Latest / The AI Ecosystem Podcast / 23 New Examples of AI Innovation
Transcript
- Bob Moore: All right, everybody, welcome back to another star-studded episode of the AI Ecosystem Podcast. What stars you might ask? The same two stars you have every other episode. Myself and Lindsey DeFalco, Chief Product Officer here at Crossbeam. How you doing, Lindsey? Lindsey DeFalco: Great. You know, ready for another day, another podcast. It's going to be a good one. Bob Moore: Amazing. So we had an episode last time about AI hackathons. We were just on the cusp of throwing our first AI-themed hackathon here at Crossbeam after years of pre-AI era hackathoning. And now it's done and our minds are kind of blown. And we thought we might spend a little bit of time at the top here just going through what we saw happen in this thing. a couple of observations and then maybe walking through a couple of these projects that just, you know, some of them blew our mind a little bit and some of them shook to our core in ways that I think is awesome and also might ⁓ further accelerate decline of humanity into the matrix or wherever we're headed. How's that for an opening? ⁓ Maybe ⁓ one thing I gonna comment on and maybe ⁓ you can talk this, Lindsey, a little bit is, Lindsey DeFalco: But yeah, it's great. Bob Moore: Historically, our hackathons are very much the realm of product and engineering, and frankly, engineering. And we saw something very different this time. ⁓ you want to talk a little bit about what ⁓ up? Lindsey DeFalco: Yeah, so first off, there were 23 hackathon teams or hackathon presentation submissions. I think the most we've ever had in the past is like maybe 10, maybe like 12. Bob Moore: Yeah, I know in the most recent one, at which point in time, by the way, we had an almost identical head count. was probably about a year ago. I think we had eight. Yeah. And even back in ⁓ RJ Metrics days, there's a bingo card, ⁓ we did these hackathons religiously with much more ⁓ regularly scheduled cadences than we do them now at Crossbeam. It was every quarter, no matter what. And... Lindsey DeFalco: Yeah. Yeah. Yeah. Bob Moore: I think maybe 14 or 15 would be a ridiculous number of teams. So to have 23 teams submitting is unprecedented. Lindsey DeFalco: Yeah. And I think a lot of the energy around hackathons had gotten kind of like blah. And I think it's because there weren't a ton of submissions. There's that conundrum that I talked about last time where it feels like you're just being asked to work faster and harder for 24 hours, ⁓ doesn't really like sound fun all the time. And so this one felt like it was completely re-energized. ⁓ ⁓ think about half the teams were ⁓ ⁓ product ⁓ engineering. and there were members from the go-to-market team that were embedded in these other teams. And so that was like, I think we talked about wondering if that was going to happen. ⁓ And it did. There were a lot of folks who either by themselves or on teams with a couple other go-to-market folks, like put together full prospecting tools that they can use. And that's something that like we have not had happen before. Bob Moore: Mm-hmm. Yeah, don't think, like we've most certainly had salespeople, like quota carrying salespeople and like, ⁓ you know, leadership from the go-to-market or participate in hackathons. Absolutely. But it's almost always paired up with somebody that is in some kind of like a technology role or, you know, ⁓ least something like content marketing or product marketing. And they end up building some kind of collateral or new design system or something. ⁓ We had quota carrying sales reps three days before the end of the quarter enter in the hackathon, standalone single person team entries. And they were good. Like new skills in Claude, interesting kind of incremental efficiencies around how to collect and consolidate data. Like document builders, like all of these things that just made me smile ear to ear. ⁓ I, you know, we've never really been in that position before. It's crazy. Lindsey DeFalco: Yeah, think that's something that was different. I think one thing that was the same that I think was surprising that was the same ⁓ is that a lot of the projects still weren't in production at the end of the hackathon. Some were. ⁓ And I think, I don't know if that speaks more to our own internal ⁓ readiness on the product development front of being able to just push this stuff to production. Bob Moore: Mm-hmm. Lindsey DeFalco: But that I thought was like an interesting thing. And I think this will be very different when we do the next one in a quarter or six months. I think the more we do these, the more it's going to be like people are literally just pushing things directly to production. But yeah, I found that interesting. I did like a follow up poll of the the projects that we want to make sure we take over the finish line and are worked into our plans. And it felt really ⁓ think they're further along, ⁓ but in a lot of ways similar that. Bob Moore: Mm-hmm. Lindsey DeFalco: You know, it wasn't like, okay, we have these 12 new features that are live already. Although you did launch something live, so you're an exception there. Bob Moore: Yeah. ⁓ Yeah, you know, the more on the fringes of the core product you get, the more likely things went to end out into production. even the things that weren't like immediately available to our entire customer base got so much farther along and ⁓ so much more real than a lot of historical hackathon stuff has ⁓ So anyway, look, let's, I'm going to say some random team names. This will be a great pop quiz of your memory recall. because I'm going to need help with mine. I'll say the team name and do you want to try and recap of what that team worked on? And I can try and