segment our customers and how we think about ARR performance.

19:26It also is going to look through all the different data sources that we've given it access to and read all the metadata about that data

19:33It's also going to find relevant projects, which is really cool and worth me pausing on because I think a lot of the tools today that are incorporating AI do this first bit.

19:46Reasonably well.

19:47Yes.

19:49Give us the metadata that we know exists and give us the unstructured context that we know is good practice, like skills or guides.

19:58But as you're building stuff, that also becomes context.

20:04You create a wonderful project with a great analysis

20:07and a well thought out methodology and you publish it and you share it, then the agent should be able to read it as well and see it as, oh, that's a great

20:17example of SQL that I should go copy and that's a great example of Python that I should go replicate here

20:24And so makes sense.

20:26This this is by far, I think, one of the like the coolest pieces that I think keeps Hex feeling really tied to like what we care about today as an organization.

20:37Because as you're building things, that context is getting more rich and it's like the system like gets smarter the more you use it, which is pretty cool.

20:47And it's also great from like a I'm assuming there's things like logs and metadata that comes off this activity as well.

20:53So it's makes observability maybe a little bit easier.

20:57People also want to know that their investment is going the full way.

21:00So if there's something that's already good context for what you're doing, it helps for the to auto-surface it at the right time, not unnecessarily make you have to go find it or bring it together and then start your work

21:10Yeah.

21:11That 'cause I think that was the that's the old world was like I think there's a thing that sort of says what I want

21:18And it's got a cell in it that I think is what I'm looking for.

21:21Let me go find that so I can copy and paste it into my project so I'm not spraying from scratch.

21:25This essentially like removes the need to do any of that.

21:29Yeah.

21:29So as you can see, it did a fair amount of thinking.

21:32It wrote all of these new cells.

21:35Uh-huh.

21:36It created a summary, a segment in a risk summary cell.

21:41It also did like a renewal cell, et cetera.

21:44And so now it's asking me, hey, to I want to finish this out, but I also want to clean up this layout because it created some cells that it actually doesn't think it needs anymore.

21:52Um so you can see that there's this empty Python cell I created, and then there's empty text cell, which I created, and it's like, hey, let me clean up your workspace and so accept it.

22:03And then it'll just go clean those up.

22:06And you can also see that all of the things it's building, I have the ability to keep or reject, which is pretty cool.

22:15And so I'll just interest it.

22:17And this agentic stuff like this is working in the product today, right?

22:21This is not.

22:22I asked that.

22:23You'll understand why I'm asking that.

22:24My audience will understand why I'm asking that.

22:27It's not a hand wavy demo, I probably.

22:29Yeah, yeah.

22:29And I I like that it's also suggesting, hey, here's here's here are a couple of things that are I don't need anymore.

22:35Like that that is a very useful step because in the syngentic workflow it's very easy to create

22:39But I think one of the things that isn't often discussed is sometimes the process of being an analyst is about the process of reduction and really refining the output and in that refinement it's about actually making good choices about what you no longer need.

22:54And removing it and that's part of performance tuning, but it's also part of visual analysis.

22:59And so I like that the agent is at least doing its contributory part.

23:02It's creating a ton of work.

23:04I'll ask a question later about the sequel that it writes, but

23:07It's also, let's say, self-reflecting and say, hey, I don't think you need these two and here's why.

23:11But giving you the choice to to really drive that, which is nice.

23:15Yeah, exactly.

23:16Yeah, yeah.

23:16Good.

23:17And now it's handed itself off.

23:19So it's written all of the code, the SQL code blocks that it thinks it needs.

23:23And now it's handed off to the visual, the data visualization sub-agent.

23:28And so that agent's job is to create beautiful charts that are legible, that are well formatted, that have nice color kind of story behind them.

23:37And so now it's doing all of that work of creating this chart, this one down here.

23:42So as you can see it's like being intelligent about okay, low should be green because that's good, medium is yellow and red is high.

23:51And so it's got they've done such a fantastic job of tuning the visualization sub agent.

23:56Our visualization team was actually like a big part of this.

23:59And so they had a lot of no, it has to be good.

24:02Like I I

24:04that you it's gonna look at these labels here and be like, these are hard to read.

24:08I'm going to remove them.

24:10Yeah.

