The Data Stack Show - 236: Ringing Out the Old: AI's Role in Redefining Data Teams, Tools, and Business Models

Episode Date: April 9, 2025

Highlights from this week’s conversation include:The Impact of AI (1:25)Historical Context of Technology (2:31)Pre-existing Infrastructure for Change (4:42)AI as a Personal Assistant (7:10)Future of... Company Roles (9:13)Managing Teams in a Dystopian AI Future (12:31)Business Architecture Choices (15:52)Integration Tool Usage (18:07)AI's Impact on Data Roles (21:53)AI as an Interface (24:04)Trust in AI vs. SQL (27:12)Snowflake's Acquisition of Dataflow (29:54)Regression to the Mean Concept (33:49)AI's Role in Data Platforms (37:04)User Experience in Data Tools (44:41)Future of Data Tools (46:57)Environment Variable Setup (51:10)Future of Software Implementation and Parting Thoughts (52:10)The Data Stack Show is a weekly podcast powered by RudderStack, the CDP for developers. Each week we’ll talk to data engineers, analysts, and data scientists about their experience around building and maintaining data infrastructure, delivering data and data products, and driving better outcomes across their businesses with data.RudderStack helps businesses make the most out of their customer data while ensuring data privacy and security. To learn more about RudderStack visit rudderstack.com.

Transcript
Discussion (0)
Starting point is 00:00:00 Hi, I'm Eric Dotz. And I'm John Wessel. Welcome to the Data Stack Show. The Data Stack Show is a podcast where we talk about the technical, business, and human challenges involved in data work. Join our casual conversations with innovators and data professionals to learn about new data technologies and how data teams are run at top companies. Before we dig into today's episode, we want to give a huge thanks
Starting point is 00:00:31 to our presenting sponsor, Rudder Sack. They give us the equipment and time to do this show week in, week out, and provide you the valuable content. Rudder Sack provides customer data infrastructure and is used by the world's most innovative companies to collect, transform, and deliver their event data wherever it's needed, all in real time. You can learn more at ruddersack.com.
Starting point is 00:00:54 Welcome back to the Data Stack show. Today, you get me and John talking about data topics. We thought it would be fun for us to just shoot the breeze on a bunch of data stuff. And so we're going to talk about a number of different things on the show today. John, we don't have a guest, so I'm just going to say welcome to you. John Merriam Welcome to you as well, Eric. Eric Bischoff Thank you. I feel so welcome.
Starting point is 00:01:17 Okay. I feel like we're ringing the rag of AI on the show. No pun intended. I actually didn't plan on that joke. I don't know. That seems planned. I really didn't. I was thinking ringing a jive. Oh, it worked really well.
Starting point is 00:01:34 Yes, thank you, thank you. Talking about AI, but really we have to because we're living through such a fundamental shift in so many things and it's happening in real time. I feel like when this stuff first started coming out with the first couple iterations of GBT, it was really over rotating on so many podcasts and news articles about like, okay, this is crazy.
Starting point is 00:01:59 But it's come so far that it really is, I think the big topic, right? You know what I've wondered recently? What other thing, if we had as much hype It's come so far that it really is, I think, the big topic, right? You know what I've wondered recently? What other thing, if we had as much hype about it as AI, would progress really fast if billions and billions of dollars got put into it? Right? Because part of the success of AI is not like, yeah, there's a lot of advancement there that's cool, but it's also the crazy amount of investment from every major technology company to make it progress. advancement there that's cool. back on the history of technology,
Starting point is 00:02:49 even the conversations around data specifically, and there just aren't, it seems like there aren't that many fundamental changes as significant as this. I would agree, and I don't think we've talked about this before, but one of the my original attractions to data was that it wasn't going to change so much right so like it was on the list because like you get into tech and you're like oh front end like web frameworks the joke's always
Starting point is 00:03:18 like oh that changes every like five minutes there's always a new one there's always something new right and then it's like well databases and, those have been around a long time. That's not gonna fundamentally change and there's practical reasons. As far as on the front end stuff, you can change things with a lot less consequences, typically in the backend. And then you get into the world today,
Starting point is 00:03:38 you're like, that might not be true anymore. He, I love, I mean, I love that generally that you were like, this data's not gonna change that much. Right, right. more. one comparison that comes to mind is e-commerce Okay, so The and the reason that came to mind as an example of the fundamental shift is that It was an Entirely new way
Starting point is 00:04:22 To do something right it fundamentally changed the way that people shopped you're talking like we went from in person or a catalog to being able to shop online to being able to shop online right and I mean there I would say there are generally some interesting parallels for any sort of major change like that. The environment needs to be there. Internet and browsers and other things like that. There are a number of interesting parallels there.