help plug in if we're struggling to remember because there are 23 of them. All right. Let's talk about the team whose name is team name. Lindsey DeFalco: Yeah, let's do it. the dark horse. ⁓ This is this one was hilarious. It was a one person team. They submitted their team as team name. And when I tell you that I went in with my expectations at zero, I was like, this is going to be some half ass project where, ⁓ you know, they just threw their name on the list at the end. And when I say this project, it blew my mind. And actually, should I spoiler and say that it was the winning project? Because I just. Bob Moore: Yeah. Yeah, this actually ended up being, we do winners based on the whole company votes on like a ranked choice voting system. So this was the, this was the. Lindsey DeFalco: Yeah. So what this project was, is essentially we have connectors for the Crossbeam MCP server with Claude and now OpenAI. So we've got one in ChatGPT and in Claude. We just launched that this week. You probably saw Bob's LinkedIn post about it. Super exciting. And so this project ⁓ used the ability to have like a light UI on top of the Crossbeam data show up in Claude. and in chat GPT. So essentially if you want to ask for, you know, what accounts are overlapping with this specific partner, then it'll return ⁓ cards that actually show the account with more of the metadata on the card. And so it was about applying the visual layer. ⁓ And the, those are like standard in chat GPT and Claude to our MCP server to have more of a visual presentation of the ⁓ of the MCP and it was really, really freaking cool. Yeah. Bob Moore: Yeah, this is like a classic, brilliant product management for a hackathon choice, because it's a surgical task. it is, it's, it's achievable in the timeframe of a hackathon. It has a visual component that for the human voters introduces some kind of like real wow factor. And it's directly related to like immediate strategic priorities of the business and moves them forward and like checking all three of those boxes. And, and by the way, it was so I think this is one that was launched into production at the time it was complete, wasn't it? Or is it, ⁓ it's close. ⁓ Lindsey DeFalco: It's close. It's close, but no, it needs a little bit ⁓ more love before it's live. Bob Moore: Yeah, so it's, ⁓ you know, the other thing that was kind of cool is I think the overwhelming majority of the company saw what this project was and said, I didn't know you could do that. And it really is like, you know, we were a company that hands people very, very curated specific contextual data in response to their queries, which is actually pretty good use case for like an MCP, except for the fact that when you're in the literal UI for ChatGPT and Claude the best it can do is send you a bulleted list or like a bunch of like line space separated data points or try to draw like an ASCII table with a bunch of vertical bars. project, ⁓ yeah, just turned it into like visual components that show a table or show like when the metric that matters better, it can be a little bit bigger and it can be inside of a box and it can have color to it. And it's not a crazy snappy overly engineered UI, but it's a way to organize. It's the, ⁓ The visual display of quantitative information ⁓ is intentionally done instead of just spitting it out as text. And it does really, really level up the experience you have of using RMCP through open it, through Chachi Vittier, through Claude. So that was a great one. Lindsey DeFalco: Yeah, and this, and that direction in general, I think is so interesting. And you can see Claude leaning into it more ⁓ recently with their launch of the artifacts, which are like the, ⁓ you know, mini UIs that are kept up to date and access, you know, one or more NCP servers and other ⁓ external data. And it is, it's fascinating to me because I think we're, I think we're in the middle of some of the swings we're going to see in terms of like the UI. versus data and UI and data and how people are using it. And this is such a good example of it because it's like, okay, we went all to text and everyone's like chats the new interface. And then they're like, wait, I miss charts and graphs. And so now we're pulling back charts and graphs. And then there's gonna be some balance there. think people are gonna be like, oh, it all looks the same. I don't like that it all looks the same. So then we're gonna introduce branding on top of it. And it's this like cyclical thing that if you kind of look around, we're like right in the middle of, but it feels like. Bob Moore: Yeah. Yeah. Lindsey DeFalco: All of the companies that are building these core models are kind of playing with what the value of the UI is. ⁓ And then everyone who has an MCP is also like, hey, we're able to actually create a UI now ⁓ through this tool that just spits out JSON, which is crazy. Bob Moore: Yeah, it's like a, I feel like Claude has had such success in the enterprise that it is probably now the scales have tipped where, nine months ago, the overwhelming majority of their users were probably coders using Claude code. They've been able to get such incredible adoption inside these enterprise contexts. And so many people using the native desktop apps and interacting through Claude cowork that the pendulum is kind of swinging back to. hey, can we be the glue layer that gets people out of their Salesforce UI and into just living inside the quote unquote chat environment. But sometimes there's visual components that exist for a reason and it's not just like design cludge. And they're trying to figure out these like choice decisions to glue that together. And it's kind of fun to watch. It's fun to participate in too. Admittedly, everybody says they're going