24:10Um if I were to ask it to come back through and do another round of cleanup.

24:15Yeah.

24:15And so it's done

24:17So well we chatted, it built all that stuff.

24:21It and then it gives me sort of the summary of what it found.

24:24We've got 258 renewals.

24:26contraction expected of 120K, call-outs, and then now I can take the next step, either keep going in this analysis, ask it more follow-up questions, or

24:37I could flip over into what we call our app builder, which is where I was gonna say.

24:41Yeah.

24:41Yeah.

24:41You get you get this sort of more dash dashboardy experience.

24:45Yeah.

24:46So I'll pause you here because I feel like this is about the limit of my knowledge at the moment with HEC.

24:51So if I allow me to play back what I understand and then you can tell me if it's right or wrong.

24:55Our notebook is where we do our, let's say, not development, but yeah, is where we curate our story.

25:01That's probably a better way to put it, right?

25:02You're doing your analysis.

25:04You're curating what you need for that analysis.

25:06And it's got this sort of top-down approach.

25:08So it's almost like a notebook, exactly like a notebook in fact.

25:11You start with the highest level of analysis and then you're drilling down as you go further down

25:15Building up your story, the app builder is basically taking those same components.

25:20And what I understand is that the thing that is the SQL query on the left turns into the underlying data source here.

25:26And so

25:27You're switching modes, but it's actually the same asset just with different perspective.

25:31Is that a fair way to put it?

25:32No, that that's on point.

25:33And the way that I can explain

25:35Expose something to the notebook or excuse me to the app layer is I we click this little button here.

25:40I see.

25:41There you go.

25:41I'm learning as well at the same time.

25:43So that's like that's how

25:44I was gonna say, how did you toggle what goes on and what doesn't, but that makes sense, okay.

25:48You can tell I didn't follow the instructions.

25:51Oh, it's fine.

25:51It's a very hidden secret.

25:53little button here, but yeah, yeah, but it's the it's easy to skip.

25:57But I think what's also nice is I maybe I like this, but I also maybe want to add like inputs and I want to let people be able to maybe filter

26:06So kind of getting to that more dash 40 experience that's not just here's the data, but also like how do you interact with it.

26:14I could by hand do things like add our inputs here.

26:17So I could put if we want a multi-select or we want a drop-down.

26:23And then in those inputs, I could give it a name, I could make it dynamic, and I can choose data from my upstream outputs.

26:33Steps.

26:34So it's uh limit to list, it's all like self-referencing, which is cool.

26:39So I could pick my data frame and I can pick the column.

26:42So maybe it's account segment type and then

26:46Yeah, that's it.

26:47And so now I've got a drop down.

26:49And then I would have to go and I could give this a nice name like account segment.

26:55And then

26:57The way that I let this filter interact with the data below it is I start to add what we call I think it's gin ginja.

27:06It's something I'm not.

27:07Oh I see.

27:08Yeah, I've got a ginger.

27:09It's used in DBT, I think, as well.

27:11Yeah, makes sense.

27:12And so now this filter is calling the value from this input.

27:17And I could do that hand

27:19Or I could just have my agent do it for me and say, hey, I just created this input called account segment, add the filter to every data frame or every SQL cell.

27:28Yeah.

27:29Not to compare your product to other people.

27:31Sigma has something similar with actions and filters and Tableau has the similar thing with parameters, right?

27:37It's th there are like analogous

27:39concepts.

27:39I think what is interesting here is where you do it, how you do it.

27:42Here in the sequel it is pretty interesting because I think it goes back to this concept you caught talked about and an environment.

27:48So what what I think isn't obvious out of the gate is that an environment

27:53runs the process, if that makes sense.

27:55So if we go to the app builder, what I understand to be happening is everything we've done so far

28:01is like part of a like a process.

28:03So when we when you hit run all to refresh the data, I'm assuming in this case, that parameter, that SQL, that is actually all running

28:12from that sequel and the sub the sort of downstream steps are the charts that we get.

28:16Whereas something like if I compare it to Tableau, that it that is sort of just it's not a process.

28:21It's more of a each thing is like a

28:23query so it's it's more more back and forth it's less of a package thing and there's actually a lot more that can happen outside of the asset compared to this where it's all contained in the asset which this is a different paradigm not saying one is better or the other I'm just I think it's an important distinction to make because

28:38It maybe makes this more of a sort of known entity in terms of what is this report doing to our infrastructure.