Starting point is 00:05:00 You have compute power and you have the advances in the actual large language model technology themselves. You have the transformers. Like there are a number of things that were precursors to create the environment in which this could happen. You know what's really interesting around the compute? I have been thinking about this too. So crypto first, big boom, lots around that. And then the AI stuff. Like I don't actually know enough about the background here,
Starting point is 00:05:27 how much of stuff that maybe was provisioned or thought of, yeah, we're gonna use this for cryptos, like, oh, throw that in an AI problem instead. Yeah, that is interesting. Because that's interesting, because there's this huge boom, and we know that it's true from a power consumption and the baseline stuff of things you can use for cryptos.
Starting point is 00:05:42 Yeah, that is interesting, yeah, yeah, sure. I wonder if some of that is, and we'll continue to shift. I mean, crypto's still a thing, obviously, but we'll continue to shift to AI. baseline stuff of things you can use for crypto. I wonder if some of that is and will continue to shift. I was thinking about my grandparents. My grandmother is still alive. She's 97. My grandfather passed away a couple years ago and they didn't use a computer. Really. They had a computer but they just they didn't use it right. They just it wasn't part of their day-to-day life. And my grandmother shopped a lot on QVC right like call this 800 number. Yeah okay yeah, but what's so interesting is that it when ecommerce comes about you have it's
Starting point is 00:06:28 There are a it was a very large group of people I Mean, it's kind of crazy to think about this who didn't really have a personal computer, right? They went to work. They use a computer at work but like they don't really have a personal computer. And so the very concept of shopping online and not having to use a phone is a wild concept. And of course that changed very rapidly, but it feels the same because it's hard to even imagine that you would have an AI agent doing these operational things. It's just like, whoa, that's kind of, it's just fundamental. There's sort of this fundamental shift and it kind of seems like e-commerce, I don't these operational things. There's sort of this fundamental shift,
Starting point is 00:07:10 and it kind of seems like e-commerce, I don't know, that was the main thing that came to mind. The interesting thing about the AI operational agents, let's just say in a personal, like for personal assistant or whatever. But from an accessibility standpoint, you just opened up kind of a luxury service, right? Like not like, like not everybody's going to like be able to afford or want to use a travel agent or not everybody's going to have like a personal shop or whatever. But you've got this like what's historically been this like kind of luxury thing that if the AI gets there to be able to do those things like you open it up for a ton of people. So that's an interesting like space thing versus
Starting point is 00:07:55 like opening up something that's more like mundane that like yeah we used to like do this manually now it's more automated. Like that's one thing but taking something that like people valued highly enough to pay a lot of money for in commoditizing, I think is fascinating. I think one of the really fascinating things about this technological shift is how many things it's impacting.
Starting point is 00:08:19 Simultaneously. Simultaneously, which is in, I mean, completely unrelated, it's probably an overstatement, but in in I mean completely unrelated It's probably an overstatement, but let's just say completely unrelated spaces So for example with ecommerce again, it changes the way that you shop and I mean there are a lot of things right the way that Payments and I mean there is it's so much that that grew sort of out of that, right? But if you think about the example that you just gave of okay I can use AI to book a reservation at a restaurant.
Starting point is 00:08:46 It's also fundamentally changing the way that we think about developing software, right? It's fundamentally changing the way that people are even thinking about finding information generally, right? Which has an impact on the way that you even think about searching the internet at all, right? And I mean, it's changing the form factor there to some extent. And so you have all these different areas where it's having sort of disruptive impact. We had that conversation the other day of like, like Google it, right?
Starting point is 00:09:17 Is like we say that all the time, but you're like perplexity, like, well, it's the AI perplex? Just like fundamental things like that. I don't know what the verbal is there, but there will be one. Specifically with data. I had this honestly horrifying thought the other day. We were talking about this before the show. And it's like, all right, you take all this stuff out five years or ten years or however many years. Like, what do companies look like? five years or ten years or however many years.
Starting point is 00:09:45 One of my horrifying thoughts was that it might really drastically change a bunch of roles, which a lot of people think that. But it might turn into essentially most companies have sales people that sell things to other people and then like operations people that operate and like and beyond that like of course there's still gonna be some levels of specialization and you probably still have some kind of finance accounting things some kind of thing but I think a lot of companies will consolidate down into less divisions and we were talking about SAS specifically I mean the dream like me being from like a technical background, the dream is like yeah like we're gonna put that landing page out there, drive some inbound traffic to it, people put the credit card into stripe
Starting point is 00:10:33 but I've got a SaaS product and we're gonna scale it and grow it. You don't have to interact with people. Like because of the fundamental change of like this barrier to entry, I think probably continuing to lower into being able to create a SaaS product. I just think it's going to fundamentally change the growth trajectory of most of them and it's going to be sure you'll still have the viral stuff and you'll have like influencer driven stuff. I think the other part is like, well, you're gonna have to sell a lot, like probably have a lot more people selling, which if you're a founder somebody that's technicals like Yeah, this is terrible like I don't know. I don't like this future. Yeah, I think it's a real like yes ability
Starting point is 00:11:11 I think it's already happening. Okay. Let's can we explore this topic by By digging into this dystopian future yeah, and by the way I want royalties if this turns out We want a book that you're to be a book that you write. A book that you use AI to write. But okay, let's dig into this dystopian future. So I'm thinking about our listeners who are very similar to you. They are managing data at company XYZ, right?