headless, but the winner of our hackathon turned out to be the thing that actually took our headless functionality and like plopped a head on top of it, which is kind of funny. ⁓ Cool. Next up, I wanted to talk about truss, which also, ⁓ I don't have time for this, was basically a different flavor of the same project. Lindsey DeFalco: Yeah, yeah. Cool, yeah. Yeah, this one I think is probably the most interesting from a, you I mentioned not every project landed in production fully done, fully baked. ⁓ And a lot of the projects felt a lot like, you know, getting much further than you could previously get. And these two projects are asking the fundamental question of what if we don't? do anything and stuff is getting built behind the scenes. So Trust was about ⁓ essentially the, it was two people who are one's backend, one's really heavy DevOps. And it was all about getting our environment set up so that we can actually have autonomous coding agents. And then the, don't have time for this was our CTO who literally said, I don't have time for this hackathon. And he had ⁓ these, agents running during the hackathon to build something, but he didn't actually build anything during the hackathon himself. So these two are really, there were a few projects actually that I thought were like really undersold in how they were probably received by the broader company just because of the nature of the projects, which is kind of always true when you get into more technical ones. But yeah, the potential of trust here is like the... ultimate goal of where I think every team wants to get right now, which is that side by side of you have humans coding, but then you also have humans that are running all these other agents ⁓ that are doing all the coding autonomously. So that one was really exciting from a ⁓ strategic perspective. Bob Moore: Yeah, I get very excited about ⁓ this. I think this was actually the one that I put in the number one spot. And it did fairly well, but yeah, I don't even think it was in the top three. ⁓ But it got my number one vote because it felt like it was reflective of probably the most transformational thing for the company if it works and if it becomes real. you're pretty much like, we're using this tool called Linear Now. Lindsey DeFalco: No, I don't think so. Bob Moore: to track our product roadmap and kind of our sense of what we want to get done in the product and with what prioritization and organized by teams. yeah, you're basically building these things that would have been called Ralph loops like four months ago or something, which are kind of these autonomous loops where they go in, they look at what's there inside of linear. which thanks to linear itself being kind of LLM powered has likely been fleshed out in a fairly big way and in context of how our system works and everything. And it's able to, you know, consume that task or that ticket or that roadmap item that's in linear, maybe even have a conversation with linear, ⁓ you know, to kind of build it out even more and resolve any questions. And then it just opens up a pull request and builds it. ⁓ And probably loops through, you know, like, Other LLMs like cursorbugbot probably gets eyes on it and gives feedback and like what ends up landing in the engineers lap is not the linear ticket, but a fully fleshed out V1 implemented using kind of the best practices of our org. And then the engineer gets to apply taste and like very, very kind of elevated executive function style thinking about is this actually the right thing? Did it do it appropriately? And even if it's right 50 % of the time, You know, you're chewing up tokens in the back end, but those tokens would have been chewed up anyway by the engineer manually doing this work via Claude code once they picked up the ticket. But this thing can run way out ahead of everybody all the time and even do experimentation and, you know, pick up things that we normally wouldn't have been able to fit into a sprint. it's, yeah, I think, you know, this is the one that had everybody looking around really saying like, the nature of our job fundamentally changes if this becomes commonplace. And I think they were also looking around saying like, can't imagine a world where this doesn't become commonplace because it's already good enough to do it. The only thing missing is the glue. And like what these were was the glue ⁓ more so than the actual technical ability to write the code or audit it or anything. Lindsey DeFalco: Yeah, and this really is like, I mean, this is like, this is it, right? Like this is the thing that every... ⁓ Every company is asking itself right now is like, what do we do when, when this is the reality and how does it change how we work? And I think it's, I think we talked about this a little bit before, but it's fascinating to me because you look at these, ⁓ you know, AI native companies and even like the big model companies and like, they also have product design and engineering in some capacity. ⁓ And like those functions are not going away, but the way that they operate is radically changing. And so, you know, when you have the, thing that used to take the longest, which was generally. Bob Moore: Mm-hmm. Mm-hmm. Lindsey DeFalco: engineering in most cases, ⁓ and that's not the fastest thing, then how does that change how you get stuff done and your ability to get stuff done? And so even conversations we're having internally are like, hey, it's more important than ever for us to have, you know, like a well-documented long tail backlog of stuff for an agent to pick up and evaluate and work on. Whereas that, you know, isn't as much of a need when you know what your humans You know what humans are doing, you know how much time they have, you know how much capacity