Latest / The AI Ecosystem Podcast / The AI Data Stack: Who is the System of Record?
Transcript
- Lindsey DeFalco: Hello, everyone. Welcome to our next episode. I do not know the number of the AI Ecosystem Podcast. It is myself, Lindsay DeFalco, with my co-host, Bob Moore. You want to say hi, There we go. Great. OK, so we are going to talk about something today that I think is going to be pretty interesting and exciting. It's been coming up in a lot of conversations for both of us over the past. Bob Moore: Hey hey, I did it. Lindsey DeFalco: few weeks and I also think it ties in nicely to our topic that we went into last time where we're talking about what integrations and what integrating technologies looks like now in this AI MCP era. So a lot of this is going to have to do with really what is the system of record now when you're building things that involve multiple technologies that are integrated together? What's the source of truth? and some more details around the that we're seeing in terms of who builds what, how it all works together. And with that really vague description, ⁓ Bob, do you wanna give us like your opening kind of rundown spiel on this whole topic? Bob Moore: Yeah, what a mess. It's like a beautiful mess and I feel like it is, we're watching the new of building ⁓ products kind of congeal in front of us in real time. But in the meantime, ⁓ don't think anybody really knows what they're doing or what they're talking about. ⁓ I have a very hard time. I think it's easy to find people with conviction. I think a lot of people ⁓ especially established companies that are doing a decent job of kind of playing ball in the AI ecosystem ⁓ are finding themselves like triangulating between how to make their legacy products relevant in this universe and also having enough of a ⁓ stake in the way people get value in this world that they can actually continue to build big businesses. ⁓ those two things don't always necessarily agree with each other, right? So to be more concrete about it, it seems like It seems like there are three very distinct roles to be played in the world of enterprise AI solutions. And to keep it simple, let's maybe just use the one analogy here of like the internal chat bot that helps make your sales team more effective, right? The ⁓ chat bot, the agent, whatever, like the entity that exists internally that's able to help, ⁓ you know, decide what deals to go after and how to win them and all that stuff. So powering that thing takes these three layers. One of the layers is systems of record of data. And in, a of these ⁓ and in terms, this would probably be your CRM ⁓ or kind of the knowledge base of who you're selling to and what the state of those deals are and who owns them and ⁓ things like But there's also call recordings ⁓ and there is ⁓ ecosystem data, like can get out of Crossbeam and there's all kinds of third party data enrichment data and ⁓ There is, you know, these are, these are mainly well established, kind of well tried and categories, but you know, there are companies that are kind of more AI native or AI knowledgeable in how they've been able to bring these data assets to bear through MCP servers and other approaches. And they end up being pretty important, right? So like you you need that stuff to exist somewhere. And then the next layer is like this data aggregation piece, which is something has to be responsible for pulling the data out of all those systems so that your actual agent or chat bot or whatever can then access it to apply logic and take actions and things like that. So this is this very interesting middleware layer. It's not unlike what we built at Stitch Data ⁓ in kind of the classic ETL sense. It goes and talks to all of these various places that your systems of record are. and pulls the data out in a format that's usable and puts it somewhere that you can use it. And if you look at the N8Ns or the Zapiers of the world, you look at different ⁓ that exist to get data into vector databases or to kind of do more ⁓ of organization specific storage of data that can be accessed on demand. Or even you look at like the classic ETL tools like the 5Trans of the world. They do a decent amount of this heavy lifting to kind of prepare data for AI and extract it either in batches. So it's stored in a way AI can get at it easily locally in one spot, or by enabling really good MCP routing and knowing what MCP server to talk to and making sure that all that data is available. So there's kind of this like routing layer of how do we get all the data from these systems of record available to my, my agent, my chat bot. And then there's the actual agent layer, which is like. the building of the, either the LLM itself in the case of Anthropic doing this, or kind of the scaffolding around the LLM, that companies like Glean are doing very successfully to basically help organizations make a very, very well, ⁓ equipped and ⁓ surrounded, and in some cases well contained. entity that can kind of like be the interface point where the prompts, whether automated or coming from humans are going to kind of then go through this, right? So the net result is Sales rep has a call in five minutes and says, Hey, I need a dossier on this company that I'm about to talk to. They go send that request to, know, call it the bot that's powered by Glean that's sitting inside of their company Slack. ⁓ It will go and ⁓ go out to the various data sources that are available to it and go and retrieve that information that it needs, which may be multifaceted in nature. The underpinning systems of record that contain that data are then kind of serving it up via these MCP layers. And then it all gets pulled together and delivered to that end person. this actually, you know, I started out by saying this is a mess. It sounds like kind of elegant and frankly, it kind of mirrors like the modern data stack of a decade ago. ⁓ of like systems of record and ETL layer and a storage layer and like layers of action, like a BI layer on top of it. The thing that's a mess is that the LLMs are so powerful that all these other layers are suddenly weakened. So like where you used to be able to be just a piece of ⁓ call recording technology. but layer on all of this additional like SaaS functionality and workflows and other things on top of it and build kind of a full featured SaaS package you could sell for hundreds of