Starting point is 00:11:39 Which is roles that you've had, okay? In this dystopian future, let's actually make it specific. roles that you've had. several engineer type roles and some specialization in that, like one that's more focused on pipelines and integrations, one that's more focused on front end stuff. A team, we had an offshore group that we used for certain parts of the tech stack. What else? Yeah, so that was the basic tech side of the house. What am I talking about? I'm missing something. No, but yeah, I think that was the basic tech side of the house and then like on the digital side marketing paid advertising Yeah agency right as you manage all that side. Yeah, you might that side of it is all eventually part of it So yeah, that's interesting. Okay, you manage both of those teams. Okay, so which is not very common
Starting point is 00:12:36 That's not very common, but it's actually interesting because it'll make the dystopian future spicier, right? Okay in this dystopian future. What is dystopian future spicier. Well, and you have the sales team that's taking calls from customers who want to order 10,000 of this particular part that you were selling. In the dystopian future, let's start with the data team. What does that actually look like? Right? And let's just say, let's just say for example, because we need a protagonist, that you're still a human in this equation, right? As the leader of data, right? What other humans are there and what do they do
Starting point is 00:13:16 and what's been replaced by AI? So I won, I think there will be more merging of like, like in this particular thing, there's like maybe a digital ops type thing and then maybe that because we did like warehousing an actual physical because maybe there's like physical ops so there could be divisions but I think it's like this digital ops thing and you so you do you sir someone reports to you their title is digital ops something yeah director of digital ops yeah yeah it's a very flat company at this point. Yeah.
Starting point is 00:13:45 Right. But Director of Digital Ops, and you essentially have what people that used to be in marketing type roles rolling up to that. You essentially have people that used to be in more specialized technical roles. And that Director of Digital Ops, maybe should be Manager of Digital Ops.
Starting point is 00:14:03 Because it might only be a couple people. One person that's a little bit more on the marketing creative side that's managing an army of AI, more creative type things. And maybe some outside help as well on areas that still require specialization or deeper knowledge. And then somebody that's a little bit more technical leaning that's doing data movement and transformations of data and things like that. But I think the spread of hey, right now the graphic designer could never do the
Starting point is 00:14:36 database administrator's job, I think your spread is going to be a lot tighter. You're going to have the ability to have people that are way more down the middle that like, yeah, this person is better, they're more people that are way more down the middle. This person is better, they're more creative, they're more on the marketing side, this person is a little better on the technical side, but is going to be way less extreme than it is, is what I suspect. I want to know a couple things. I want to take a practical look at a couple of things in the Suspian future just as far as like managing data and the data stack itself okay and let's just say okay also do you have any analyst and so you have the
Starting point is 00:15:14 director of digital ops you have a creative you have a creative person right a creative person. And then a generic technical person. So they're essentially a team of three. The DigitalOps team. Potentially. There's a lot of things that would have to go right or wrong. I'm not sure which for this to happen, but yeah, potentially. And what was the team size previously just across? And order of magnitude is more like, so we're at three, I don't know, call it like 18. Wow. Okay. Yeah. So that's significant. Right. Okay. And each level of, and it will drastically depend, I think, on what you're doing.
Starting point is 00:15:56 And it will depend on how you want to architect the business. I think there will be a lot of businesses that you can architect like, hey, we want to optimize for at least people possible. Yeah. People opt for that model. Others will opt toward like, hey, human touch, like this is a really big part for us. Yep. We're going to optimize for another model.
Starting point is 00:16:12 Yeah, yeah. I agree with that. I actually think that, I mean, there are already entirely new business models and ways to think about operating a company. Yeah. We talked to a founder recently like single one person founding a SaaS company doing extremely well doing all of it.
Starting point is 00:16:31 Yeah, Mike Dragalis, Shadow Traffic. Yeah, Shadow Traffic. Yeah, yeah. And then can probably scale quite a while. Yeah, yeah. Just him. Okay, on the practical things, so you need to set up a new data pipeline. So you get into set up a new data pipeline.
Starting point is 00:16:45 So you get into work on Monday, there's a meeting with, I guess, a small number of people. Everyone fits in a company. You need to create a data pipeline to do something. A feed of inventory to some system to do something. Let's just say update inventory in real time or something of that nature. to do something, right? set up a quick meeting with your digital ops team, who does what, and what does that process look like, and how are they in this dystopian future creating the pipeline, all that stuff. Yeah, so I think you're going to buy tools of like, hey, we have an integration tool and we bought a tool. what are your specs, okay, the origin, and then destination, like vendor B or thing, like show me your docs, show me your specs. And then like
Starting point is 00:18:08 all right, we use X, like middle layer tool, and essentially like feeds all three of those with a little bit of guidance to some kind of like AI type tool, or maybe that will get built into the integration layer eventually, and says, okay, like go build this thing. And I do not think it will be perfect the first time for a long time, and says, okay, go build this thing. And I do not think it will be perfect the first time for a long time, but I do think somebody slightly technical will be able to coach it through a few quick iterations
Starting point is 00:18:33 and get to something that is pretty good. And then it's gonna depend on how important is this? Should we have some kind of code review step? Can AI do the code review? I think right now that's a really tricky step how important is this? Should we have some kind of code review step? Can AI do the code review? I think right now that's a really tricky step because you can get pretty far with vibe, code, and concept. But it doesn't make sense, and I wouldn't be comfortable. I think most people are not comfortable for true production use of a lot of this stuff. But, I mean, actually there's a business model out there already.