they have. ⁓ And that's just like one small example of how it changes how you think about stuff. And I think there's ways that just will radically shift and is already radically shifting how people iterate on feedback that they get. Because it's such a good candidate for iterating on like the insane amount of feedback that comes in from customers on ⁓ any product on a daily basis. Bob Moore: Yeah, so cool. Alright, I wanna, there's three more that like I can't ignore. And they're so different and also uniquely interesting. The first one ⁓ is Match. And there's like a specific reason I wanna talk about this one, but I wonder if yours is the same as mine. Do you remember what this one was? Lindsey DeFalco: ⁓ I do, but you're so excited about it. feel like I want you to give the description on this one. Bob Moore: Here, so, all right, fine. I'll take that self-inflicted tea up. This was a team we, okay, if you think about how Crossbeam works, we help companies compare their data to each other and find what's at the intersection of that data or what's not at the intersection of that data. And you can be on Salesforce and I can be on HubSpot and our third partner can have a Snowflake data warehouse and somebody else can upload a CSV and Crossbeam has to figure out are we all selling to the same company at the same time? Or am I selling to Delta Van Lines and Lindsey's selling to Delta Faucets our other partner selling to Delta Dental and the other one selling to Delta Airlines, right? ⁓ And this presents a very, very tricky problem at the core of our intellectual property, which is solved by our matching algorithm or matching engine, we should say, ⁓ which is this very ⁓ clever, ⁓ ⁓ complex ever evolving logic system for determining whether or not two companies are actually the same, even if they come from very different ⁓ provenance. So ⁓ that system, we fine tune the crap out of it and we've been fine tuning it more than ever recently thanks to various... ⁓ growth of the company and like the training data that we can pull in and us expanding into new industries which give us new like matching dimensions we can match on and all those things. But it's a lot of work and like a lot of these theories that might exist as to how to make the matching algorithm better, they're risky because you worry a lot about false positives and false negatives and if you tweak the matching algorithm in one way to fix one weird corner case issue where we're missing a match, it might have these ripple effects where over in some other corner in some other account, you're accidentally saying that things are a match that aren't. And we always kind of have to strike this balance. What this team did is it built an engine through which it could take enormous amounts of real world kind of anonymized aggregated like safe sandbox playground data that's reflective of the massive, massive superset of the data that exists inside the greater Crossbeam world. And it could actually like up and run ⁓ theories of how to modify the matching algorithm against that real world data ⁓ ⁓ LLM's ⁓ and expectations about what the right and wrong answer is to actually ⁓ ⁓ statistically compute whether or not certain improvements or changes to the matching algorithm code resulted in better outcomes universally. And previously to do this, was a total nightmare. And there was always a little bit of this like, okay, we think we like this and we've done some one-off testing and we're flip this live in production. This system, what I like about it and why I wanna call it out is that... we cannot use LLMs in matching. We cannot ask every single match like, hey, this person says Delta and this one says Delta.com, is it the same? It would be so expensive. We would like, you know, consume every GPU on the planet every day or something to write, like there's trillions of these, it's like an N squared problem ⁓ inside of Crossbeam ⁓ to do that. So it has to be deterministic. But what the LLMs allow us to do is build Lindsey DeFalco: Thank Bob Moore: ⁓ like the framing, the theory construction, and like the test data sets and the test data expectations in such a way that we can really, really rapidly iterate through different versions of deterministic code to make it really strong. And this is where like LLMs as rails, as an accelerator to make extremely effective non-LLM powered in production day-to-day deterministic code becomes this incredible, incredible superpower. And this is the thing that I think people that get it, get it. And I think it's being deployed all over. I think it's underappreciated. When you talk about AI native companies, yes, you can build a product that's just like a wrapper on top of an LLM or that is delivering value directly because of an LLM. You can also build companies better by using LLMs in the development process to build wildly more deterministic outputs that maybe humans would have never been able to construct themselves. But then those outputs actually run deterministically and very predictably and therefore very cheaply like our matching algorithm has to. And so this project, I picked it and it was in my top votes partially because I love the matching algorithm and it actually does like seem to be a really effective thing that'll help there but more for what it represents which is bringing LLMs into the iterative process of how to create the non-LLM powered parts of our code. Very, very powerful. Lindsey DeFalco: Yeah, I knew you'd crush that description. You have almost as much passion about the matching algorithm as you do about RJ. Bob Moore: It is a real, you know, it's a real, it's a real battle. Lindsey DeFalco: Yeah, my favorite thing about that project is exactly what you said about how like the big question of