0:00All right. Hey, everybody. Super, super pumped to have you here with us with Mother Duck. We've got special guest Ryan Dolly here with us today. Ryan, say hi to the crew. What's up, everybody? How's it going? Awesome. Awesome. What are we talking about today? We're going to talk about, you know, how things are going with BI and the BI landscape and how AI changes things.
0:23It's kind of wild. You know, obviously, you know, from my background, I started my career on the accounting side. And so I kind of came into BI from that side of the world. You know, heavy Excel, Power BI didn't exist at the time. You know, I had the pleasure of using things like business objects. I'm sure, Ryan, you also did.
0:48For those of you who don't know, by the way, I actually worked for a BI company on the QA side as an intern when I was in college, a company called Melmo that was acquired by SAP. And the funny story there is when we launched the app, it was called Roambee on the iPhone, May of 2009.
1:11The feedback we heard from everybody was, how do I get this on my BlackBerry? So there's my, I've actually been working in this space for quite a long time, dating myself a little bit there. Ryan, what about you? Tell us about, tell us a little bit about yourself. You know, maybe not all the viewers know you.
1:31Yeah. Yeah. Yeah. Yeah. So I've been in I've been in BI like a similar amount of time. I think I got my start in in 2010. So and I was mostly a Cognos guy. I spent like a decade basically as as the Cognos guy or one of the Cognos guys. And in that time I did. So I worked on a BI team.
1:53I started there and then I took a brief stint at Oracle and then I went to a BI consulting firm where I eventually became a partner at the firm. And then kind of got sick of that and and moved into startups. So I worked at Account, which is kind of it's a cool tool.
2:15It's kind of like Miro meets a BI tool in a way. It's like like what if BI were an MMORPG sort of. Amazing. Yeah. And then and then I was VP of product strategy at GoodData for three years. And now I'm an independent advisor, analyst and author covering the BI space, working with companies like MotherDuck. So, yeah, I mean, I have similar like similar experience.
2:37I actually remember Roambee. I had a friend. I had a friend who worked there, actually. And so, yeah, I remember that. I remember that well, it was like mobile, mobile BI. Now, what's hilarious is that everybody wanted it on a BlackBerry. That's what's hilarious. That was the that was the.
2:58Yeah, that was definitely the through line. I mean, you know, we were working really hard on something really cool. At the time, I remember when I got my first iPhone, the popular app was the app where you would tilt it back in the beer. The angle on the beer would change. Yeah. Oh, yeah. So, yeah. So somehow we had the idea of what if we put charts on that? You know, it was wild.
3:19It was wild. We had we had a lot of fun, though. Down there in Solana Beach is where we were building that thing, which is just between San Diego and Carlsbad.
3:30Anyways, a lot of fun. So I think I'll set the stage a little bit for, you know, some context here, which is really just like. Opus four point five launched in November of last year.
3:50And what that brought along was what I've heard people say is like, well, basically, like solved front end. Right. You could basically say, hey, I want something that looks like this and react and it would build it. And pretty rapidly following on that, you know, we saw it get integrated into like Claude Web as this notion of like artifacts.
4:12And then, you know, Mother Duck obviously brought out. Dives and a lot of other companies have shipped similarly shaped products, which let you vibe code, you know, or prompt your way into into charts. And I think this is leading us into a crazy, crazy new world. And I know, Ryan, you've written a bit about that.
4:37And, you know, what's really cool about your experience is you're kind of connected into all of the all the leaders in the space. And so, like, tell me a little bit like what is the vibe? Like what are people thinking about, you know, both on like, hey, we're building the thing and then also we're buying the thing. Like what are where are people at?
4:56Well, that's that's the thing is is like leaders are all over the place, it seems like when I talk to people. So there's so there's some there's some common threads. I would say, first of all, it's it's a mixture of excitement and worry, which is probably doesn't just apply to be. You know, I think leaders across data and business in general are maybe feeling a mix of those two things.
5:18When it comes to be, I in particular, you know, I see kind of a two things happening. One is this this influx of kind of vibe coded front end stuff that the team is maybe producing some of it, but they're getting a lot of it flowing in from their their end users.
5:40So it's like the classic when I when I cut my teeth in this industry, and I'm sure it's still true today. You know, the main thing, like 50 percent of the projects I ever worked on started with an end user coming and saying, like, here's this Excel spreadsheet that I built that has become operationally critical to my department and I don't want to maintain it anymore. I want to get promoted. Yeah, exactly. I need a new job. This job is not my job anymore. Exactly.
6:04Exactly. So like, hey, B. I. team, like turn this into a B. I. thing for me. And I think that's happening now. But but the form factor has changed where it's it's HTML react apps that someone vibe coded in Claude or codecs or, you know, even even, dare I say, copilot. Yeah, sure. And and so you've got that going on on one side and then kind of the other side.
6:28What I'm seeing is pressure from the executive team and finance looking at the existing B. I. stack and saying, look, these tools cost a ton of money. And, you know, my the people who work for me are giving me incredible dashboards in HTML.
6:48Why are we paying for this tool? And so it's kind of the you know, it's like it's the convergence of those two things. Like the existing tooling is viewed as clunky and expensive. And why do we still need this? And I think there are reasons you could make arguments you can make for why you still need it or parts of it.
7:05But, you know, I'm sure we'll get into that. And then and then just this kind of excitement at what you can do now with with the models, but also real concern about, OK, how do we you know, how do we manage this? Like, you know, when everybody's got their own HTML dashboard on their desktop, like literally literally an HTML file,
7:27like saved to the desktop next to a picture of the kids. And like that, that's how we're running the business now. Like, how do we manage that as a team? Yeah, that's wild. I mean, I think I think Slack just, for example, added support for static HTML now inside of like embedded inside of Slack. So that just tells you where the momentum on this is going. Yeah, it's wild.