28:45It would all be sitting under that one process and therefore you could go back to Snowflake.

28:49look at your logs or look at everything that's coming off this and it's a little bit easier to say that whereas with something like a live connection it's it's hard to know without doing a lot of work to pass that out.

28:58Yeah.

28:59True.

28:59And I think to help kind of offset some of that, because there can be moments where you need to pull down quite a bit of data to do something.

29:08The other kind of advantage or I guess just

29:11thing to know about the way this is set up is in this environment, once you've executed or run a cell and it becomes an output, if it's a what we call a data frame output, it's stored in memory.

29:24And so then all the subsequent filters and queries hit against it are not hitting the database anymore.

29:30I see.

29:30Um That's really important actually.

29:33I guess platforms like Snowflake and Databricks have an element of caching and I don't I'm not sure customers always set those up in the best ways, but this is another

29:41sort of insurance policy in the product to make sure that you're a good citizen on your customers' infrastructure as it were.

29:47You don't want to be unnecessarily causing spikes in a customer's

29:50costs just because you haven't designed your product efficiently.

29:53Yeah, exactly.

29:55And that's all to say too, like that's you're locked into that.

29:57I think we do let you

29:59do what we call like query or SQL query cell that every time you run it it hits against the database.

30:07And there's trade-offs there too.

30:08There's, you know, mem in memory can you can get

30:11It pretty full if you've got a lot of data and you're running a bunch of cells concurrently.

30:16And we try to differentiate with the colors of the output.

30:19So To give you some signals, yeah.

30:21Yeah.

30:22So like the green is gonna be in memory.

30:24I believe like purple is

30:26um something that's hit against the database.

30:28So we'll give you some clues there, but definitely there's knobs and toggles that you can pull depending on trying to achieve and like how much you want to

30:36Lean on the database versus how much you want to be able to just do stuff in memory and make it feel really fast and snappy.

30:43Yeah.

30:44Amazing.

30:45And so, yeah, like I guess the next step is how do I share this?

30:48So I've built this.

30:49Yeah, what's that sort of final step look like and how does your platforms handle that?

30:54Yeah, okay, so we love this.

30:55This is beautiful.

30:56I'm actually gonna add my filter in here just because even though we

31:02Don't have it wired up to much of anything.

31:04I'm just gonna drag it just so I can give you a sense of like how that experience is.

31:08Yeah, yeah.

31:09Drag things

31:10Move them around.

31:11I could add tabs if I wanted to.

31:13Yes, I've seen that, yeah.

31:14So different pages as it were of the report.

31:16And then in order to get this ready for publish, not just get rid of that tab, I can set a few things

31:23So I can set the app theme.

31:25So let's say I want it to be dark or light, depending on maybe what I've designed it to be.

31:30And then the app runs settings.

31:32So this is also really nice because

31:35Maybe the data itself doesn't get refreshed but every day for some upstream process.

31:43I could come in here and set it to the workspace default or change it to some interval of time, meaning

31:51When you visit this, if it hasn't been rerun inside of that interval of time, go ahead and run it before you show it to that user.

31:58But what's really nice about this is if nobody looks at this, it doesn't get refreshed.

32:03Oh I see.

32:10Painfulness, yeah

32:11I'd love to know uh metrics from your customers, like how many refreshes have been set when no one ever went to them.

32:16I'd love to know that percentage.

32:18That'd be really painful for an analyst.

32:19Here's an exact number for the amount of work you do that no one cares about

32:24That is true, yes.

32:26But uh we'll we'll assume everybody's gonna be good corporate citizens and they're definitely gonna check it every hour.

32:32But this kind of lets you, I guess, just again have a bit of insurance on Yeah

32:36Yeah.

32:36So to play with that, if I said that's every sixty minutes, but no one looked at it every sixty minutes, when someone comes to it, let's say after four hours, would that be when the refresh actually happens?

32:46Like it waits for an interaction to then say, Okay, I should update this and then

32:51Over time it's just monitoring that trend and tapering up and down depending on that, yeah?

32:56Correct, yeah.

32:57Yeah, cool.

32:58Perfect

32:59Very nice.

33:08comes back as expected before I'm even allowed to publish it.