thousands of dollars. The LLMs are so good at the SaaS bit of it that you end up kind of getting reduced down to, you know, just the product that has the call recordings. ⁓ So then you're like, well, I, or look, I'll use Crossbeam as an analogy here, cause it's kind of like, you know, what we're seeing as well. ⁓ the authority for ecosystem data. And, but in a world where our customers have one of these stacks in place, we're basically just like a data source and the data is extremely valuable. And maybe you can just build an incredible business around the data alone, but the user interfaces and the workflows and the integrations that exist for humans inside of legacy SaaS tools, a lot of these things don't matter anymore to a lot of our customers. So the question becomes for us as it would for the call recording people, as it would for the third party data people. Should we then get into the business of aggregating? Should we be playing the role of talking to all the other MCP servers and pulling all the data in so that we become that one system of record where all the stuff is and we know how to go get it. And then if that's true, should we build an AI chat bot that becomes the one bot to rule them all? And I think that more so than ever before, those next two layers are easy to enter and largely undifferentiated from one another to a point where they're also really, really devalued. But what we're seeing right now is all the system of record companies getting into the ⁓ agents and building kind of ingestion of MCP game, but this big ⁓ question of like Spider-Man pointing at Spider-Man, like who's powering ⁓ the bots here? Who's powering the thing at the core of it? ⁓ it feels like maybe Anthropic is like winning that game right now in B2B, at least in... ⁓ like tech B2B enterprise circles, but ⁓ it leaves a lot of these legacy SaaS tools trying to play ball in AI in kind of the middle of nowhere. How's that for an eight minute rant on, I ⁓ mean, somewhat, yeah. Lindsey DeFalco: At least. Yeah. I mean, great. I think that's it. I think that was the podcast. So thank you, everyone, for it. No, I'm just kidding. ⁓ Yeah, like you mentioned, it's all of the kind of classic integrations between tech companies, ⁓ problems and questions just amplified. I feel like pre-AI, this would always come up with activity timelines or any sort of timelines. ⁓ And it's probably like the most rudimentary example of it, but the conversation was always like, everyone had a timeline and they wanted this timeline to have everything that was happening for, uh, you know, everything that had to do with that account, let's say for go-to-market tools. And at that point, the integration conversation was always like, okay, well, can you send us an event? And it's like, well, why don't you send us an event? Well, whose activity timeline do you want them to look at? and if you explode that out into what's going on now, it's like. Bob Moore: Yeah, yeah. Lindsey DeFalco: The existence of MCPs has just brought that, okay, well, where should people actually be doing the thing with the data question to the forefront? ⁓ And I also think we're seeing more of this with like doing any of this stuff as one question from one person at a company is different than doing all of this stuff at scale. So like if I go into Claude and I say, hey, tell me about this ⁓ account, tell me all the calls that happened, tell me all the emails that have happened. What are all the interactions? Tell me about the ecosystem. Tell me what feedback this customer has had. Like it'll go hit five different MCPs. It'll take at least a minute to come back. And that's like an acceptable experience for me. But I think what we're seeing is that people are now trying to like automate and do these things at scale. And that's not really an acceptable process at scale. So you can't go hit five MCPs and take. two minutes to return something for every single time you want to put together a brief about an account across your entire organization. Or if you want to turn that into a workflow, that's not an acceptable ⁓ both time for it to return, but also ⁓ amount of steps for it to take. And then people get into token usage with MCPs versus other. Bob Moore: Yeah, was going to say time is probably not perfectly correlated, but probably heavily correlated with a bunch of tokens being consumed and like more information than you really need changing hands and entering and exiting the context of like, cause you make an MCP call, you're talking to another, like you're, getting hit both ways, right? It's like you are an LLM talking to another LLM. It's like tokens in tokens out tokens sideways. all to probably have a imperfect exchange of information. That's like a superset of what you really need. Lindsey DeFalco: Exactly. Mm-hmm. Bob Moore: And it is expensive and that crunch is gonna bear down on us at some point here. Lindsey DeFalco: Yeah. Yeah. And there's been a whole like a narrative happening in the AI sphere. I don't know if it's the LinkedIn influencer sphere, whatever we want to call it. Hey, there we go. Yeah. Where people have been talking about like CLI versus MCP. And I've seen that like exact headline a bunch of different places. And I think that's like a Bob Moore: Let's call it the AI ecosystem. We'll try to get some SEO going for their, AEO going for this podcast. Lindsey DeFalco: a bit of the wrong dimension, but getting at the right thing, is CLI is command line interface, and then you've got your MCP. And so really what it gets down to is like, what is the right tool for the job? And I think that's a lot of what we're talking about here is like, MCP is just a tool and it does make integrations really easy to set up and do, but it's not the right tool for a lot of use cases. And so as people are trying to roll this stuff out, you know, at the enterprise level and at scale, the question comes like, ⁓ why do I need to use an MCP if I can just hit the API directly? It's a very deterministic thing. I know what I want. I get what I want. I don't have to blow through tokens to do it. And so I've seen that narrative pop up a lot more where it's like the hype cycle around the fact that ⁓ you can ask a question in two seconds with an MCP is great, but rolling something out at scale as an integration. It's not a magic wand or a magic bullet for it. And I think going back to the