7:49It's wild. I think like, you know, I'll share personally, you know, on the mother duck side, when we built out dives, we were kind of like, all right, this is like long tail analytics. You know, this is for the really hard, annoying, thorny stuff that your BI tool is not flexible enough to solve. And then, you know, like for us, you know, that then put us in that exact spot that you're talking about, which is like,
8:09oh, actually, it's more convenient and easier for us to iterate on like the important things in the business in here than it is in the BI tool we're using. And so I think like it's also causing lots of churn, I think, in in existing BI tools, because people are like, all right, let's see if we can actually make this like have a go at this this new new frontier.
8:32Is that what you're hearing, too? Yeah, I am. I mean, I'm so it kind of part of it depends on the size and culture of an organization. Yeah, of course. Right. Like everything else. So anybody, you know, tech forward organizations and smaller organizations,
8:50I think are already making a serious run at can we just vibe code this and trying to solve, you know, they are running into some of the unseen things that BI tools do that people really don't acknowledge or appreciate. But those problems are increasingly solvable, I would say.
9:11And then larger organizations, you know, what I see is a mixture of of openness, like, hey, let's try it. And when I talk to a leader at a larger organization, I I don't tell them, hey, ditch whatever BI tool. And I say, like, pilot it, OK, just pilot to figure out because it can be different for everyone. Everyone's at a different maturity level and different technical capabilities.
9:34But just do a pilot on your team of like, how much of what we do today could we divert to more of a gentic vibe coded workflow and just be open to learn? Right. It might be hardly any. You might be shocked that it's most of what you do. Right. Just go into it with an open mind. And that's kind of what I'm seeing. So, yeah, that's what I encourage big companies to do, especially like don't don't don't be shy.
9:57Don't be scared. Give it a shot. Like, yeah, there's there's stuff. Yeah. You know, you have reports that are like sock to type to compliant reports. And OK, don't vibe code those yet. Right. Right. Exactly. But a lot of other stuff. Try it. Just try it and see how far you can get.
10:15You'll be surprised. Yeah, it's interesting. I think like, you know, the other side of this is that like it seems almost like B. I. teams over time became invisible to their execs. Yes. Like, can you can you help me like talk, talk, like explain that point a little bit?
10:35Yeah, I think that it's a weird it's a weird thing that has emerged in B. I. But because of the way B. I. evolved and and the way that especially like B. I. just became about dashboarding and visualization and that really became all anybody saw.
10:55And there's good reasons for that. Like, you know, hiding the complexity of what's going on under the hood from executives and end users is good. Right. You don't want them to have to feel the pain of whatever you went through to develop it as a user. That's good. But the end result of that is is that B. I. became synonymous in the minds of people consume it with with being just charts.
11:18And so when something came along that could generate charts much easily and more cheaply, it immediately engendered in the minds of executives and consumers like, oh, well, we can just replace that thing. Right. Because it's just charts. It's just charts. And Claude can give me charts, then we can just replace that thing.
11:38And it and it really started to highlight the cost of procuring and maintaining these tools in the minds of of decision makers and people who control budgets, you know, where suddenly they're just thinking, I mean, B. I. tools cost a ton of money.
11:56And and if you've they my experience of B. I. tools is when you buy one, it tends to be a pretty good deal because they're trying to win your business. Yeah. And then starting sometime year three, when you're just a renewal stream, like you get turned over to a different team who manages you and suddenly things change.
12:16Right. And so if you've been on a B. I. tool for 10 plus years, it probably costs an enormous amount of money at this point, especially if you're a large company. And so, you know, so I think that that's kind of what's happened. The minds of executives, it's just charts. And and and so the value of what's happening under the hood,
12:39especially when you get away from anything related to like regulated data sets, you know, maybe the value of that in a B. I. tools a little more clear because of auditability requirements and that sort of thing. But when you get to pure analytics like, hey, I've got a question and I just need it answered, you know,
13:00the fact that the B. I. tool is just charts makes it look pretty low value in 2026. Yeah. Yeah, yeah, yeah. Right. Right. Right. I think and I think, like, you know, the hard part is, like you said, there's and there's a comment in here, you know, like charts are just the tip of the iceberg in terms of what the value is. But like also a lot of those things have been abstracted, you know, away from the user.
13:21They don't even see it like. Yep. So I think I think this is a really, really interesting, you know, the actual complexity inside these platforms is quite high. Right. Yeah. You know, I think like when I think about what's actually inside of a B. I. tool, right, is it's often like a cache, right?
13:41That might be a full OLAP database in the case of like Tableau. Right. They bought hyper, for example. It might there might not be one depending on the tool. You know, some ones that are built on Snowflake now don't really have a cache, although I think they're they've all built one kind of as a result. But that's more like a result cache versus like a database.
13:58Right. There's some sort of notion of a semantic layer. Right. There's this notion of like security and governance, like who can see what, when, you know, and then there's charts and then there's exporting those charts. Right. And so like but but the thing that users mostly interact with is like, you know, either the exports or the charts, you know, kind of the dashboarding part themselves. So they don't see all those other pieces, which are quite hard.
14:22And like no one wants to think about those, you know, frankly, very, very challenging, especially when you can get a better chart, a better narrative more quickly, you know, almost instantly with A. I. Right. Yeah. And so, you know, I think this notion of like, oh, like A. I. , you know, can just replace this for cheap.
14:42Like, is this is this correct? What's your opinion? No, no, it's not. I mean, I think what I can do today that that I like the workflow that I really would never pass through a B. I. tool, honestly, is kind of the the ad hoc decision making like the hey, I was wondering X, you know,
15:06that's just so much faster and better and more efficient through something like Claude. And of course, there are dangers there. Right. There's the danger of hallucination. And then there's the danger of, you know, you not understanding the data or having access to the right data. But I would I would argue that there's ways to solve that, you know, context like data, I mean, tools like Mother Duck.