like determinism versus when to use an LLM. And so many people just like throw AI at problems that don't need AI. And I love the approach that this person took because I do think it was so well thought out in terms of exactly where we can get the most ⁓ out of AI versus, know, ⁓ this entire thing needs to be redone ⁓ with AI. ⁓ And that's like, I mean, that's really the question a ton of people are ⁓ running into. Even in my hackathon project, actually, I hit some things where I was like, ⁓ I didn't need AI for that. It was like fun to figure it out with AI. But then once you get into it, it's like, ⁓ okay, this is just, you know, ⁓ an if then statement, or it's just like a really, really deterministic ⁓ workflow. ⁓ read this blog post, it was talking about the different types of agents. And this is relevant to this because it was like the ⁓ first type of agent, and it was like everyone should start with this, is really a deterministic workflow with AI embedded in some part of the ⁓ workflow or the process. And that most things that you think you need a level two agent for, or something that's more complex and is doing more of its own reasoning, you really don't. figure it out and once you map it out most things that we're trying to apply you know agentic like ⁓ building agents for ⁓ just end up being these deterministic ⁓ workflows with AI in one part. And that's a great example because it's exactly what the hackathon project was. It was like, OK, we have this deterministic process and workflow. And then there's certain parts along the process where we want to pause, ⁓ have the LLM do some stuff, and then come back and then continue on in the workflow. Bob Moore: Yeah, that makes perfect sense. ⁓ I think the best evidence of that is like how frequently the LLMs themselves want to hand you deterministic code and not actually incorporate LLMs into the solution. Like they'll want to use LLMs to figure out how to get at the solution. ⁓ And in some cases, like surgically insert calls to LLMs on Lindsey DeFalco: Yeah. Bob Moore: like when all other possible routes are exhausted, but it never starts with just like, why would you build a time tracking app when you could just like make an LLM call to say, hey, start tracking my time and then hey, stop tracking my time. Like they know that the deterministic code is the more like production ready, reliable thing. ⁓ And it's like these leverage points are not all just let's slap something on top of an LLM and run it that way. It's like. Let's re-envision how to build a very reliable thing way more cheaply and way better because LLMs are kind of playing quarterback. ⁓ Lindsey DeFalco: Yeah, well, it's funny how much stuff like, you we have our, we have the MCP server at Crossbeam. And when you, when you talk to people about how they want to use it, ⁓ so much of what people want to do is just use a web hook and have a workflow go off of the web hook. Like no. Bob Moore: Yeah. Lindsey DeFalco: AI whatsoever needed in the whole process. Maybe in how they tailor the message at the end of the workflow. I know we talked last time a little bit about the whole MCP versus, are you just calling it API through a CLI? And it's that same thing. It's like people are discovering these use cases that have been possible for years. It's just that how technical they were, we're out of reach. Bob Moore: Yeah. I love, yeah, like headless is like such a hot term right now. And there's so many people out there using it ⁓ like it's new. ⁓ Like headless has been around since like the 90s. It's like a paradigm around like communication interfaces and yeah, it's pretty funny. Cool. ⁓ go ahead. Lindsey DeFalco: Yeah, yeah. Yeah, there was a, I was going say headless was trendy. Like, ⁓ think somewhere around like 15 years ago or something, just talking about front end, ⁓ like testing deployments. ⁓ that was like the last time I remember it being super, super trendy. And I think even then it was old and borrowed from the nineties. Bob Moore: ⁓ Yeah, like unit, like integration testing when you could like, not like a browser in code without ever seeing it, like headless integration thing. And then headless CMS's became really big. Like, Hey, went on a blog, I'm going to buy a headless CMS. And like, it's not going to be prescriptive about the front end. It's just going to be like content organization. We had one of those at Stitch headless CMS. Yeah. What's old is new again. Speaking of what's old is new again. Lindsey DeFalco: Yeah. headless. Yeah. Bob Moore: I want you to talk about interface. Lindsey DeFalco: I am so excited about Interface. there are a few folks from the ops and product within the growth area of product team set out to build an internal analytics and BI tool. So this has been a... This has been very, very top of mind lately for a few reasons. And a lot of it has to do with what AI is enabling people to do in, I'll keep using Claude as our example here. So. Now that you can just query Claude and ask questions about analytics and get answers back, it begs the question, what do you need BI tool for? And then generally the answer is, okay, well, we have all these dashboards that are stored in this tool and we want to share them, we want to monitor them, we want to track them, we want to go back to them. ⁓ you end up with this kind of like split of use cases where you have ⁓ things, ⁓ you just want to get at query time. You want to go into Claude and you want to say, ⁓ are our monthly active users right now? ⁓ But then... there's the cases where you're putting together a board presentation, you wanna have a chart, maybe you're monitoring something that you wanna go check day over day over day and you wanna have a more persistent view for it. So we currently have