three point structure you're talking about is exactly that. It's like, is the tool that we should actually load all the data into a warehouse? And then the AI layer sits on top of that, or is the tool that we should use in MCP because way this certain information is going to be collected is going to be better suited for that. ⁓ then yeah, we get back to your question, which is like, okay, well, where should that happen? Which has kind of been like the, the SAS fight from the beginning of times is like, where do people work? Bob Moore: Yeah, I guess it is a little bit of that. I think that the thing that feels novel here is you could get away with not being the place where people work throughout a lot of the history of SaaS and still build what seemed like a decent business and even like use a really specific entry point to grab a foothold of users and then parlay that into some kind of like big enterprise play. like, this is... It still blows my mind that like, God bless them. Like DocuSign is like a giant company. Like e-signature feels like a wedge solution, right? And yet turns out it's like pretty darn complicated. And there are ways to make it very, like anything being deployed in a large enterprise environment suddenly adds all these interesting layers of complexity and customization and like a reason why somebody might spend millions of dollars a year in order to like have a. you know, organization wide global access to like a new signature company. and that's great. it feels like in this AI era, these wedge solutions are going to be easier to stand up actually. that's the problem. ⁓ like the inclination to build these things internally will be stronger. and the willingness to pay for external solutions will be less ⁓ and what you can potentially charge will be less as a result ⁓ and you can parlay that into a bigger play will be less if everything just ends up feeding into some LLM driven system ⁓ one way or another. like it does, I'm a very ⁓ data person. So it does make me think it all just all roads lead to data at the end of the day and ⁓ something There's a difference between something that inherently creates data and that is in fact a core asset. It's like an aggregator and system of record for data something that creates data exhaust, right? Like something like a call recording solution or like a Crossbeam. These are things that are basically like they are becoming these data archives for these very valuable ⁓ of actionable things or categories of knowledge. Something like e-signature. That's like a task that needs to be performed. It's a workflow, but it only creates data downstream as data exhaust, as people use the product. And it's really just data about the usage of that product. And yes, that data is ⁓ useful in various ways, but it's harder to come by and build and construct and all those things. So like the hope for a long tail of companies that ⁓ grow themselves up to some modest size and then find a home, ⁓ getting acquired by some bigger thing just feels very, very weird. And I know this from talking to friends who work in the venture capital industry that it's like, I met with a company this morning that went from like zero revenue to over a million in its first year with no outside funding and is getting passed on by ⁓ like the VCs because they're just worried that within a year or two, people will just. click a couple buttons and stand up what they do on their own. like a company going from zero to a million of ARR in its first year, know, five years ago, that's like, you're raising a series A or if not a series B on that, right? Yeah. Lindsey DeFalco: Do you think like in a weird way that the current climate is actually like strengthening the behemoths because there's all of sudden this like patience. There's like, I feel like there's more of an appetite for an appetite and optimism that larger organizations and maybe it's just the, you know, the big AI ones like open AI and thropic, but you know, let's toss like a Salesforce in there. It almost feels like there's like more ⁓ optimism that they can and will just create these things really quickly. And it's like, I'm thinking about if we look at, even just Crossbeam's customer base and ⁓ the feedback we get on what people are doing to orchestrate these, like AI operations internally. And it really is these, like a set of big companies ⁓ you know, people are still trying to do it with Salesforce. They're doing it with, Claude, doing it with Gemini, they're doing it with, you know, Glean. ⁓ And It's interesting because hearing you say that, it's like, OK, that like a dynamic shift from before where it feels like it leaned much more heavily towards startups being the ones that could obviously overtake these companies a lot faster? It's like has AI shifted that? Bob Moore: Yeah. Well, to hear you say that, does feel like there's two buckets. There's the giant incumbents where a bunch of operational infrastructure already lives that have these incredible deep moats just by virtue of entrenchment. And then there's the new players who are so close to the bone on what the new technology is. Basically the model providers, like it's opening I ⁓ DROPIC effectively, and Google. Lindsey DeFalco: Yeah. Yeah. Bob Moore: who are kind of intrinsically a very obvious place to be deploying new infrastructure to because no matter what happens in like the chaos of the rest of the stack of this market, like those companies are probably gonna gonna be there and gonna be continuing to grow. And like it's very, very expensive proposition enter into that market. And frankly, those those three or four companies that are doing it are consuming ⁓ last iota of compute ⁓ that comes off the assembly line. like, it's just like, feels unrealistic that other players are going to magically ⁓ appear unless they're very, very specialized. And that's not, you know, it's not what we're talking about here. So, you kind of, you have to pick like, do we keep throwing energy into ⁓ these places we're so entrenched that it just doesn't feel realistic that we will unwind it. or we can throw it into the new thing that will inevitably consume some of that entrenchment, but maybe not all of it, or maybe it'll work in tandem, but you don't really want to put that energy anywhere else. ⁓ So everybody else that used to be in all these like middle layers of SaaS ⁓ who have these