Starting point is 00:19:00 I think it's pullrequest.com where you just have a third party review all of your pull requests. So you can imagine something like that integrated into your system and you've got one person kind of coaching AI like docs, docs, integration, and then you send it off to pullrequest.com. They review it, they happen to have a specialist as an expert in whatever tool that you're using to integrate, and they're like, all right, it looks good. basic flow. Okay, yeah, so we got all the data flowing, and we want some insights on the orders we have flowing through the system or something.
Starting point is 00:19:48 I think so. I think it would start with some kind of like okay, what does the ex-executive want to see, or what does whoever want to see, and then there's probably again, somebody that's responsible for feeding that into the system, and then coaching it through a couple iterations to get
Starting point is 00:20:04 something that they know is the the system, and then coaching it through a couple iterations to get something that they know is the right thing, and then saving it off for the executive to look at, and then theoretically, the executive has an easy way to tweak it a little bit more if they want to. Yeah. Okay, what we're kinda getting at here, one of the interesting threads, thank you for giving me a glimpse into your dystopian future.
Starting point is 00:20:22 Yeah, I do think there's another version of this that also is likely to happen, for giving me a glimpse into your dystopian future. I do think there's another version of this that also is likely to happen, where we have lots of horror stories that come out of where AI really screws stuff up. And this will happen, it just depends on at what rate, and who's to blame, of security breaches And depending just how rocky that gets, I could definitely see another version of this where like the future is actually five years from now is not that different. But the reason I think that is less likely to happen than I would have believed previously is how much money is tied up in all the major tech, it tends to happen. If fill in the blank with some other company, things just tend to happen. Because that's who the big companies trust.
Starting point is 00:21:29 With like, what should we do with our tech? Right, right, right. Fascinating point. So what areas, that's a great, that's actually where I wanted to go next is, which areas do you see in data, AI having the most impact, and where is it gonna have the least impact?
Starting point is 00:21:55 I mean, a couple specific examples of that, right? Like one of the things we talked about was producing the analytics around this, right? And so you don't need a team of analysts anymore, One of the things we talked about was other areas. You probably still have analysts depending on what it is but they'll actually be analyzing. Most analysts don't actually analyze anything. Yeah. So I do think, yeah, it's true, right? They clean data, they move data around, they copy data. Yeah. So I actually think you probably still have analysts in some form or fashion. It's either combined with another job because it doesn't need to be a full-time job or to your point or to what I was saying earlier,
Starting point is 00:22:44 like they actually start analyzing data versus just moving it around. Yep. And what do you think, what are examples that come to mind of things that won't change? Things that won't change. I mean what's going to be largely the same in five years? I mean I know the actual answer. I largely the same in five years from now. There's one glaring answer. Storage, essentially we will probably still use similar commodity storage for data. Yep. And there'll be a lot of noise happening above the storage, but I don't think the storage changes fundamentally.
Starting point is 00:23:23 I agree. The the storage changes fundamentally. I agree. The spreadsheet. Oh, sure. That's why he had never die. It will never. She'll probably never die. But but I also think that one thing that AI can't kill. Yeah. But I also think since like essentially like S3 or S3 equivalent
Starting point is 00:23:37 is behind every one of these things. Yeah. Still. Yeah. That probably doesn't. Yeah, I agree with that. But on the user facing side, yeah, some version of a spreadsheet still, that probably doesn't change. show notes. My laptop, the way we're sitting, it's too far away for me. It's too far away to Google that live, to perplex it live. Sorry. So one of the big things that I think is going
Starting point is 00:24:11 to be really interesting as we think about where AI is going to have impact and where it's not is the dynamic of, how do I want to frame this? Essentially being an interface to all sorts of other platforms and tools, which is really interesting. So here's an extreme example. If you could essentially manage your infrastructure, let's just say Snowflake or whatever it is, if you could just manage all of that through AI in the same way that you talked about, right? Okay, here's some documentation, here's whatever, like just go do this thing. How much are you logging in to Snowflake?