a very outdated solution for our charting essentially in our dashboards. ⁓ it's been something that has just felt like too big of a lift to get off of. And at the same time, we've spent a lot of time looking at a lot more of the newer solutions where ⁓ there's ⁓ approaches to the kind of BI process. problem now, maybe it's all your data lives in a data warehouse and there's an AI layer and a charting layer on top of it and you get a bit of the dashboard, a bit of the query, the querying through AI. anyway, we, then also this has always been kind of a pet peeve of mine, but I get it. Every single tool that is built for a product team, they try and do product analytics and it's never connected to the core analytics system at your company. And so therefore it only has a value and it's never been something that's really like complete enough. So very long intro to say this hackathon team was like we're just gonna build an internal analytics and BI tool and it was so... ⁓ freaking cool. I don't think I can drop the F-bomb because there's probably restrictions around that, so I won't. But it was so incredibly badass. And what they were able to achieve in a 24, 48-hour time period was crazy. they essentially presented this full internal analytics tool where you can ask questions. So they built a semantic layer on top of the data so that it can actually make sense of all of our data that's stored in ⁓ our warehouse. And then ⁓ they also built in this charting. And so it's achieving both, where you can ⁓ ask natural language questions about the data, get back responses, but you can also build these dashboards. And it's in beta right now ⁓ with our internal team, but it's one of those like, holy crap. Can this work if we build it internally for ourselves tool where you would never two years ago say let's build our own internal BI tool? Like that would be insane. That would be an absolutely nuts undertaking. And it feels really, really close to reality. Maybe I'll eat those words in six months, but ⁓ maybe not. Bob Moore: Yeah, this is where every, you know, every old friend of mine from my business intelligence days is like, ⁓ you know, they're still with us thankfully, but they're rolling over in their graves, right? Like listening to all this, because it sounds like all of the tropes that ⁓ should not be true and should be like mockable and eye roll-y. Lindsey DeFalco: Yeah. Bob Moore: about when people say, I'm just gonna build my own like dashboarding and BI tool. Like I don't need to spend, you know, 50 or a hundred grand a year for this, like this layer where we just have dashboards and charts and people can collaborate on them and whatever. I, like hurts my soul a little bit to say this, but like, I think in the span of that, 36 hours of the hackathon, they built a better product than RJ Metrics was after like eight years of working on it. ⁓ Like really at its core and look, our data stack, have a Snowflake as a data warehouse and then there's like a pretty well managed DBT layer sitting on top of that. So there's some like semantic organization of the data. ⁓ But what these guys built, Lindsey mentioned that semantic layer, it's kind of a translation of the DBT views into business logic and business concepts. And then, like this UI around being able to create and organize charts and then dashboards of those charts and then, figure out the sharing and communication and being able to construct those things ⁓ through a interface. But also, ⁓ if you need like doubling back to what the actual sequel is that's being written. ⁓ I think the ability for it to be like a hot, swappable replacement for our current BI solution, which granted is one that is ⁓ not the bleeding edge, like most innovative on the market these days. But ⁓ yeah, but it's still, you know, it's still a product that is functional and that kind of serves a purpose. I think they built something that's gonna be able to like hot swap out of that. And now the big question for us ⁓ ⁓ know, ⁓ do we hit limits on this thing and what's the maintenance burden and the overhead associated with keeping it running and keeping it alive. But the reality is it's not Lindsey DeFalco: That's an understatement. Bob Moore: because the real workhorse here is the data warehouse that sits underneath of it. And really like the glue that exists inside the semantic layer is what's really important. In a lot of ways, this approach that it's taken, like you almost want this layer to be intentionally thin. It's really just like organizing probably the most like downstream concepts ultimately that are the interface by which business users get at the data. But you also want to be able to like, If something changes in the semantic layer, you want it to be able to ripple through really effectively globally across all the charts that exist so stuff doesn't get out of date. And that's where the LLMs are really going to shine. sure are ⁓ BI solution companies out there way better products than we're able to build. And I think ⁓ at certain scales, it'll make sense to use one of those. But compared to where we are today, And in terms of like our direct immediate needs as it relates to this stuff, this is like kind of functionally awesome. And the maintenance burden doesn't feel very high and it looks really cool. And it's like live and deployed. I was, I think this team got second place, right? With the popular vote, yeah. ⁓ Lindsey DeFalco: Yep. Yeah. Yeah. Bob Moore: challenges my core values and beliefs around like, should you just build your own thing to replace all your SaaS tools? ⁓ I have a very allergic reaction to that idea. And then somebody went and did it and I poked and prodded at it a bunch. And I was like, damn, this thing might just work. ⁓ work for us