value stories, as their value story emerges in this new paradigm, it just basically gets reduced down to being a data source. And to extent that there are still humans in the mix, certainly there's of action. ⁓ where humans do stuff still are, you know, have certain levels ⁓ of But it feels like if you ⁓ forecast enough into the future, those just taper down, down, down, down, down. we end up like, the game theory takes you to a place of just like, yeah, to your point, like massive, players who have amassed all of this authority and... kind of point solutions that are effectively data providers who are very, very good at ⁓ aggregating and being known as the owner of certain classes of data. ⁓ That frankly, even those like, while those may make it longer, those are not immune by any stretch. Cause once the low hanging fruit of like getting rid of all the workflows and all the human work and change the way work is done and all that stuff is like well trotted, you better believe that Lindsey DeFalco: Mm-hmm. Bob Moore: you know, the big players and the new players are going to start going after how do we become the owners of these data sets as well. And they're either going to do it through acquisition or they're going to go, they're going to go try to build stuff. Lindsey DeFalco: Yeah, it is interesting because you look at like, you you mentioned how well-entrenched these big players are. And it's like a lot of the AI tooling, at the model, what the companies that are creating the models are doing, plus what's available otherwise is really like sniffing away, if you think of it like a spider web of how well-entrenched they are, it's like sniffing away at some of the webs of what... hooks them into those companies. ⁓ And so I think there's a bunch of think pieces on how long a company like Salesforce has before they're not the full system of record as a CRM anymore. ⁓ But I think it's gonna... it's going to take that layer happening, like you said first, like all the workflows need to be taken over and then building an actual source of truth for something is easier. But it's really all the stuff that's on top of it that makes everything more complex. And that's what's, again, interesting to me about MCP specifically, because in a lot of way that just abstracts it. It abstracts that layer much more simply than something that... has more infrastructure built around it, like you need to build with APIs ⁓ and other mechanisms, but I think that's only true now. So I've thought, yeah, go ahead. Well, I was going to say we. Bob Moore: No, go ahead. Up, up, up, up, up, up, up, up, up, up, up, up, Let's make sure this gets left in. You go ahead. You go, you go. Lindsey DeFalco: I was just gonna say if we do the thought exercise that you kind of started this whole thing with of like, you know, if, at the end of the day, so it's the, there's this like source of truth, right? Everyone talks about context. And it's, you know, is that context stored in one place that's like a database or are you getting that context from, you know, other places via MCP or whatever, just grabbing content. Then you need to have this in a accessible and some sort of like one chat rules them all. ⁓ And I guess my question, like the thought, the immediate thing I go to is like, okay, well, so who should build that? Who should own it? And I guess after this discussion, my question is like, do you think there's room for anyone besides any of those big players to actually have something that people use or does it become I keep thinking of like Trey, right? Like Trey is kind of a ⁓ glue connects everything. It's like, is it this other ⁓ tool that sure some people use, really the core adoption is still going to end up in all the big players. Bob Moore: I think there is room. I think what's interesting is that the... As much as it feels like should be, there's not really network effects. If I ⁓ Anthropic as an example, the deployment of internally at Crossbeam ⁓ is ⁓ a of ⁓ connecting a of these systems of record where all of our data lives to Claude. ⁓ And the job that Anthropic has done in building kind of like onboarding and experience management and user access controls and user management and of their own marketplace of integrations and ⁓ like that in order to making it a really quick, seamless rollout to give access to Clog to everybody in our company and have them only have access to like the data that they're supposed to have access to. The fact that they own the model, like the fact that Opus ⁓ is anthropic model is kind of irrelevant. to doing all that stuff that I just described. Like they don't have to be the same company. And I think there's reasons why it's easier that they are the same company. There's reasons why they can like, you know, always maybe have an edge. ⁓ But you look at somebody like, if you go to the coding world, you look at cursor, which doesn't their own models. Like they're developing some and you could argue they're like doing some fine tuning and like downstream stuff, but they're not like. you know, they're not a research shop that's like making frontier models. but they basically have just like built a, a better mousetrap for being able to do coding that's powered by these models. And they have an enormous user base and huge amounts of revenue. And like, I think you could do that for enterprise deployment of kind of like chat bot and chat and agentic assistance for workforces net new, like someone, and I bet there's 50 companies building it right now. And we just don't know who any of them are, but like what People are always talking about like when we were this far into the internet revolution like Google hadn't even been founded yet, right? And there's a I think there's some real truth to that where I think there is room ⁓ I think there are like ambitious plays that could be made I think of slack, right? I think of like when slack came onto the scene Which I was like probably a decade into building companies at that point and we had like, you know, we had used Yammer in the past and like Lindsey DeFalco: Yeah. Bob Moore: Everything else was kind of like a bunch of email and all of sudden Slack was there and within a week it was like, how the hell do we ever run the business without this? Right. And there's like, I think there's some of those that are coming that will kind of take the world by storm. and that