Starting point is 00:25:08 How much are you logging in to these platforms? It's just interesting to think about the interface for that changing. Whereas essentially to all these platforms we use to directly interface with, essentially you never look at anymore or you only look at if there's a problem. Right, right. Well, yeah, it essentially becomes the platform's utility becomes troubleshooting and other things like that. I mean, visibility, observability, those sort of things. But that's really interesting. I mean, one of the ways this is already having an impact is like general decreases in website traffic in certain areas. I was talking with a friend who works in and
Starting point is 00:25:52 around that industry. And it's like, oh wow, the actual inquiries that people are making, I think are dramatically increasing. It's just that some of that is shifting over to GPT, right, instead of going into Google and then going to a website, right? It's just that it's delivering that end user visit directly to you, right? Well, I mean, I think it's fascinating because we've talked recently just in the customer data space and attribution specifically, like a tool like a rudder stack doing server-side attribution and looking through like kind of raw data around that seeing open AI and seeing like these other tools pop up in the attribution is really interesting. Like and I'm actually seeing
Starting point is 00:26:38 that happen especially in like in satchel and, wow, like there's a decent amount of traffic being driven from these AI tools. Yeah, yeah, that's super interesting. I do wonder though, what the impact of the trust factor is gonna be, right? Especially when you think about things like production pipelines. Yeah. Where, I mean, it's just hard to trust, right?
Starting point is 00:27:03 Like you just, it's really hard to trust, right? It's really hard to trust. As opposed to writing SQL, right? Which is sort of exploratory by nature in many cases, and iterative. And I think that's the thing, right? If you can get to a result that's abundantly clear if it works to a human. console you can still have a security problem. But I do think it can be less evident further down the stack of, oh, we made a major problem here and it's an edge case that we'll find eventually, but no idea when. And then the question becomes, well, that's true of humans too, right?
Starting point is 00:28:00 We're going to make mistakes that show up later to bite us. And the question becomes, is it more frequent with AI? Could we use the AI to try to catch those in a separate tool that doesn't have the knowledge of the original tool? Just like you would with humans, like a double-blind audit situation. Right, right, right. Yeah, super interesting. We're going to take a quick break from the episode to talk about our sponsor, Rudderstack. Now, I could say a bunch of nice things as if I found a fancy new tool, but John has been implementing Rudderstack for over half about our sponsor, RutterStack. customer data can get messy. running production instance of Rutter Stack at six years and going.
Starting point is 00:29:10 Yes, I can confirm that. And one of the things about the implementation that has been so common over all the years and with so many Rutter Stack customers is that it wasn't a wholesale replacement of your stack. It fit right into your existing tool set. Yeah, and even with technical tools, Eric, things like Kafka or PubSub, but you don't have to have all that complicated customer data infrastructure. Well, if you need to stream clean customer data to your entire stack, including your data infrastructure tools,
Starting point is 00:29:40 head over to rudderstack.com to learn more. Okay, well, continuing on with the discussion about AI, Snowflake made a big acquisition, DataVolo. Is that how you pronounce it? Yeah, DataVolo. They announced it late last year and it seems like they're starting to kind of roll out and have customers using it more
Starting point is 00:30:00 in this first quarter going into the second. Yep. Okay, industry-wise, we'll just do industry pundit. it more in this first quarter going into the second. industry-wise, we'll just do industry pundit. We'll do an industry pundit segment. What's the move by Snowflake there? How are you reading it? Yeah, I think... What does data actually mean. at a really high level. So the specialization is unstructured data.
Starting point is 00:30:44 Right. So you can move data that is in various different formats that you would want to run AI workloads over, which is different than a typical. Yeah, than a typical ETL tool, for example. Right, exactly. Exactly. And it's interesting because I've
Starting point is 00:31:03 seen there's a number of tools in this space. And then there's some interesting ones like Ultrex is one, there's several others that have been around a while. This Ultrex is like more of a graph, graphical, like you can, and they've got these little modules. I'm not, I haven't super kept up with what they're doing, but I believe they're getting into the AI piece as well. And it's interesting with that space where there's going to be
Starting point is 00:31:29 these all like all in one tools of like data, AI use case, great ETL, great, like whatever you want to do. Yep. And they're like a graphical, more of a graphical interface. Then I think there's these specialists, more specialized tools like, hey, I want to move customer data around. I want to move, like I said, AI, hey, I want to move customer data around. I want to move, like I said, AI data around.
Starting point is 00:31:46 I want to move structured data around. And so it's like, there's two accesses. There's like the no code, low code versus like as code, like movement. Like, hey, I want drag and drop. I don't want to do any code. Or hey, I don't want to interface at all. I want it all to be like YAML or something.
Starting point is 00:32:03 So that, those two accesses. And then there's the specialization access as well, I don't want to interface at all. I want all to be like YAML or something. So those two accesses and then there's the specialization access as well as far as like really attuned to the customer data movement problem. We really nail that use case versus like we're a generalist tool. We'll move data wherever you want it. It'll be interesting to see what happens there. Yeah, it really will be interesting. I mean, it is going to be fascinating to see how, because the cost is decreasing as well. And if we go back to our previous example of like an interface on top of this tool, and you think about Snowflake, and actually, interestingly, we were talking, who was I talking with someone? They were running Neo4j inside of Snowflake. And actually, interestingly, we were talking, who was I talking with someone? They were running Neo4j inside of Snowflake, which was really cool.