15:28OK. And then having good context management and that sort of stuff can make that a much, I think, better way for people to do that one off or even like small department reporting. You know, anything that doesn't have to leave your little group kind of begs the question, OK, why use a B. I. tool at all for it right now?
15:48When you get outside of that, you know, and people are talking about in the comments here, you know, all the stuff someone someone had a list of all the stuff, you know, caching, data performance, real label security, audit, cloud front end, CSS. I mean, there's so much stuff like you just think about, OK, we semantically are in a B. I. tool.
16:06Yeah. Right. Like how do we guarantee that someone in Department X and someone in Department Y and someone in Department Z uses the same metric and gets the same answer 100 percent of the time when they ask a question? Right. And that that's right. That is not a solved problem today when everyone's using Claude. There are obviously practices you can put into place to help.
16:29But but I guess my question would be, well, you know, how far away are we from that being a solved problem? Yeah. And I think a lot of people are working on it. Like you guys included. Right. If you look at what dives do, you know, dives.
16:46I mean, I you know, it's I work with you guys, but also I I run my solo consultancy and media brand on Mother Duck and Dives. Yeah. And I as a guy for what I need would live. I will never use a drag and drop B. I. tool again for for what I need. I feel quite confident about that. I love that. So never.
17:10So first off, yes, I don't ever want to use a drag and drop B. I. tool ever again. Like I think I think one thing that that did end up happening with these B. I. tools is that they they got everyone got locked into these U. S. that were very inflexible for power users. Right.
17:27And I think like one thing that that we maybe went too far on the pendulum song too far on, I think, was like, oh, you have this lockdown, you can only do this. And it turns out like some of the best charts that you want to build, you know, require tuning very specific knobs in Matplotlib or something. Right. Right. So, yeah, I think it's definitely funny to kind of see to see how these things are going.
17:51I think it's really interesting to talk about like what like what things need to happen to like what primitives need to exist inside of a tool for it to be able to interact, you know, for you to for you to discard the the drag and drop interface. Right. Like what what what pieces need to fit in there?
18:11Yeah, that's a great question, I think. So do you take the drag and drop interface aside? What you need, like the first thing you need is some sort of content management system. Right. We still need a place to put the objects where we can people can find them. We can version them. You know, they're discoverable, all that sort of stuff.
18:34And that's in many ways the first and most obvious problem of vibe coding, you know, HTML, JS dashboards. It's like, well, where are we going to put them? And I think this is actually an underappreciated problem in BI because BI really has two roles. Role one is to answer ad hoc questions and help with like point in time decision making.
18:57But rule number two is, OK, we've made a decision. Now we need to agree on what metrics will tell us whether we're succeeding or failing. And then we need to agree on some representation of those metrics that we all look at over time. You know, the thing you put on the wall that you can argue about in a meeting. Right. That thing needs to be standardized.
19:20It needs to exist somewhere where everyone who has a stake in it can see it. And that is that's a part of BI that I think everybody just has taken for granted forever. That suddenly is a problem again, because, you know, because of the just huge proliferation of dashboards and that sort of stuff.
19:40So that's part that's number one. And then you get into, OK, there is this security thing. We need to keep track of who can see what. And it gets quite complicated in BI tools. It's almost like as someone who's worked for BI vendors, there's an infinite, an infinite. degree of security finer and finer ways. Someone out there
20:04wants to slice the security model. Yeah. Yeah. Um and and so that can get quite complicated to manage um, but I think the the the then the fundamental thing is if I ask a question and you ask a question. Ideally, it gives the same answer every time the same input renders the same output right. That's right. That's right. This item potency thing
20:26and uh and that is not an easy thing to deliver um when you you know, like BI tools deliver that today and and so if we wanna slice off the drag and drop interface and we say, hey, we're just gonna vibe code this. I'm totally in favor of that. I think because it's just as a builder, it's a vastly better experience than clicking and dragging. I
20:49cannot believe what I can accomplish in a short amount of time compared to when I was using old BI tools, but you know, there's all this other stuff we gotta um we've gotta solve that um that just chopping off the the head of the BI tool doesn't solve. Yeah. I mean, I think like you know, what was always interesting to me is like the highest calling for a BI report in my experience was like, okay, we screenshot this and put it in the board deck which breaks
21:14the entire lineage of the entire thing in the first place. Yes, right. So, it's like now, now, we're having the discussion, you know, on the board or you know, or in the board meeting about something and it's like, well, hang on. Is this right? It's like, okay, well, now, now, we've gone full circle uh on this on this uh on this debate. Although, I
21:31do think it's very um interesting to see um uh how uh you know, because a lot of like what I'm hearing from teams is basically like they're getting they're getting um in in previous um like in the previous world, it took way too long to get data into like their let's say quarterly quarterly WBR like uh or like quarterly, you know, business
21:55review meeting for their marketing team or whatever. It took way too long to get the specific things. They always did the the old ways on the things that were kind of the prescribed path and now with AI, it's like they're getting all this stuff but everything is conflicting um conflicting in the deck and all the conversation is kind of like reverting back to wait, how did you get that? Cuz I got it this way or like even worse, you know, people are using these charts to like beat each other with a bludgeon. Um I don't
22:19know if you're if you're hearing folks either in that world or like concerned about that. Oh yeah. Yeah. I hear that a lot. I mean there's a lot of concern with that right so it's um like the one of the goals of BI has always been we walk into the meeting with some shared understanding of what's happening in the business right based on data and um and so
22:42there's a pendulum that swings in BI and and has been swinging since before my career started between okay like that's the real problem. The real problem is conflicting numbers and people arguing my charts right now my charts right so we got like that's the problem we gotta solve and then what ends up happening is we develop these pretty rigid systems and these rigid methodologies and then people can't get the data
23:06they need in time to do anything with it so then we swing the other way and it's like it's all about enabling decision making fast decision making uh but then we create you know we we feed the fire of my charts right now my charts right and so um what AI has done is it it has like taken that that fire of whose chart is right and turned it into
23:27like an epic inferno. Yeah. I live in the Detroit area and 2 weeks ago like we couldn't go outside for 3 days because there was so much wildfire smoke from Canada, you know and and I feel like it's in some organizations. It's that level of inferno right of of whose numbers are right um and so uh I mean it's a real problem to solve and and it's like people are having a very active debate in the comments right now about