will lean on this stuff in ways that just kind of re-imagine the way that work gets done. actually, I actually think internal communication is like really, really ripe for it. I don't quite have a vision on like what that product is, but Lindsey DeFalco: Yeah. Mm-hmm. Bob Moore: The way in which we're kind of like, just like duct taping AI into Slack right now in ways that basically amount to like, it's a fake person that may or may not be in channels. It just feels secondary and bolted on and not like innovative in a way where if you built like the next generation of like internal communication and collaboration, that's a Gentic native that treats like ⁓ agents and LLMs as like first-class citizens inside of these. ⁓ like internal communication collaboration environments. Like somebody is going to build something like that. That's going to be an amazing, amazing business. That's super valuable. So yeah. Lindsey DeFalco: like an actual, ⁓ an actual fake employee that people don't know isn't a real employee that's like in Slack. Bob Moore: Yeah, well, think that would be like in there, right? But it's even more like, how do those fake employees interact with one another? And how do you do that in a way that's very transparent to everybody? Like the thing that our Claude instance doesn't give me that I wish I had was like, I kind of do want to see more about what all my colleagues are doing and what their chat conversations are. creating in terms of like net new knowledge and how they're fine-tuning all of their skills that they kind of keep private to their instance and like that collective Kind of consumption and collaboration the thing about Claude is you really don't get to work in public you like work inside of your silo and like you get some very top-down push skills inside your org, but slack allowed people to work in public and I think it broke down a lot of communication silos and other things and it led to better work getting done There's like some version of the way in which, you know, Claude for enterprise or Claude for teams has been built that is more collaborative and that does have like communication as a core crux. And that does allow like a more complex model than just like everyone has it or just you has it. ⁓ That I think it really beautiful and elegant and like changed the way humans work together. And I'm not seeing that out there in the market yet, but it's coming. Lindsey DeFalco: Yeah, it's like if you slap Slack on top of Claude and I'm like literally thinking about Claude's UI right now. Yeah, there's something that's yeah, there's something insane there. Yeah, it's funny with the. Bob Moore: I would go, I would stop. I would never log into Slack again if they just had, if they just layered in chat, like Slack quality chat into Claude, because all the integrations I care about in Slack are already now connected to Claude too. It's actually more of them. ⁓ And Claude's able to process that information in much better ways because of the LLM layer. Like everything in Slack is so analog and like linear and deterministic. ⁓ It's just like, when I'm over in Slack, I feel like I'm like living in Lindsey DeFalco: Mm-hmm. Bob Moore: and then I go over to Claude and I'm living in the future, but I'm alone. ⁓ Lindsey DeFalco: Yeah, actually, I had this this morning, I was like, wait a second, there's a Claude app for Slack and no one's using it. And it's this it actually came up because we have this channel where you can ask questions to product. And some people will say, hey, I asked Claude and it said this, but other people don't. And it just helps cut out me being like, hey, can you ask Claude if I can see that they asked Claude? But like you said, everyone's working in private. Bob Moore: Yeah. Lindsey DeFalco: I don't know if it told them something different because of what connections they have connected versus what I have connected and what it knows from me. and it's this whole thing. I was like, all right, we need to get Claude going in Slack. and like a, you know, a super kind of, brute force of, ⁓ of the two things that you're talking about. And it is, it is funny because it does create this cyclical thing again, which is kind of the pattern that we're talking about here, which is like, where do you do it? Do you do it in Slack and you're talking to Claude? Do you do it in Claude and you're talking to Slack? Bob Moore: Yeah. Lindsey DeFalco: because you can do both. And ⁓ a funny thing is I keep trying to test Claude's skill for answering product questions, and it keeps going to Slack and grabbing my answer to the product question and giving it back to me. And I'm like, no, don't do that. But that's like a... Bob Moore: Yeah. Yeah, it's a good like free, free, crazy company idea you could probably go raise venture on right now. Yeah, it's like everybody's putting Claude into Slack, but you should be putting Slack into Claude. ⁓ And like there's a, there is a there there. ⁓ There was this, ⁓ going to attribute this to ⁓ Andreessen or Zuckerberg or somebody, but like there's a quote about network effects, about like everybody loves network effects until they start to unwind. And like ask Friendster how they felt about their network effects everybody started moving over to Facebook, right? It's like the network that have built Slack into something. I feel like if you think about it our company, if all of a sudden like ⁓ five important people all started collaborating on an alternative to Slack that had like some very particular reasons why it was better and some really specific hooks to it that ⁓ got them excited. how quickly the entire company would end up moving over to whatever that new thing is, I think it would be scarily, scarily fast. Lindsey DeFalco: Yeah, there's a, mean, you know, even working on large projects in Claude and not having a good way to send that to people if you're not doing it in like the perfect right way that they support sharing ⁓ is a huge pain right now. So it does, it's like your Claude context becomes another layer of that context that we're talking about that is only valuable if it's shared with everyone. Bob Moore: Yeah. Whew. Maybe we should not release this episode and just go start building that thing. It's one of those things I just wanted as a customer. But you're right. It's different flavors of the same conversation, which is like, where the