Starting point is 00:32:47 Huh. A graph database. Yeah. We were talking about some identity resolution, identity resolution type use case. Anyways, they were talking about run, it's super cool, right? You just, you can run Neo4j inside of Snowflake and actually you can sort of like push stuff out as tables or views, a couple other things. I don't know all the specifics, but just like, oh, that's cool. Actually you can push stuff out as tables or views, a couple other things.
Starting point is 00:33:10 I heard that and then you think about DataVolo doing unstructured data for AI workloads, etc. This is clearly Snowflake's long-term intent. Okay, what do you want to do? You mentioned customer data, right? Do you want to build an identity graph? Do you want to do something with generative AI over this or whatever? It's got the pipelines, it's got the query engines, all that sort of stuff. It's pretty fascinating. Yeah, yeah for sure. And this actually reminds me of, you're familiar with Thinking Fast and Slow, the book? I haven't read it. You haven't read it? Okay, it's a good read and I am probably
Starting point is 00:33:52 gonna butcher this, but it got me thinking about, there's this concept that he talks about the book, not specific to this book, but like regression to the mean, right? So it's essentially you've got like outliers and there's this like principle like things regress toward the middle. Yep. The example he gives on the to the mean, right? So it's essentially you've got like outliers and there's this like principle like things regressed toward the middle. Yep. The example he gives in the book, which I think is fascinating, is it talks about coaching. Say you're coaching you up, say you're a baseball player and you like strike out and you've got a guy on third base and like, ah, like, come on Eric. And the coach yelled at you, right? Yeah. You go up and like next time like you play better. It's like,
Starting point is 00:34:22 oh, that must have worked. Opposites for you, like you go up, you like you play better. It's like oh that must have worked Opposites for do it like you go up you hit a home run. It's like good job Eric Yeah, and the next time like you do worse So what so you're the coach like oh, I need to be hard on Eric He does better every time but the actual principle here is you regress to the mean So you hit a home run and then odds are you're gonna do worse next time you struck out odds are or whatever struck out two times Right. Yeah. Yeah, this is interesting thing. You're going to you're going to you're going to eventually reach your batting average. Yes, exactly. Yeah, exactly. Exactly. So like the way it relates to like this and AI that I've been thinking
Starting point is 00:34:59 about is I think there's gonna be a stronger pull to the mean if people are using AI tools. Because if you're thinking about this like AGI concept and stuff of like, and this is kind of a pushback on like the generalist concept that we like launched off with. There is I think going to be this like these unique scenarios where like there's such a pull to the mean of like, oh, we should solve this in this one common way. Where people are going to be like, no, like no,
Starting point is 00:35:24 actually there's this like novel way that like is much better. It's not pulled to the mean of like, oh, we should solve this in this one common way. Where people are going to be like, no, like, no, actually, there's this like novel way that like is much better, not pulled to the mean. And that's where I think a lot of the engineers, like really good engineers are going to gravitate toward those problems. Whereas like like ETL of like, oh, OK, cool. Ninety eight percent of the time, 90 percent of the time, like, yeah, use this generic detail tool. It's the right tool. And there's going to be a stronger pull there where that could have been
Starting point is 00:35:47 60 percent of the time before maybe it becomes 80 or 90% of the time. But the last 15 or 20%, I think will exist for a long time. And then engineers will work on those like really interesting problems. Because they I don't think that goes away completely. And I don't think that strong pull to the mean like, well, you might get to 80% or some high percentage is going to go to 100%. Yeah yeah super interesting I need to borrow that book from you. Yeah it's a good read it's just got a lot I should read it again this got a lot packed into it. Okay let's dig into that I want to dig into that topic a little bit more in terms of specialization and generalization.
Starting point is 00:36:25 Okay, there's kind of an accepted narrative. Take Snowflake for example. We'll praise them and then we'll pick on them a little bit here. Perfect. Okay, and they've acquired a number of companies. I mean, they've been very acquisitive, which makes sense. Streamlit, DataVolo, a number of companies.
Starting point is 00:37:06 And it just becomes this like really big platform to like right you said to build something right and so this the narrative that sort Of generally accepted is as that happens though That it becomes a big platform right that you can do anything in and that's actually part of the problem and what creates the opportunity for a smaller specialized companies right to Disrupt and so in the world of data like I mean it clearly the storage aspect in the thing smaller specialized companies to disrupt. In the world of data, clearly the storage aspect and the things that we've talked about with Snowflake and with Databricks, there's consolidation there because they want to be large cloud platforms. What are the other tools that you think are going to get generalized like that.
Starting point is 00:37:45 I don't know. I mean, Microsoft has already, from a marketing perspective, approached it like, hey, Microsoft Fabric. And then like, that's the marketing thing, and it's just one thing, you have all these components to it. In reality, it's essentially just a bunch of different components that they branded as Fabric. But I think that happens for Databricks, for Snowflake, for others, where it's like, cool, like data stuff, like do it in our platform, you can do it all. You can have the VisLayer, you can have the pipelines, you can have the storage, you can have the AI,
Starting point is 00:38:12 LLM built into it, you can do all the things. And the question in my mind is, does those companies being able to use AI internally change the equation where it used to be like, oh, well, yeah, that happens. You become a generalist as a company, you grow big. Great. But then opens up a bunch of doors for specialization to do X piece better. Yep. Is that still true when when these companies have these sophisticated AI models where maybe they can juggle more? I think probably. Yeah, it is.