23:52like different aspects of this, which I love to see um but but that's that's kind of where we are and and we're really in like uh I would say the exciting thing about all this is. We're in a if you have an idea, you can try it phase right. That's right. That's right. Like these problems are not solved. They will be solved and there's been a lot of
24:15in solving them and so you know if you're listening to this, I would say identify those problems within your organization and take a run at at how they work for for your organization within your culture and your data stack because that like there are promotions to be had right by solving these problems and new products to be built. That's right to solve these problems and and so that's the exciting
24:38thing about it. Yeah II totally agree. I think um you know I think one thing that that everyone everyone who's who's adopting AI this way is speed running is like you would only you know it took you. It took you maybe six or 12 months in the old world to get to the point where someone would have a conflict in metric definitions because it was like alright. we gotta get all the data in the data warehouse. We gotta stand up the BI tool. We gotta migrate the existing
25:01charts. We gotta do all this stuff and then we can be like. Oh now these two don't match each other right um that took a long time at least you know when I was working on this stuff uh you know six or 7 years ago now um we never really got to that place, but AI just lets us speed run all of that right um so I think like uh you know everyone's showing up with a
25:23different number. It is a real problem and I think you you you had there was a great comment you made Ryan on LinkedIn about like defining words for the book you're working on and one of those words was like metrics and I think the thing the thing that you had said about it was like um uh well, I'll I'll let you say like how would you define metrics? Yeah. Yeah. I mean II think um I'm trying to remember
25:46the exact words I used uh but but in my mind, there's there's two elements to a metric one is there's like uh like a logical mathematical element to a metric that is literally how is it mathematically defined so that the calculation renders the correct result and that is I would say about 95% of what people currently put their effort into, but it's not 95%
26:10of the value or purpose of a metric. The other the other part of that is like a metric is really the difference between a metric and a calculation is a calculation is just some math you run to get a number. A metric is really a social contract between data producers and data consumers about what
26:31about what is important to track within this organization and why and what should we do about it right like the best metrics combine all of that stuff. Yes, it's mathematically accurate and it has lineage and it's traceable and tracked over time. Yes, all of that, but the more important part is why does
26:51this number matter to whom and what do we do when it changes in positive or negative ways right and so that's right that element of metrics has always it's never been codified. It kind of exists ephemerally in in like the social arrangements of an organization. That's right and and one of the things that I argue really strongly for in my
27:15book is like agents don't know any of that stuff and so you actually and they can't right. They're not human beings and so you actually have to write it down. You know if you're trying to build some agentic BI system in your metric definition, it can't just be the math. It also has to be like why does this matter to whom and what will they do with it and and that needs to be included in your metric definition so that an
27:38agentic system can pull that metric definition and actually be able to properly reason on it and do the right things with it the same way a human being would. It's just you don't think about that stuff. Nobody in your organization thinks about that. It's it's like implicit knowledge a lot of the time. Yeah and and and and then you know when it's not implicit knowledge or or you and I realize our assumptions about what this metric means sociologically are different.
28:02Yeah. Well, then we argue about it and we come to some conclusion right, but the machines are not going to do that for us. so we need to give it to him. Yeah. That's that's totally right. I think you know one one thing that I think we sometimes get lost on you know as technical folks is like the data we have is a shadow of the physical world. Yes, right and we need to we need to realize that the way we define these
28:27metrics is constrained by the fact that we're interacting with the shadow, not the real thing right and I think like that is a you know it's easy to get lost in the sauce on like how to define a specific metric in a SQL definition right, but the the reality is like well, does this matter? This is number matter. What will we do when it changes? I think those
28:49are the important things and I think you know it's easy for us to you know descope our focus a little bit and drill into that. Yeah, very very interesting. Yeah. I mean, I think it's there's people in the comments talking about product management and and actually chapter two of the book is a chapter two or chapter one. Yeah, because I
29:13think chapter one one of the two is is titled product management for data people. Okay. I love it. I love it and so like II think really the solution to these problems for data teams is to embrace the product management mindset and and philosophies that come out of product management and data product management and I don't advocate strongly for certain
29:37standards or data contracts. I got friends whose whole careers are about data contract standards right like I don't that's all important. It's good. You know my opinion is kind of yes do that stuff cool right, but but it's the mindset. It's the way you think about managing data, especially as as we see like these models just keep getting better and better. so you know more of the job will be about
30:01really understanding the what's the word invented in the data mesh book socio technical. Maybe she didn't invent it socio technical. I first encountered it in in her book. you know like that understanding that becomes a bigger part of your job. I totally agree. I totally agree with that. I think. You know kind of what I heard from a
30:26from a customer mother duck, you know, maybe a few months ago, they were like maybe there's like less kind of like BI analysts and more like they use the term like context analyst. Yeah, like I really think that like that is a lot of the work is shifting towards. Understanding all of the things around the thing because now we can just define the specific thing with AI so much faster. Yeah, I agree. In
30:50fact, that's I was a real struggle. I've been working on this book now for. Coming up on a year and and honestly everything I wrote in about the first 8 months, I ended up throwing out because of Opus four five and and and and the book really changed to be about context management. So so like II turned in the chapter on data pipeline building data
31:14pipelines with AI and last week and and really the shape of the chapter is like here are the key decision points you're going to have to make like the technical things you might get hung up on right and then here's how to manage the context around those around data pipelines in general and and the book really doesn't have a lot of prescriptive like here's the prompt you type in to get this output because that
31:38kind of became useless. That was that was like a really good thing for me to teach you a year ago. Yeah, but it's not anymore right. It's because because the models got so much better. Yeah. I think yeah definitely hard to write a book. I think right now things are moving so fast. You know you have to you have to write fast and be willing to throw things