hell is all this stuff supposed to be happening? the reason it comes up so much for us is because we have an identity crisis as a company where like... Lindsey DeFalco: amazing. Yeah. Yeah. Bob Moore: our data is more valuable than ever. The market pull and demand for what we do is greater than ever. But if you take like a game theory view on it, I would argue that our, the breadth of the way we deliver value is narrower than ever. And it's like we're kind of becoming like ⁓ very, very highly concentrated value providing point data solution. Lindsey DeFalco: Mm-hmm. Mm-hmm. Bob Moore: Whereas I think we've kind of built the business up as like a fairly comprehensive, ⁓ UIs and workflows and integrations and like all of these things that are the extensions of the data that we provide and the ways that value gets realized. And like the stuff we built for people to consume our data and take action on it is very quickly becoming not the best way to consume our data and take action on it. In some ways it's the dream, right? It's like. hey, we don't have to build all this other crap. Like we can just do what we're here to do. But it's a little bit of like, hey, is 80 % of what we built historically, like basically doomed, which doesn't mean the company's doomed. It just means exceptionally large amounts of focus on a much narrower set than what we used to work on. Lindsey DeFalco: Yeah, yeah, and I think like you're seeing this pattern across a bunch of different companies in like how they're rolling out their pricing models, which is like ⁓ very good way to see where companies are going directionally. ⁓ But see all these like more ⁓ standard SaaS companies rolling out credits. ⁓ And that's really, to me, an indicator that, you know, they're ⁓ trying monetize people. ⁓ actually consuming the data more so than the more legacy seat model, which I think has been a big talking point about moving over to consumption or usage-based versus ⁓ just seats. it's definitely something that we're talking about, obviously, ⁓ what Crossbeam is seeing in terms of our direction. But I've seen that a of other similar tech where they're trying to this out in terms of how and where their data is valuable. Bob Moore: Yeah. Lindsey DeFalco: The thing I think is like interesting too though, cause as you're talking about that and you know, maybe 80 % of what we've already built in our experience isn't, you know, isn't, isn't useful anymore. It's like, I also think we are still in the industry over influencing how much people get it internally. And I think you'll see, you see this with plugins, right? The fact that plugins exist in Claude. Why do you need a plugin if everyone gets it? A plugin is a template of how to use, you know, one or more MCPs with some more context around it. Bob Moore: Yeah. Mm-hmm. Lindsey DeFalco: ⁓ and sure, it makes it easy to have things be repeatable, but I think I touched on this in a past episode, which is like, it's this, let's actually use Notion as an example, because Notion is a great example of this. they created this, ⁓ a world of point, like point solutions, they created this tool that's incredibly flexible to do any number of things across any number of personas. They very quickly ran into, ⁓ we are overestimating the fact that people understand how databases work ⁓ as. ⁓ lists and how databases relate to each other and the power that exists there. If you have a technology background in any way, it's incredibly obvious that what they did immediately was really, powerful. But it took another step of product thinking to get it to actual use cases. And so the next thing you saw Notion focusing on as a business was like rolling out all of these persona specific templates and creating a template marketplace and templatizing things to give people. ⁓ issue trackers and to-do lists and task management templates. And that's what I still think we're running into now. And I still have a lot of conversations on a daily, which is like, we're sitting here being like, we've got this great data, you can do whatever you want with it. But if we don't create the first most simple version of actually doing that, not everyone has ⁓ an ops team that is focused on creating AI workflows for them. And so we're in this funny point still where I think we still need to handhold people ⁓ more than like the linked influencer sphere tends to think that. Bob Moore: Yeah. Yeah, this is maybe my, my opening comment that like no one really knows what they're doing. It maybe is like a by-product of, ⁓ it, it feels like we think about these things in terms of how sophisticated companies are, but we don't talk about it in terms of how sophisticated like the median person at those companies is. And it seems like, at least in our experience, you know, we, get to interact very often with. Lindsey DeFalco: Yeah. Bob Moore: kind of line level execution focused people that are on sales teams, that are on partnership teams, that are on marketing teams. ⁓ And I feel like there's almost like three personas that I encounter very regularly. One of them is the performative CEO who like has a bunch of Mac minis running and is like toying around with open club, but not nothing that they're actually building. Hey, hey, hey, let's not get personal here. ⁓ There is like that zone, right? Where like people Lindsey DeFalco: Is that you? Bob Moore: that want to experiment and want to talk about it and have the privilege of ⁓ control over their schedules so they can carve things up and create time to go do that. I don't know that many those are actually transforming their businesses so much as tinkering. And ⁓ there's people that are actually transforming the businesses, which feels like it is a very elite squad of a low single digit percentage of people inside of these companies that are driving it. like huge amounts of transformative change and really, really, really get it. And they are the most important and valuable people at a given company in this era. And those are the people that ⁓ a real crisp clarity on how this stuff works, why it works, potentially have a vision around where it's going and selecting tools and migrating the way work gets done and all that stuff. And then there's everybody else, which... ⁓ By my count, it's like 90 % of the business. And what I think we encounter 90 % of the time, even at companies with incredible AI brands, even