for now. ⁓ the now ⁓ be a year, might be two years. And like, that's a material amount of economic value saved and value delivered, ⁓ which is pretty, yeah. Lindsey DeFalco: Yeah. Yeah, I actually, ⁓ this just reminded me of, ⁓ this is gonna be like a completely downer take on your take because I share your sentiment about that project and it being right for us right now. ⁓ But I was ⁓ listening to a podcast and ⁓ there was a guy saying that we're gonna have a challenger moment with AI in the near future, which is just a ⁓ a terrible, sad analogy, but it's essentially talking about how. Bob Moore: pretty awesome. ⁓ it's in like the Challenger shuttle. All right, go on. Lindsey DeFalco: Yes. Yeah. Which is, but I liked how he described it because it was basically like, everyone knew that the O-rings were faulty, but they kept having successful launches enough that people gained confidence in it. And so ⁓ that comes to mind here, but it's because it's like the more that we're building ⁓ these tools, the more we're like, okay, this works, this works. And then at some point there's going to be some catastrophic failure and maybe it's a, Bob Moore: ⁓ yeah, yeah. Lindsey DeFalco: security ⁓ incident with a capital I at some large company. Maybe it's, you know, I don't know, worse than that. ⁓ it's this interesting thing of like, we're currently kind of playing with like how much we trust AI to build out these different tools and for what purposes. And, ⁓ know, it's right for us right now. But, maybe it's not later and bit of a bit of a dooms take it. But it is interesting because we're like this hackathon is kind of like pushing the boundaries of our trust of how we want to use it internally. Bob Moore: Yeah, it does become very much like a mission critical system. there are probably very important questions that we want to ask around the pipeline for bringing things to life inside of that. But again, interestingly here, at the end of the day, what they shipped is a bunch of deterministic code. It's a bunch of Python and JavaScript. And we have great engineers that can code review it and get it to a point of you know, at least ensuring that there's not some extremely dangerous like prompt injection ⁓ issues or other things that might exist there. And we can even use LLM tools to do pretty exhaustive security tests against it, at least for that kind of stuff. ⁓ So it's not like we're leaking customer data out to some kind of like LLM thinker, like no LLM had to see any customer data or even our own internal like private data in order to build this thing. And that's what's pretty wild about it. And then it just becomes this layer that runs in a deterministic way without any hooks into the LLMs besides like, you know, the kind of chat interface to create charts. that's interface. And what's that? Lindsey DeFalco: Can I segue to one that I don't know that you're going to touch on, but I want to? Bob Moore: ⁓ great. Yeah, yeah, yeah. I was going to go to, these go to 11 and that was going to be my last one. So if it's not that one, ⁓ segue away. Lindsey DeFalco: do it. Yeah, I'll do a quick before we go to the last one, which is speaking of ⁓ hackathons that rebuilt an entire company and hackathon. I was very tickled by another one of the projects called Newsbeam. Because I used to work at a ⁓ get bingo board out for me. ⁓ Yeah, I used to work at a company called TrendKite ⁓ got acquired by Cision and it essentially did PR analytics. So it was like the Bob Moore: Yeah. This is right up your alley, yeah. Lindsey DeFalco: It scraped news sources from across the entire web. It found company mentions and then, you know, it told you how much your company was getting mentioned in the news, how much of that is owned or earned media. You know, some attribution sprinkled in there, blah, blah, blah. So I see this hackathon list come in and there's a team and it's like, news Crossbeam. We want to find company mentions in the news and then, you know, do 10 other things, like kick off all these other all these other workflows about things that are relevant to the thing that happened in the news. And my immediate comment was like, ⁓ my God, I love this. You're going to recreate the entire company that I previously built in a hackathon time period. And these two are such good examples of how that's, I mean, that's literally what's happening to companies ⁓ right now ⁓ ⁓ time. But I got a really big kick out of that. And ⁓ they ended up making something pretty cool. I think would probably get a lot of people like 80, 90 % of the way there from, you know, the product that we built and sold 10 years ago now. Bob Moore: Yeah. Unbelievable. It is really funny, isn't it? It's like your last company and my last company both basically got recreated from scratch in 24 hours in this hackathon. guess as ⁓ you get older, or maybe I'm only one getting older, don't know, as ⁓ you get older in this business, it's you live long enough. to see your life's work become meaningless or like recreatable in the blink of an eye. And it's like a very interesting fact of life about living in a time of such incredible technological progress condensed so short. It's a weird feeling. I'm less depressed by it than I should be on paper. Yeah, it's like there go my 20s in one 24-hour hackathon. Lindsey DeFalco: Yeah. Yeah, cause Yeah, well, it's crazy because like the whole, you know, the whole value prop of a TrendKite was like that people were physically cutting these news mentions out of newspapers and magazines, pasting them into books and then dropping them on a desk. And there was like a thud factor of how heavy the desk was. ⁓ we're like, guess what? We can do that on the Internet. ⁓ now, you know, 10 later, it's like, guess what? We can