32:00away. I think and I think the same is actually true. I saw I saw a really interesting tweet from Charlie Marsh who's now you know working at open AI and he's like sometimes you just have to be build. You just have to build the thing that's going to get you know it's going to get thrown away in a month because it needs to get built right now. Yeah and so like on the flip side actually on like the BI side we get to do that sometimes like you know what alright. I know that like some
32:23other team is going to build this into the central thing, but I need an answer right now and I can spend some tokens on this. It's all good. I'm totally in favor of that. I think that I I think that you know what I go back 15 years. I in ye olden times, I designed like we had a very rigid SDLC at the place I was working. Yeah and then I designed like a quick access SDLC because we were taking people who said, hey, I just
32:48need one more chart on this dashboard and we were putting them through the regular like the eight month SDLC process. It was like what are we doing here you know and the nice thing so the great thing about about AI is like when you when you kind of when you develop your meta BI thinking. Yeah, you know you can take that thing that you just built. Hey, we just we just needed this thing and we built it and we
33:13know we're going to throw it away in a month, but maybe there was something in there that would be good to plug into the overall like designed workflow the the designed BI environment and and it's actually much easier to do now. You know you can literally ask Claude like hey was there anything in this that I should like think about putting into my my global metrics layer and it'll be like. Oh yeah, actually this one this one metric is actually might be
33:37kind of useful. Go ask someone if you know and it's it helps you enormously to bridge that gap between hey, we just fired this thing out cuz we needed it now and and I'll actually maybe there's something from this that yeah, we don't need to keep the whole thing, but there's one component of it that actually would be worth working into our overall BI environment and it just makes that path easier. Yeah,
33:59totally totally totally. you know what's interesting is is so you've actually been building a lot as you as you mentioned, you know for your internal internal for your company and then kind of examples understanding the tools you know with Claude you know and and something like mother duck. What has been you know I know you built that briefing book, which I thought was really cool. What kind of
34:22surprised you about working through that now you know kind of building building BI in the frame from the framework of someone who like you is you know has a lot of experience in the space. Yeah. The first thing that stood out to me was just how shockingly easy it was to get to go from nothing to
34:45you know data like in my case, I built this briefing book so so you'll have the context that it's an example of kind of a classic BI report built in a mother duck dive. So you know most dives justifiably are really flashy visual cool looking things. I built a dive that is like doom running in a dive right so so it's it's all sorts of stuff like that, but
35:06but I know from my experience that a lot of what BI does is still financial style reporting and regulatory style reporting and so I set out to say okay instead of building something really flashy. Let me recreate what I spent a lot of my career building, which were these more table of numbers regulatory style reports. So what I did is in about a day day and a half
35:30maybe I went from nothing to extracting from the Yahoo Finance. There's a Python library that that just extracts data from Yahoo Finance so great library. Yep. so using that building a pipeline to do the extraction to land it in mother duck to then create some dives on top of it that that
35:52are like financial statement reports and then from that to automate the distribution of those financial statement reports over PDF, which is something I was asked to do a lot with this style of reports over the course of my career and so the first thing was like I did that in a day and this was literally a quarter long project at the beginning of my career or longer and and maybe before AI we had had advancements in tooling and that sort of stuff that okay.
36:17It wouldn't be a six month project. It'd be a four month project right like sure sure, but now it was a day and and look in a day. Did I do everything you would need to do in a true regulated environment? Of course not, but I did probably 80% of it in a day so that that was the first thing that really stood out. It's just yeah. I mean, it's
36:37just insanely easy to do. I was I will say I was comforted like there's people who are oh well BI analysts even exist in the future. I mean I was comforted by the fact that my experience really did matter in in building this like yeah, my ability to point out bad patterns or or know the words
36:58to use the things to request you know. To be able to suggest to the AI no, you know the way you've designed this like identity management to to manage the automated PDF distribution. Not the way I would have done it right like that still mattered a lot and I don't think I would have gotten to the outcome I had without that stuff. so I you know, I don't I don't think your CEO is going to be able to
37:23vibe code what I built so and really I mean in my mind, it was like all wins. It was my experience my taste. It all still mattered, but I was able to do what used to take me months in in an afternoon. Yeah, that's right. That's right and I think like you know, and I think the hard part and I think a lot of people have talked about this right is basically like okay. We have these really powerful tools,
37:45but like how do we help? This is not really a question we need to get into too much, but like I'm definitely thinking about like how do we help junior people get those skills too? Yeah right like we were in the salt mines man like like it sucked. Yeah, I would never wish that on anyone, but like you know you do get like those the really good experience
38:07building those things for you know for people who have really high standards and you know as a result, I'm very very good at using an Excel spreadsheet these days, but also on the flip side, how do we help everybody you know up level is really really interesting. Yeah. I don't want to get stuck too much on that. I guess like maybe read your book. I guess
38:30it's yeah. Yeah. Yeah. Shameless plug. There's a book coming out that will help you help you with that. We'll talk about we'll let you plug it at the end. So like one thing that so like is your is your belief that like in maybe like small and midsize companies that they're they're pretty much going to start ripping out some of these older tools that
38:53don't have the right affordances and start kind of being more AI build in this specific space. Yeah, I do think so. I think I think obviously it will start with you know SAS companies and and people like that because there isn't a degree of engineering rigor. I think you need to have to make this work like if if you're a small midsize company
39:15and you know you're a SAS company and you are like everybody on your data team already knows how to use get okay like go go now right. You can do it now if you're someone out here in the Detroit area and you're a midsize tool and die manufacturer, you know, maybe you don't there's a little bit you need to learn of just engineering principles
39:38and and that sort of stuff. Yeah, first but AI is a fabulous teacher. So I think you can learn it pretty quickly before you get rolling and and I think yeah, I mean II honestly don't see why in for most of what you're doing today you would not choose to go down the.