33:11And then we've also got a we're letting you do like a Git sort of PR check-in process where

33:16Yeah.

33:17Let's say I put this together, but my organization requires me to have somebody review it before it's published.

33:23Maybe I don't even have the ability to publish it.

33:25It has to go through an approval process or at least a review process of some other analyst on my team.

33:31Um if not, I can just simply hit publish app and then now it is available and live.

33:38So I'll switch my tabs really quickly.

33:40And now I'm looking at the live app.

33:43I could share it with somebody.

33:46I could share it with the entire organization.

33:49I could do some embedding options if you're like really keen on putting this inside of other surfaces and like letting the two more like an iframe experience.

33:58I could actually like add comments, which is pretty cool.

34:02So let's say somebody's looking at this and they realize, hey, this looks weird or this looks great.

34:06I could like add mention.

34:08somebody like Hunter on my team, say, hey Hunter, can you take a look here?

34:13This looks weird.

34:14And so you're getting that sort of layer of collaboration as well in this whole workflow.

34:21Amazing.

34:23And we've got agents everywhere.

34:24So I could have I have an agent now that can answer questions about this specific app or riff on it and go beyond what's on the page.

34:34So I I really think it can overstate like how much we believe that agents are gonna be the key to unlocking anyone's ability to ask data, ask questions about data.

34:45And so we've put them everywhere and been really thoughtful about

34:48where and how and why they would interact.

34:51Yeah.

34:52And I guess you maybe we'll come onto this after the demo, but this term app

34:57F I find this phrase very interesting.

34:58You actually wrote a little bit of a a blog about how apps are finally here, right?

35:03So it's I'd love to spend a bit of time just coming back to that after you after the demo because I think there is interesting nuance there and the framing of

35:10agents and AI is also sort of linked to that.

35:12There's something about workflows getting things done, action and all of that.

35:16So put a pin on it, but let's come back to it when you've done the demo.

35:20Cool.

35:21Yeah, so that's really the full end-to-end workflow from like an analyst looking to come in and do a deep dive.

35:26I think the other side of the coin is what we call threads, which is our more like conversational.

35:32So this would be like

35:33for a business user or honestly even an analyst to come in and just ask a question.

35:39It's like how are we trending?

35:43For our next quarter.

35:45Renewals.

35:46So the same question, but maybe from the less I want to do a deep dive to more of a business user

35:52asking it offhand and maybe not wanting to go hunt and peck for the dashboard that would probably give them this answer or find the email that was sent three weeks ago.

36:02They're just gonna come in here and ask.

36:04And so one thing we recently launched is agent memory.

36:08And so you see our little brain icon here.

36:11So this agent looks back at past conversations I've had with it to get a sense of who I am and what I care about

36:17Which is pretty cool.

36:18And so it could it realizes that I was just in the notebook doing something and it has all of that in its history and it's actually gonna go use that project that it just helped me create to answer this question.

36:32It's amazing.

36:33It's pretty cool.

36:34That's good again from a reusability perspective, right?

36:37Like sort of building incremental knowledge is actually way more powerful than starting afresh every time.

36:42That concept is so key.

36:44Yeah.

36:44Yeah.

36:45And so as you can see, like this experience is far more stripped down.

36:49We are writing everything in the hex is a kind of a notebook under the hood.

36:53So like this thread is actually a notebook, but

36:56Me as a business user, I don't care to know that.

36:58I don't need to see that.

37:00If I wanted to go deeply, I could see all of the different things it's looking at.

37:04And you see here, it actually didn't write any SQL at all.

37:07It just went to my project, it looked at three cells, and it pulled out the values from those cells.

37:14Nice.

37:14Um

37:15And then I could come in and ask it a follow-up question.

37:19You can see how it says, you want me to dig in to a specific AE.

37:23But if I hit continuous project.

37:25just to show not telling tales out of school.

37:28Like this is what it's building under the hood.

37:30Oh wow, yeah.

37:32And as my conversation goes, it's gonna be more and more filled out.

37:36Yeah.

37:36Cool.

37:36Does it send you this at the end of the conversation?

37:38As a summary or

37:40Yeah, how does it yeah.

37:42Yeah, I think now is a good time.

37:43Let's like move into like context studio because none of this works unless you've got a good eye on

37:51What conversations are happening, what context the agent's using.