at companies that are AI native in nature that we interact with, they can't even hold like a discussion, a complete discussion end to end to envision the way to like make a product collaboration or a workflow or a potential like new ⁓ service or agent ⁓ ⁓ like they have pieces of it. They have pieces of vocabulary. They know how to maybe work in the areas that they've been trained in, but like we act like, because a company is really sophisticated or doing cool stuff with AI, inherently everybody who works there is. ⁓ I think that's really not the case. ⁓ and in a lot of these cases, you've one extreme of the other and they're both bad. you know, the one extreme is that people just haven't been. Lindsey DeFalco: Mm-hmm. Bob Moore: trained at all or are not intellectually curious enough to train themselves or not given the right access to be allowed to do that training or they're getting way too much and just go and you know, chew up a whole bunch of tokens building the same silly skill a hundred times over and over again. And if you fast forward three months, all that code is going to get thrown away. And yeah, so inside of these companies, it's just like, I don't feel like Lindsey DeFalco: Yeah. Bob Moore: The average human even working in the tech industry actually really had a click for them yet necessarily. I'm ⁓ proud of what we've done at Crossbeam because I feel like we're, you know, are a of people that I would put into that second category at our company at this point. And they kind of span a bunch of different operational areas, but it's been worked to do that. And we also have the privilege of being small, right? If we were 10,000 people that probably would feel next to impossible to pull off. the luxuries of having a small team. Lindsey DeFalco: Right. Well, I think that's like the trends you're seeing with the massive, massive companies that are going through their, they're doing huge layoffs is like, ⁓ my guess, I don't know, cause I'm not inside of them, but the, ⁓ energy capacity money that they have to get everyone up to the point, like get up to speed with where they want them to be ⁓ likely not worth the cost of that scale, which is like really depressing. ⁓ math that I think a bunch of these ⁓ massive organizations are doing. ⁓ Bob Moore: Yeah, there's this funny thing. You know, we have access. can look in, we can look in Claude at like the token consumption of everybody at the company. it's, it's funny because when people are at zero, like they're at zero and it's like, ⁓ you know, it's, it's kind of, ⁓ telling it is what it is. And then you get like a universe of people have like pretty, reasonable, clearly like using it all the time level token consumption. And then whenever we see somebody that's like, 10x that. But I don't immediately jump to, holy crap, this is a 10x performer who's like working 996 and like killing it. I'm always like, what stupid skill did this person write that's going out there and like re-downloading our entire Salesforce data set every single time it asks for like a deal update or something. I think it is more often than not, like this, yeah, this thing where you've almost over empowered, but under trained people. ⁓ And it just creates these situations where it's, you know, you're going to have lot to clean up later. Lindsey DeFalco: Oh yeah, it's like when you, it's the classic, know, enough to be dangerous situation. I maybe at one point during my engineering career accidentally restarted all of our production nodes at the same time and took down our platform. And I think that was a case of a probably had more power than knowledge I had. But yeah, I do, you know, with what you were saying about just the range and expertise that it always strikes me when I go on. Bob Moore: Nice. Lindsey DeFalco: LinkedIn. I know I keep bringing up these LinkedIn influencers but there's this like discourse right now that you don't know anything if you're not like fully automating your own personal life at home and if you don't have like 10 clod bots running your ⁓ life ⁓ ⁓ funny to me about that is like there's a lot of stuff that I want to keep analog ⁓ like I really like I like physical products I like my to-do list is a ⁓ card it's manual another interesting thing we could spend a whole ⁓ talking about probably is like what should remain analog that is not and what AI is doing for that. But yeah, think it's like even ⁓ think I see about that are like eight steps ahead of where most people actually are. ⁓ And they're if you're not fully architecting and hosting your own ⁓ applications ⁓ your at home, then like you're behind. ⁓ I'm ⁓ we take like 10 steps. ⁓ back from that and make sure that we're bringing everyone along. And I think seeing that kind of stuff is really intimidating for people that are still at a more baseline understanding of what AI can do. ⁓ that disparity right now is huge. Bob Moore: Yeah, yeah, is, you know, gets into the, like LinkedIn has the same perverse incentives that Instagram does to emotionally rile up the consumers of its content to make it more addictive and, you know, keep people engaged and on the site more. And I'm sure they're not dumb over there, right? Like there's, there's a reason why these ⁓ people like, I've never heard of and never interacted with and don't even necessarily know people who know them are showing up on my feed and on your feed. And yet posts from people that I know aren't. because like they do drive, you know, this particular type of engagement that I think is partly, there's like a FOMO component, there's a panic induction component, there's ⁓ you know, self value assessment thing going on. They're all like these very toxic. Lindsey DeFalco: Mm-hmm. Bob Moore: like emotional reactions that happen to these things, but it's basically there to make you feel like you are behind you are not enough. But yeah, like the more ⁓ time spend and like really understand what strategy feels like and what good strategy feels like in this space, the more, yeah, I like roll my eyes when I see that stuff. Lindsey DeFalco: Yeah. ⁓ yeah, for sure. And I think, you know, we've talked about this before. It's totally role dependent too. ⁓ there was a, there was a company you referenced at some point who had like everyone in the company go get an NN license and build a workflow. And it's like, do sellers really need to know how to do