39:59of vibe coded path and, and, and a tool like mother duck provides a lot of the infrastructure that you need to make it happen so that you're not just emailing HTML files to one another, right? Like that, that's, that's the world you can't find yourself in. Uh, and to the greatest degree possible. And, and there is tooling out there that will help with that.
40:21Right. Um, and so, yes, I mean, my, my, my thing is like, yes, do it now. Yeah. Yeah. Okay. Now just, yeah, I think it'll be really interesting to see how far these things can push, um, uh, over the next, uh, over the next few, few years for sure. I mean, even the next few months,
40:40it's wild. Um, I think here's what I want to do next. Let's, um, any, anything you want to hit on in summary here, Ryan, and then let's, um, let's, let's jump into the comments here. There's a bunch of really good conversation and questions that I want to hit on. Um, so yeah, any, any parting shots here? No, I mean, I mean, I think like the, the big thing is if you haven't,
41:01if you haven't played around with this yet, like you need to immediately today, um, you, what you will find is that it's really easy to upload a spreadsheet to something like Claude or connect Claude to MotherDuck and just fire out a pretty good looking dashboard. And, and if you're
41:23as a data person, especially if you know the data well, you will immediately find things in that output that you're like, okay, it got this wrong. It got this wrong. It made this assumption about a metric that was incorrect. Like that's all true, right? I've talked to a lot of people who have that experience and then their conclusion is, is like, okay, and this is why we can't do this. We have to go back to the old way of doing things. That is the wrong conclusion. I would,
41:46I would try it, find all of those problems and then start to think to yourself about, okay, how would we solve these problems within my organization? Right. And then like that needs to be the path that you go down because I think the gains to be had from this way of working are so huge that, uh, that you have to embrace it and you have to be, yes, apply skepticism to
42:10the process, but skepticism in service of problem solving, not shutting it down. Yeah. Um, I think that's a really nice transition into, into the questions here. Uh, I think, so Shane just asked, you know, what would you recommend data team does tomorrow? Give a new AI for BI world. I think you just answered that very, very cleanly. Yeah. Yeah. Yeah. And just be open to it, right? I mean, this is, this is a really exciting time
42:31to be building in this space. And so embrace the excitement, recognize that the objections you have in your head, like, Oh, how are we going to ensure metric consistency? And, you know, LLMs hallucinate. I mean, the way I recommend you solve that in my book is, well, to the degree you can, don't let them write, like, don't let the context window calculate stuff, say write Python and then, and then save that Python to disc.
42:54So the next time you come in, you fire up a new session. It reads that Python from the last session. Right. Um, that that's the way to approach this sort of stuff. Yeah, I totally agree. I mean, I think, you know, a lot of it is like we use, I think a lot about, um, okay. So, so a little context for me. So I used to run a ERP team at
43:13a public company and, um, not everything fit into that system, right. As hard as, as hard as the ERP software provider wanted us to get everything in there, not everything fit in. And the BI tool was a critical and the analytics was actually a real, a real critical part of what we were building, right. Because we didn't perfectly fit into that. You know, we were
43:34not a commodity trader that fit perfectly into that. Um, and so, you know, I, I think one thing that I kind of think about here is, um, what is it like, uh, how do we fit these pieces in together and like, let the, let the magic of the company work better. Right. Not everything
43:56fits in these like beautiful prescribed workflows. If it did, you know, every company would be the same and they're not right. Um, and so how do we, how do we let the magic still happen? I think AI is, is, is going to let us get there. Um, speaking of which, here is a really, uh, good, good question from YouTube. Um, will BI analyst roles and similar roles go the way of the graphic designer type roles did five to 10 years ago? Um, what do you think?
44:21Uh, I don't know is the easiest answer. Uh, I don't know. I think what I always, my, my stock answer when people ask me questions like this is I think that people overestimate short-term change and underestimate long-term change. So when someone says, oh, agents are going to be doing all data work in a year and a half,
44:43there won't be any data analysts. There won't be any BI analysts. I'm quite skeptical of that. Yep. I agree. When you, when you project out five years, 10 years, 20 years, who knows? Right. Um, who knows? I think that the role is going to change a lot though. I do think that the, I'm an expert at the complicated BI interface. Like the value I bring to the company
45:07is, is the BI interface is so complex that only one and a half percent of people can even figure out how to use it properly. And I'm in that one and a half percent. And so that's the value I bring like that, that I I'm, I'm quite certain is, is rapidly diminishing. Yeah. Um, yeah, I think I agree. Uh, great, great comment here from Frank, by the way,
45:30um, that he just threw up on the screen here, which is, you know, uh, BI analytics used to be able to get away with just making visuals, organizing KPIs and some light commentary. Stakeholders not doing this themselves, right? Uh, role has to evolve. Um, how, how it will evolve, I think, you know, subject to some debate. Yes. Yeah. I think, and then that's where the product thing comes into, in, into it, in my opinion is like, if you start looking at your data,
45:55your BI and your overall data practice from a product perspective, and then as your job goes from, I'm the, I'm the drag and drop guru to I'm kind of the person who, who best understands both the business and the data. And I'm in a position to kind of manage our data, our BI like a product.