37:55All that.

37:56So context studios are like observability 360 dashboard for how are agent conversations trending and what are people talking about

38:05And so you can see the trend over time.

38:08One thing that we do is like topic clustering.

38:11People are having a bunch of conversations using a bunch of different words.

38:15We have models running under the covers saying

38:18Let's cluster these into topics and then let's segment them based on how many of those topics are triggering warnings, meaning like the agent was confused or the user was confused or the context was just missing.

38:31And so this top one here, churn analysis is the number one topic and it's overwhelmingly causing issues.

38:37I could click into that to get like a nice little dashboard filter drill down moment.

38:43And I can see all the threads that are generated with this topic and then what warnings it's generating.

38:50So I could click into that thread, see all of the thinking, see what the agent actually built.

38:57get a summary of the conversation, the assets that it used, so all the data, and then the projects it referenced

39:05Which is also great.

39:06I can see a timeline of the entire thing.

39:08I can see all of the agent tool calls and thinking.

39:11And then most importantly, I can see the warnings.

39:14The workspace doesn't have period-based definitions for reporting churn.

39:18So the analysis mix current account stage with historical.

39:21And then it gives a suggestion of how to fix that, how to close that gap in context to improve.

39:27the quality of answers going forward.

39:31And in the old world, you may have had to like this was the end.

39:35You had to copy paste this or you had to transmordify it into what you think

39:40the right version of it might be in like a source of truth calculation or whatever.

39:46Now we have

39:48what's called the review agent that's actually going through and finding other examples of where users encountered a similar challenge

39:57And then we're suggesting a big change across a bunch of different surfaces of context.

40:04In this case, it's just guide updates to help close the gap.

40:09And what's super cool is I could edit this if I wanted to.

40:12It takes me into the workbench.

40:14And so now I can do whatever kind of different version of it I would want.

40:19Or I can come back here and just say, you know what, I I trust you.

40:24This is a draft is pending, so I can't exactly publish it from here.

40:28This is a our big demo instance.

40:30But I could just hit publish and it would Yeah.

40:32It's uh it's an interesting uh thing.

40:34We've heard tech companies talk a lot about hey, seventy percent of our code is written by AI and

40:38I've always wondered like when is this coming to the analytics space in this exact context?

40:43Hey, I've looked at your query, I've actually suggested these things and added these fields to your report.

40:48We're still gonna have to start we are going to have certain metrics for how much of the assets and stuff that people consume is been authored by AI uh rather than the analyst.

40:58Yeah.

40:59It's a bit of like um it's a good and a bad thing because I I I found that agents are probably the best at writing instructions for themselves.

41:09And so I think if you keep it in that vein of like most of the suggestions that are gonna come out of here are going to be suggestions for documentation to improve the agent's ability to use and like understand.

41:24what that piece of data means or what that metric is.

41:27So like for example, we'll also recommend like column description updates.

41:31So the existing one is pretty light, you know, the name of the AE.

41:36We recommend this.

41:38And this is gonna be a much better thing for the agent to reason about when it's trying to find the right data.

41:44I think it does a really good job there.

41:46For sure.

41:47Amazing.

41:48Yeah, I'm s slightly conscious of time.

41:50I know I know we only got you for a little bit, so um I I love the demo.

41:54I I really

41:55To me it's oh yeah.

41:57There's so many different ways of achieving the same goal and I think different workflows suit different teams and I could definitely see how the Hex workflow suits um

42:05I don't know if maybe this is an unfair thing to say, but would you say there's a certain type of customer that you see more often than not?

42:11I I uh you probably don't want to profile your customers either, but

42:13Yeah, is there is there a specific type of workflow or specific sort of makeup of team that you typically see heavily using a hex?

42:21Yeah.

42:22I think people that want to be really close to writing or at least creating the underlying data and the the logic

42:30And then visualizing it, but like making those two feel as connected as possible.

42:36And I also think that teams that have a really strong data science lean or

42:43think very deeply about using Python inside of their analytics workflows, really enjoy working in hex because the two can sit side by side.

42:53Side by side, yeah.

42:54But I think ultimately it's a team that's really I want to do a lot of different things and I don't want to have to switch taps.

43:00I think is really what it comes down to.