Starting point is 00:38:43 Because guess who else has the AI models and similar technology? The innovator has can juggle more? all this AI and it's gonna be terrible is like, yeah, sure, but so will all the security companies. Right. So it's kind of like, there's like equal force both directions. But other than like those companies like consolidation, I for good or for bad, and actually probably a little bit more for bad, I do think it ends up, you end up picking more mainstream winners where you have a bigger gap between like the mainstream winner for the most use cases is like down the middle and that's like a really big like chunk of the market and then
Starting point is 00:39:36 you still have like the like I was saying like the people that really nail like a specific painful problem on the sides, but I do think it probably makes that gap bigger where if like down the middle like CRM for example, salesforce does not have 90% of the CRM market. I don't remember the number but it's low. I think it's below half. It's really low. I might be wrong about that.
Starting point is 00:40:01 It might be like a little bit, but it's not like 90%. Interesting. One of those books. Yeah, we should think about that. I'm going to reach over to my computer and google this. One of which is, you got somebody like HubSpot, second is industry specific CRMs that pop up, and third is a number of companies that don't really use CRMs still. I am... Perplexifying it. I'm not actually should use perplexity. Sorry. I just I use raycast and it defaulted to just drop the e4o But look at raycast is doing a really nice. Yeah, we're not up to raycasts. Very cool 21.7 to 21.8 21 21.8. No, it's tiny Wow
Starting point is 00:40:44 Less than a fourth Wow, I am. No. Yeah, it's tiny. Wow. Less than a fourth. Wow. I am processing this. Yeah. Less than a fourth of the market. That seems so crazy. Yeah.
Starting point is 00:40:55 Now, the real question will be though, I say like there's gonna be more like progression to the mean and more like that middle lane. Maybe it gets less congested, but it could just get more competitive and not necessarily combined in like one product. Yep. Because you could still have a three or four major people that are in that middle lane that are essentially the same but they're still competitors and they still have fine and they're still and they're selling
Starting point is 00:41:22 different ways and people just prefer one over the other. Yeah. I mean, think about clothing brands. That's true. I'll just close and we've got a ton of those. Yep. Or car, like same with cars, right? It could become more like a car shopping experience of like, does it have four wheels? Yes. Does it have four doors? Can it track? Like, it's like, they're all, and car people are going to be super like, no, they're not. But like from a transportation standpoint, like they're all pretty car people are gonna be super like no they're not but like from a transportation standpoint like they're all
Starting point is 00:41:47 pretty much the same yep, but Technology could become more of the carbon thing of like yeah Well, like you got four four options like down the middle in this like 89% lane. Yeah, then you just go with your preference Yeah, essentially. Right. mentioned Snowflake he's like yeah I he's like there's you obviously if you're building out a data stack you have to have a data store blah blah blah right okay he's like so you get a data warehouse right for data warehouse stuff he's like I like Snowflake and he kind of paused oh he said yeah he's like I
Starting point is 00:42:58 like Snowflake just cuz I like it he's like they're just something about it like I just I like the I like it that Yeah, it's already starting. We're like a lot of these There there's definitely not feature parity So I'm not saying that between all right them, right? But I think there will continue to be closer to feature parity. Yeah, and I really think it's gonna be more of a car thing Yeah, there's still there will be definitely differences like as there are with cars about like, yep I want to optimize for this use case. I like off-roading or I don't know I mean, as there are with cars about, to the mean. in their toolkit in terms of creating competitive advantage
Starting point is 00:44:06 from themselves from giant incumbents is a dramatically better user experience. Yeah. I actually think 5Train is a great example of this. I mean, they just have a phenomenal, it's just so easy to use, so easy to set up. It's great, right? It just is great, like compared to set up, it's great.
Starting point is 00:44:25 It just is great compared to a lot of other tools. You end up paying them money because you're just like, this is just a great tool. That's not the only thing, that's not the thing that made 5Trans successful or the thing that made Linear successful. But it was a big part of it and sort of reflects like the DNA of the company. But what's so interesting is it's going, it's getting so much easier if you think about these different data tools to deliver an absolutely phenomenal user experience. Which is super interesting, right? an absolutely phenomenal user experience, which is super interesting.
Starting point is 00:45:10 I think you're going to get a stronger and harder split between audiences here. Because for data tools for me, I'm gravitating really heavily toward fill-in-the-blank as code, BIS code, data pipeline, all that stuff, because the productivity increases drastically. And it will continue to increase, I think, with AI tools, because guess what? AI tools are good at text. There's some neat stuff out there with like chat gbt's operator thing where it can browse the web and stuff.