46:16And I'm more of a product manager and, you know, maybe an agent is more like the developer and I'm the product manager who helps the agent build the right thing. I think that that's, there's an opportunity for, to contribute a lot of value there. Totally agree. A great question here. I'm going to pull up, um, from Ollie. Uh, do you think visualization tools like Power BI will go
46:37away? I cannot believe some complex logic, like how DAX implements queries would go away though. Yeah. Um, I don't know. I think, I mean, I, I think that, look, if you, let's just take Power BI as an example. I would say that Microsoft is already in, in the process of preparing for the Power BI user interface to be radically deprioritized. They, you know, they recently
47:01switched over from the PBIX binary file to this new, I can't remember what it is, PBIJ or PBIR or something. Someone in the comments will know. Right. But, but it went from like this, this binary report definition to a well-formed JSON report specification. In my mind, that is actually, it's super overdue, but it's also step one PBIP. Thank you, Adam. Uh, the PBIP file is
47:26the first step in saying, you know what, maybe we're going to let an agent build this instead of building it through the drag and drop interface. So does that mean Power BI goes away? Like, no, but it, it does mean that the drag and drop interface gets deprioritized in favor of more of a AI-led agentic workflow. Um, when it comes to like, okay, DAX, like complex DAX, um,
47:50I don't know. I mean, I, like, I don't know. I can't, I can't tell you how good Copilot or Cloud are writing DAX because I don't do it, but I know they're pretty damn good at writing SQL. Yeah. I, what I would say about semantic layer languages is the problem they have is there's a lot of them and they are all very different in their implementation. And the advantage SQL has is we have 50 years of, of training data and models are just really
48:15good at writing SQL. Um, what will it look like? You know, do we, do we need some sort of semantic layer? Like, I think you need some sort of way to make sure that programmatically you can always get the same answer, right? Ryan, you said, Hey, like, you know, put that in Python, right. Or TypeScript I saw in the comments. Yep. Right. Um, you know, I think there is, uh, there's definitely
48:35going to be a need for that. And you may not be able to express that simply in Markdown and have it reliably retrieve. Right. Um, so I don't think that need goes away. Right. Because we want the conversations when we're talking about the charts, like the most important conversation is what do we do about these numbers changing? Not like debating the authenticity of the authenticity of the number. Right. And like, I think a lot of companies are like recircling that debate where
49:00like they're going into all to like their, their calls. I've talked to a few friends that companies that are, that are becoming more AI pilled and they're like, Oh my God, everyone is, you know, shaping the narrative to match their own thing. It's very chaotic. Like, uh, and they're like, but like, there's not a clear path on that. Um, they have so many different tools, you know, not everything's integrated in the data warehouse, all that stuff. So, um, there's definitely going
49:22to be a pathway there. I think that that we, we know that we need a way for, um, uh, to get, to get those answers reliably. Um, what that would look like, you know, I, I'm not super convinced on form at the moment. Um, but we'll see where it goes. Uh, I think here's a good one from Dave, uh, for companies that have data ops, data science, data viz and data analyst groups. So we're talking to like pretty big companies who will be responsible for the product context
49:47management. This is such a good question. Yeah, I, I don't, um, so in my mind, actually the, the closer you are to the business, the better seat you're in to be the person who's responsible for this, because ultimately the product and context management is about capturing
50:09the business and cultural realities of the organization and rendering it in a way that it can help the AI develop analytics and answer people's questions. So, so in my mind, that's the data analysts, right? They're the closest to it of all these different groups. Um, you know, these other groups, uh, like data ops, I mean, that, that becomes,
50:31you know, I would say the data analysts are like the producers in this world and the product managers. You know, if you're in data ops, I mean, there's a lot to context management. Frankly, we don't even know how to properly do it right now. I mean, like, like I manage it personally. I have a directory of, of Markdown files. That's what I do. And it works really well for me. And in the book, I walked through a lot about how to do that. And then I have a
50:54chapter like, okay, you did it in Markdown files. Does that really work at an enterprise scale? You know, maybe not. And then here's some options for what you can do to scale it out, right. For more deterministic, more deterministic systems and stuff like that. Um, you know, I, but I think the ultimate answer is like, this is what I just told you is my hypothesis, right? I, I think I,
51:17we don't really know yet. And, and we don't know what of these roles are going to exist and how they're going to exist in the future. Um, but I, my, I think the people who are best positioned are those who are the closest to the business and have the most experience translating the business into the data. Uh, that's the best seat to be in.
51:39Yeah. I think, you know, my perspective on this question, I think is that, um, uh, explicitly, I would expect that we will have more kind of data platform teams. And then a lot of things that were maybe previously centralized will, will decentralize back into domain teams. Right. So your marketing team may have, um, you know, their own data scientists, whereas in the
52:00past, maybe that was a shared resource, um, or, you know, whatever. I, I think like, you know, in those cases, the challenge with decentralization, of course, is like, and especially with AI, you can move so fast. How do you keep everyone aligned? I don't think anyone's actually figured this, figured this out. I mean, GitHub is basically falling over for this core reason, which is like, we don't have the right abstractions for collaborating at AI scale.