43:02If you live

43:04in primarily one tool and that's your kind of like singular focus.

43:08Like maybe you're not writing a ton of SQL, but you are building a lot of dashboards.

43:12I don't know if Hex is going to be as pleasant for you because we're not the we don't want to be the best dashboard tool.

43:18We want to be the best tool for data work in general.

43:21But if you are somebody who's writing SQL, then copy-pasting that, then uploading it, then writing a tableau level detail or a Power BI MDX.

43:32calculation and then you're doing it this way and then you're dragging and dropping and then you're if you're doing all that and you're hopping and switching and jumping you're gonna find hex is a really fun place to be because you're now you're just

43:45switching cells, you're not switching tabs.

43:49You get to move up and down rather than over here, over here, control Z, all of the things.

43:55Yeah, so really again, really focusing on a very specific workflow and that makes a lot of sense.

44:00And I I think I w I I don't know what you think about this.

44:05I feel like I feel like

44:07more people are moving in that direction.

44:09I I it's a strange thing to say because that sounds counterintuitive for going away from this very visual focused, tool focused place to this

44:20sort of closer to um use a hardware term closer to the metal as it were right yeah a little bit closer to where the work is actually done because actually

44:29Um the tools are now capable of making sure that we do things well in that space and they can sort of connect the dots a little bit more.

44:36So yeah

44:37Yeah.

44:38I had a couple of questions, so I I'll just list them out and then we can chat about them.

44:41So I was gonna talk about uh semantics and metrics and i if there's any sort of

44:47If Hex has a view as it relates to, you know, the the the general effort around open semantic interchange and how how you sort of support that.

44:55if I use metrics as an example, how you support that in both directions, so Hex ingesting that stuff, but also Hex pushing that stuff down.

45:02I don't know if that's possible.

45:04The other one is the SQL.

45:06So much SQL sort of being being being done

45:08Is the sequel writing really good?

45:10Like do you do generally get feedback from customers to say, hey, yeah, I love the sequel that Hex writes?

45:14Um and if not, sort of where are the areas they push back on?

45:17And then the very

45:18Last one is apps and AI and just digging into that a bit more.

45:21So a whole volley of questions.

45:23They're the only two we'll answer.

45:24But uh yeah, I'll let you start wherever you want.

45:27Yeah, I don't know.

45:27Which one do you want to pick off first?

45:29I'll let you choose.

45:30I I think the semantic one's important because I think that's where our philosophy differs from a lot of other players in the space.

45:38Because I think you can go two directions.

45:40You can lean very heavily on semantic models and you can say these are the thing that you need to have in order to have trusted results from AI.

45:49Yeah.

45:49And essentially you can't go outside these boundaries and it's a very on-rails experience.

45:56Or

45:57You can say yes, and you also need to be able to do more exploratory analysis, write novel sequel, but still make use of all of the different garbrails and context that a team could give you

46:10And Hex is very firmly in that the ladder.

46:12Like we think context and semantic

46:18isn't a model.

46:19Models are pieces of it.

46:20It's more like a context layer that's gonna be more important for an organization to let you go up and down from highly explor exploratory, I have no idea what this should look like.

46:32I'm gonna let my agent and I help and figure it out together to like highly governed this has to be the same number 100% of the time and it's

46:41It's delivered the exact same way every time.

46:44And so Hex has semantic models that you can sync or build in the product, but we also let you endorse data.

46:51We let you endorse projects.

46:53We let you hide things.

46:55We let you create the guides, so it's like the unstructured context.

46:58The warehouse metadata is also context

47:01We have workspace context.

47:03So it's like at the system level, what's the prompt for the agent for your entire organization?

47:08So it's a layered approach and it's like everything has its own job to do.

47:13And once you have all those things harmoniously doing their jobs really well, to the end user, you get to do anything.

47:20Like you can ask the weird novel question.

47:23And get a maybe not great answer, but our agents are also really good at caveating and being like, hey, I don't know if this is right.

47:31This is what I think it should be, but here's like my best guess.

47:34Because I didn't have enough data and oh by the way, now that we have context studio, I can go see that conversation and say, oh yeah, I'm that warning.

47:42Oh, that's really interesting.

47:43Maybe we should actually canonize that metric.

47:45Um all the way to was our error last year.

47:48And it hits the semantic model and it calculates it and it's fine.