that? is it more for, you know, people on the, the like building side of the house where that's super valuable? I think the, same question comes up and, you know, in our, in our world with partnership teams, it's like, A lot of partnership folks aren't technical by nature. ⁓ And there is ⁓ this need right now, I think, at least to have like one tick more technical of knowledge is really beneficial to be able to know like what's possible in ⁓ like with AI tooling in partnerships in integrating ⁓ different products together. And so ⁓ there's, I think there's always just gonna be a bit of lag between something trending and something being actually known among all the people that need to know it. And that's where we are right now. Bob Moore: Yeah, and it's why this conversation cut down a video is gonna get 10 likes on LinkedIn instead of the 800 that we could be generating by saying you're in the Stone Age. Get your shit together. Lindsey DeFalco: Exactly. Yeah. Yeah. Pull up all your Mac mini Claude bot automations and make everyone feel like poop because they don't have them. So I am curious because we did, think we're ⁓ getting, near the end of this here, but like we started this whole thing talking about the, you know, who should be the system of record. You, ⁓ you Stitch, which That's like a bingo square. It's not the middle one like RJ metrics, but you did mention a past company. I do. Bob Moore: It's in the- if you're going for the diagonals it might be useful, Lindsey DeFalco: Yeah, yeah. It's one on there. joke for everyone that Bob can't go through a conversation without mentioning a past company that he founded. So ⁓ yeah, like, if we're talking about who's the system of record and then, you know, who has the chat abstraction layer. We talked a little bit about that and that there's room there. In my mind, I think I see a lot of this also just strengthening the whole data warehouse ETL. ⁓ pipeline area of business. like, I think a lot of things, like I mentioned, that people are doing with MCP right now, they'll actually realize that they can do cheaper, better, faster, stronger. There's a song in there somewhere. In a classic ⁓ data Yeah. ⁓ Bob Moore: Yeah, that's a great, I'm glad you brought this up because it is related and it is like, are twice in the last 24 hours this has come up. Cause you and I are talking to one of the big leading AI providers who we have a prospective partnership with about a really deep integration. And ⁓ solution they want is potentially not MCP at all. It's for us to ⁓ able to pipe data into their like native vector storage system. so that they can just query directly, not through MCPing us, which is like very intuitive now that they say it, but also like not something I had like seen a lot of before. And then this morning, you and Alex, our CTO sent over this memo about us investing in like the next generation of like data storage for like the Crossbeam data. ⁓ And... We're looking at one of these like heavily AI enabled, like vector store compatible solution, like next generation column or data store solutions. And it's like, ⁓ yeah. we can get at this knowledge and these insights way faster, even by bypassing our own MCP. Like our AI chat in our own product hits our MCP server. Right. Yeah. So like why. Lindsey DeFalco: Yeah. Mm-hmm. Yep. Bob Moore: ⁓ it's like, are, we are talking to our own LLM to get our own data, right? It's like, it's like, we're, ⁓ you know, bouncing a signal off of the moon, ⁓ to make it go next door. something be said there. And if we're talking about token consumption reduction and cost reduction and all that stuff, it's like, yeah, did we just end up, there is some version of the world where these, these data storage layers, ⁓ properly integrated with one another. Lindsey DeFalco: Yes. Bob Moore: are actually like a way, way more efficient way to do this stuff. And that's another area where like there's a lot of opportunity, or the million knockoffs ⁓ that are out there. Maybe ⁓ what should be on is that there's some like anti or counter MCP movement going on where it's actually, here's the technology that'll aggregate all this into like a really, really good. Vector store and just keep it on a streaming basis always totally up to date and have the results be lightning-fast and totally comprehensive and marginally free as opposed to like highly token consumptive Yeah, there's something there's something to that too Then I'm sure there's smart people working on but it's it's feels like it's around the corner Lindsey DeFalco: Yeah. And that's like very, it's very much kind of the center of the, you know, what I'm calling the CLI versus MCP, which I think is really more of like a API versus MCP discussion is like, ⁓ lot of people are using a tool for the job that is an AI tool right now. And then the more you get to actually scaling that and productizing it, especially at, you know, enterprise scale, I think we're going to end up back it a lot of. APIs, streaming data into warehouses, your standard kind of like ETL pipelines. Because really the value is all of that being aggregated in one place and not having to make 10 calls every time you want to get at it. Just kind of like classic product SAS architecture 101 questions. Bob Moore: Yeah, yeah, that is pretty wild. ⁓ The more things change, the more they stay the same. Lindsey DeFalco: Yeah, yeah. I'm also, feel like we should take bets at some point on ⁓ which of the current AI companies is going to become the AOL. And we're going to be like, ⁓ remember that? Maybe we'll take those not on our podcast. Bob Moore: Yeah. I got a few bets on that already. Yeah, we'll take those offline. Lindsey DeFalco: Awesome. Yeah. Bob Moore: Yeah, really true. Well, cool. I think we did it. ⁓ Hope somebody out there cares about all this. It sure is useful for us in congealing thoughts. And as always, appreciate the listenership and the promotions. Folks that like this, do us a solid. Give us a five-star review on Apple Music or Spotify or post on LinkedIn. We've been deliberately not really promoting this thing. And I think we're starting to pop our heads up a little bit. we'll keep it rolling and appreciate the support. Lindsey DeFalco: All right, I'll write a really controversial LinkedIn post about it to try and get some likes. Thanks. Bob Moore: Great, I look forward to being one of those two likes.