52:24Um, uh, it's so hard, like, you know, I'll speak for myself personally, but like, you know, even working inside of, you know, the marketing team and the DevRel team here at MotherDuck, I can have a conversation with someone. They say, yeah, okay, I'm going to go off and do that thing. And the next day we, we talk and it is, you know, it is complete and it is done in a way that I would have never anticipated. And it is so far down the road. Um, it, it is wild how fast these
52:48things go. And so I think we, how do you, you know, how do you collaborate in this world? I think is really, really interesting because, uh, there's a pressure. There's, there's pressure, I think, on both sides. Like we need to collaborate better, but also domain teams can run way faster. And so how do you tie those out? Yeah. Um, all right. Great question here from Juan. I'm going to pull in here. Will standard reports like financial reporting coexist with AI driven
53:09analytics? Uh, yes. Yeah. I mean, people have been telling me that standard reports are going to go away for literally my entire career and, and they're not going to go away. So, so let me give you a scenario. Uh, I am the CFO and I want to check the current state of
53:28our core four financial reports. Do I want to wake up in the morning, log into my laptop, go to a chat bot and say, you know, what was this metric? What was this metric? What was this metric? And then do it again tomorrow or do I just want to look at the statement? You know? Um, I think
53:49that I think AI will build these things, right? AI will build the standard reports. I actually anticipate a future where there are more standard reports because they're easier to build, right? The barrier to building them is lower. Um, so in situations where maybe you wanted one in the past, but it's like, you know, we're just going to get by with this spreadsheet because it's too hard to
54:09build the standard report. Well, now it's not. Yeah, I think that's right. Um, and I think, uh, um, what do I really think about this? Yeah. I mean, I think like the most important thing is that we, the social contract is intact, right? Like we are having a conversation about what we
54:30do about the thing, not if the thing is right. Um, and I think, you know, that is a very durable principle. Um, we'll see, we will see how things go, but I think, you know, uh, the, the pattern that I've seen evolving is you use AI to build a deterministic thing, right? You use the AI, which is non-deterministic to ask a question, to do some archeology, right? To discover the thing
54:54that is unique about your business. And then you can encode that in a, in a deterministic package. Yep. Um, all right, next thing here. Uh, oh, well, I'll, I'll just share a quick story. So I built a Slack bot we were using inside of mother duck. Now we call it quack bot that is connected to our database and, and you can ask questions too. And it will, it will tell you,
55:15you know, how many leads did you have yesterday and all this, all these things. Um, and I started looking at the logs and, uh, one of the things that other people are asking for, one of the top things is, Hey, okay, great. This is a good answer. Can you send this to me every Monday? So even in the, even in the non-deterministic UI interface, people are like, okay, how do I make this? Like, uh, and I'm like, all right, I probably should add that. That's probably,
55:40that's on my list of a short list of things to add some sort of cron inside of the quack bot. Absolutely. Which, which is maybe not a, uh, a standard report per se, but it looks like it. Um, all right, next question here. Uh, what about big players implementing ontology layers for human
56:00agent data access? Yeah. So, so this is an area like we step outside my expertise when we start talking about ontologies in, in any serious way. Okay. What I would say is my read on it is we do
56:15need some standard place to define the, all the soft stuff that used to just exist in people's heads. Right. We do need some standard place to define that. And I don't use the term soft, like in a pejorative sense, right? It's just, it's not a data pipeline or a metric calculation. It's,
56:37you know, what do these words mean and how do they relate to one another? And, and we all know that we don't need an ontology necessarily. I mean, we all have a shared ontology, but, but it exists, it's distributed. It's a super highly parallel processing distributed system called the human brain of all the people who work at our organization. Right. And so we do
56:58need some way to distill that into something that AI can reliably read and understand. Um, as far as who builds it and how they build it and, and what it looks like, you know, that, that's kind of where you get into the, uh, like the Juan cicada, Jessica talisman world where I
57:16defer to their expertise. Yeah. I feel, uh, I feel the same. I feel the same way. I'm not sure. I have never worked at a company big enough to need an ontology. Um, and, uh, you know, for those of you who are there, uh, please, please let us know how it's going. Yeah. Um, yeah, we're going to wrap. Yeah, exactly. We're going to wrap up here. Um, Ryan, what do you have
57:39to plug? Yeah. So listen, uh, the first thing I want to plug is if, if you like this conversation, I'm actually doing a live stream at 2 PM on Friday, where I will vibe code an app in a mother duck using dives and flights, like just live. It's kind of a low key expect things to go wrong. You know, I'll just be messing around sort of thing. So it's just like a cool come,
58:01hang out in the chat, ask questions, that sort of thing. And then other than that, um, super data brothers show I'm in season. So that's every Thursday at noon, tomorrow at noon on my LinkedIn profile, I will be, and on the super data brothers YouTube channel, it will be Matt Housley joining me. So we're going to be talking about actually BI, the state of BI. And, and, um, Matt has some pretty interesting research he's doing on incorporating like vector stores and
58:26knowledge graphs into BI. He thinks that this has gotten radically easier, but nobody realizes it yet. Um, so we'll be kind of going deep on, on topics like today, plus that sort of stuff. And then my book with Wiley, uh, tentatively titled, um, building, uh, data apps with AI. It's a highly Googleable name, um, will is coming out in April. So, so look for that.
58:50Okay, great. Well, now connect with me on LinkedIn. Yeah, please connect with me on LinkedIn. If you want to follow my work. Um, amazing. Uh, thank you so much, Ryan. I'm just going to plug one thing real quick here. I put it in the chat. Uh, we just launched our, uh, conference this fall, uh, data outpost. ai, where we're going to have lots of conversations like this with,
59:11um, lots of people that are way smarter than me. Um, really, really excited to get that out there. Um, we have a special intro price on it now. Um, would love to see some of you all there, uh, should be really, really fun. And, um, we will see how far things are along in this, in this pathway, you know, in, in a few months. So, uh, thank you so much,
59:32everybody. We're going to wrap up. Um, we will see you all around.