Starting point is 00:45:38 But that is nowhere near where it is with text. Everyone knows that. But from a human perspective, humans are like, no, I've yet. And maybe this day is coming. near where it is with text. But from a human perspective, humans are like, I've yet to say, man, this product is just like a killer user experience. It's so ergonomic and it's all command line. I've never had that feeling. Maybe we'll get there. than just nailing one or the other. Yeah, for sure.
Starting point is 00:46:23 Now, this is a really interesting point. I kind of think about Postman as an interesting example there. Because you can do a bunch of different stuff in a command line. There's so many niceties that they provide for doing all sorts of different things. Yeah, like graphical organizations, all those command lines. Yeah, and I think there'll be more of that. And I think that's when you can switch into like YAML mode and like clean stuff graphical organization. And I think that's when you can switch into YAML mode and clean stuff and switch back. That's a good example.
Starting point is 00:46:55 Any last AI thoughts before we turn off the recording? I think in conclusion, I really am torn between does the future look like that generalist future we talked about? Or does it look like that like regression to the mean where there's like the X percent that is like generalist, but there's like a ton of stuff on the edges where you actually get more hyper specialized because like the general problems are solved and like technical really gravitating hard toward the edges.
Starting point is 00:47:22 I think that's a real possibility too. And they're not necessarily mutually exclusive. really gravitating hard toward the edges. And they're not necessarily mutually exclusive. I feel like I did kind of interview you this. Yeah, old habits die hard, I guess. What's the craziest thing you've done with AI? lately Or like the thing that sort of you're like, whoa, that was crazy Yeah, I think front end stuff like messing with like hey, here's a landing page like really yeah new agreeable data website
Starting point is 00:47:55 Yeah, yeah, that's true. Yeah launch that definitely use that on some of the front end layouts, but just yeah like this general like and layouts, but just, yeah, like this general, like, very vague, because I'm not a designer by any stretch of the imagination of like, hey, make this landing page look good, like very vague language, and then like being pretty surprised with the outcome. Yep, yep, super interesting. What about you?
Starting point is 00:48:22 My turn? I think the coolest, basically building prototypes, which we do an immense amount of different things at that, but today actually I did something new. There's a tool out there that I was looking at, like, oh, I wonder if I should use this tool, if we should get this tool to use in the RutterSack platform. To sort of accelerate a feature, whatever. And so, it's a component, it's a set of APIs, et cetera.
Starting point is 00:49:00 And so I thought, okay, I go create a test account for this thing. That's great. I get the API key. And I was like, you know what I'm going to do, actually, is I'm just going to spin up a dummy thing, and I'm going to actually try to install this, like try to install this and actually kind of see what it's like to use this thing and see how it actually works on the back end and see the. Whereas before, you'd have to get a call with engineering, like, hey, let's do it.
Starting point is 00:49:27 Totally. Like, isolated environment. Right. And I'm not a software engineer. Right. I mean, I'm not a software engineer. I know enough to create problems for others. For others, right?
Starting point is 00:49:37 Yeah. Okay, but this is what's astounding. And this, to me, was just, I think we were talking about this the other day. So I create an account with this thing, I get the API key, I just hop over into Versel, which we use Versel. We've used a number of different tools, but we've deployed a number of different things on the Versel platform. And so we have an account and you can add vZero to the account. And it does a number of nice things. Which vZero to the account.
Starting point is 00:50:25 Amazing. I go in there and I click create new project in Vercell. And I was like, okay, I'm just gonna grab any, they have templates, right? And it's just like, I'm gonna grab anything or I'll just create something it doesn't know, right? So I go to create a new project. It's like, oh, there's a template thing. And I was like, I wonder what templates in there. So I go and look, it's like this product
Starting point is 00:50:43 that I had signed up for has a starter kit, right? And it's just a fully functioning Next.js app. templates in there. it does everything, right? And then the thing that I had to do to get it running locally was create a.env. Right, yeah, just fill in some environment variables. Literally the environment variable to get it running locally. I pull the repo and using cursor, I'm using how the API works and I'm seeing like I mean it was just totally astounding Yeah to actually go through with that right and then I can push it and it in Vercell will deploy it and I can share it with people on the team and like have a discussion about it, right? And everything's fully transparent and we can sort of see how this thing works just to me felt
Starting point is 00:51:41 That is a product demo I mean Holy cow because we talked about this too of like there's my last take our last hot take on this I felt that is a product demo. I mean, holy cow. There is this future where everybody in software now is selling generic things for people to use their imagination to implement in their company. I think there's a future where one of the major human value adds is like, hey, we looked up what your company does. We imagined for you what it can do.
Starting point is 00:52:07 And here's a demo of it, like exactly what it would do for your company. That's huge. That is really big. Totally. Yeah. That's wild. That's wild. All right.
Starting point is 00:52:18 We're at the buzzer. Thanks for joining us, John. Thanks for joining us. All right. We'll catch you next time. See you guys later. Grab it if you haven't. Thanks for joining us.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.