47:53And so that that's where I feel like if we want AI on data

47:58to work at scale and to do the things it's promising, reducing the bottleneck, getting teams out of the business of being a ticket taking kind of

48:09Report factory, whatever term you want to use there.

48:12I struggle to think that if a if you're doing semantic models only

48:16How is that any different?

48:18Because the moment that an end user hits something and says, oh, the semantic model doesn't have this predefined, I I literally cannot ask this question.

48:26Now you're back in the queue, now you're back in the thing.

48:29And it it's the same story, just like different tune.

48:32Yeah.

48:32So that's why we feel like I I'm particularly excited at the direction we're going here.

48:36And then it's like

48:37the context from like past conversations and all this context is like wafting off of every analysis.

48:43Like how do we capture all that?

48:44So that's all the stuff that we're thinking deeply about.

48:47Amazing.

48:48SQL.

48:48How good is a sequel writing?

48:49Is it good?

48:50Concepts you worth it?

48:52Genuinely, it's not like it's not full of itself.

48:55It doesn't do a CTE when it's not necessary.

48:58It doesn't do crazy weird joins.

49:01It is probably the most consistently like

49:04reasonable sequel I've seen written by a an agent.

49:09Or a tool.

49:09Yeah.

49:10Yeah.

49:11And it doesn't pull in columns unless it truly needs them.

49:13And if it ends up needing them, then it does a good job of adding them.

49:17I think maybe it like it overindexes on aggregating probably more than I would like.

49:23Everything becomes a sum group by having situation.

49:27Yeah.

49:28And that's I think just to save on computer.

49:30I want to pull back all the rows.

49:32So I find myself having to like be really clear.

49:34Okay, I don't want this aggregated until

49:37a specific scale.

49:38Yeah.

49:38Because I want to use that and break down somewhere else or whatever.

49:42But outside of that, I haven't really had much

49:45to complain about.

49:47Good.

49:48Then very lastly, because I know we're nearly at time, what is an app?

49:52That's a big question to end it off, but if you could answer it in uh in as brief a possible way.

49:58How would you distinguish between an app and a dashboard?

50:00I guess that's a better question to answer.

50:02And if I could challenge you to go a bit more beyond this sort of this action conversation, like

50:07What is it explicitly about it that makes it an app and not just let's say a tool, if that makes sense?

50:13Yeah.

50:14I think a big thing is

50:16Because you can write Python in it, you can have it connect and do actions on your behalf a lot more easily.

50:23So I can look at analysis.

50:26Let's say I'm a salesperson and this is a dashboard for me to look at my forecasted deals or some account 360

50:33and I notice something is off in my Salesforce record.

50:37You could wire up that API via the Python to push an update into maybe right back to the database, maybe push into Salesforce itself.

50:48to fix something.

50:50And now that in that moment I have done the analysis and done the fix.

50:53Or maybe it's handy to actually send an email.

50:56Where I need to create a PowerPoint deck, you can wire all of that up with a push of a button and it's gonna run all of that code underneath the covers because once you get Python in there, like it's really endless.

51:07what you can see grammatically.

51:10And that's I think the big difference there is yeah you can filter, yeah, you got charts, so does everybody else.

51:16But can everybody else let you capabilities the action do the thing?

51:20Yeah

51:21Yeah, that that's a good answer.

51:22I'd never really centered that around the Python capability.

51:26Other platforms rely on write back a lot more, whereas Pythons can be a little bit more dynamic, more app-like, if I can say that.

51:32So yeah.

51:34Great answer.

51:34Appreciate it.

51:35Listen, we're out of time, exactly on time.

51:37So I really appreciate your time.

51:39Thank you so much for coming on.

51:40I'm gonna be playing more with hex.

51:42This conversation was

51:43like the start of a journey, so now it's my turn to get my hands dirty and hopefully teach people the wonders of what hex can do and maybe we'll have you back on at some point in the future.

51:52Yeah, to talk about some of the newer functionality that comes out over the course of the yeah.

51:56I'm sure you've you'll be shipping features way more often now with with the way that agents and AI work as well.

52:01So it'd be great to have you back on.

52:03I would be happy to.

52:04That'd be great.

52:05Amazing.

52:05Thank you very much.

52:06Take care.

52:07Thanks, Jen.