do that in 24 hours. Bob Moore: Okay, now you sound old. I was trying to protect you before, but I think you've crossed the line when the scissors came into play. Yeah, that's incredible. All right, before we to this last project, I'm gonna just like speed round down the other teams without double clicking on them, just to give y'all a sense of the areas of coverage. yeah, so we had one... Lindsey DeFalco: Yeah. Yeah, this hat's covering my graze. Bob Moore: ⁓ about enterprise outbound messaging, where this is like a solo member of the go-to-market team built an engine to create highly customized outreach to enterprise prospects. ⁓ had this one, ⁓ that was like technographics visualization. So like being able to look on like a timeline view of like what tools, what companies are using over time ⁓ and view that in like a beautiful visual way. It's pretty neat. One related to Financial forecasting and being able to predict where we're gonna land at the end of a quarter based on a huge comprehensive universe of signals coming in from all our different products is super cool. One related to ⁓ an centralized interface for ⁓ subscribing unsubscribing from notifications in Crossbeam because there's like marketing email lists, there's system updates, and then there's like alerts you can have from lists that you build. And those are all in different spots. ⁓ And they centralized all that into one zone in the UI, which was really cool. ⁓ That got third. Yeah, think great product market fit with the audience. Because I think every customer success rep jumped on board like this is life altering. Because that's like a death by a thousand cuts about Crossbeam prior to this improvement is like you can think you unsubscribed and then the emails keep coming. Lindsey DeFalco: That got third, by the way. People always love it. Bob Moore: There was an interface for managing feature flags more efficiently and doing it through Claude. ⁓ this talk about headless, right? This is like administering our client accounts with ⁓ MCP commands. So not just reading data, but actually doing some writing to turn feature flags on and off and check statuses through Claude. It's very cool. Do ⁓ you remember what overlap and close was? This is Reese's project. Lindsey DeFalco: It was either a Claude skill or an n8n workflow that was getting triggered from Claude that was about essentially actioning on partner signals. Bob Moore: yeah, partner signals, that's right. ⁓ Like translating signals into like actions and alerts and stuff. How about what we call chaos is just patterns we haven't recognized. Johnny. Lindsey DeFalco: Ooh, which one? ⁓ that one. Okay, this one actually, think ⁓ this poor team suffered from a like hiccup. Your classic when you demo it, breaks. But this one was essentially ⁓ creating hoc visualization from the MCP server. So. ⁓ Bob Moore: ⁓ right. Yeah, yeah, yeah, yeah. They had that like spider graph thing. Yeah. Lindsey DeFalco: Yeah. So you can apply any sort of visual to the data that's coming out of the MCP server, which is like a really, really cool use case for being able to create custom charts and dashboards in product by just having a conversation with the AI chat or having something more custom that comes out of something like a Claude or a chat GPT, because it's like generating the visual in a mini UI for you. really, from a potential standpoint, like very exciting. Bob Moore: Very cool. There was one called ProRate, is a credit by our legal and finance teams for when contracts were new early ⁓ or have an upsell on a contract, there's this complicated map that has to happen to prorate the existing contract term and then incorporate it into the new contract to like credit people appropriately. ⁓ And it's pain in the butt. And our finance team usually has to do like ad hoc work ⁓ in Excel. is like really powerful. calculator that runs just as like a, I think it's like an HTML powered, just like in browser calculation engine. ⁓ And it is, talk about using LLMs to build something incredibly precise ⁓ like unit tested out the wazoo. ⁓ it's gonna save so much time and allow our refs to self serve in a lot of ways on creating order forms and stuff ⁓ contract prorations. So good to solve, good one. TNE or to not TNE, this was some folks from kind of our people and culture team trying to build some skills related to helping people book their work travel. We do a lot of off-sites and almost everybody at the company is on planes at some point during the year. This one didn't get as far as they wanted because of limitations in the MCPs of a ⁓ of the travel providers, but it'll ⁓ there. ⁓ Deal Brain, ⁓ remember that one? Lindsey DeFalco: That's a good one. That was a good You know what's funny is I remember the AI generated graphic that they used to advertise the team and now I'm blanking on the rest of it. Bob Moore: But. ⁓ Yeah, Deal brain, probably It was, was was ops, it was Collette and Greg too. ⁓ I'll say that it probably has do with what it sounds like, ⁓ which is like very, very strong around deals, ⁓ my project ⁓ called Like A Look, I built an to create lookalike. audiences and populations based on lists of companies. And that was using a mixture of technographic and firmographic like third party data signals that we aggregate. And the idea here is ⁓ try and allow companies on Crossbeam to have a prospect list, even if they don't have a CRM they can connect or they haven't built their own prospect list yet. ⁓ We could like intelligently build a prospect target population for them based on their target market, their ICP. ⁓ customers of their partners and things like that kind of an initial foray into dipping your toe into Crossbeam ⁓ More to come on ⁓ We talked about a lot. Monday Morning Quarterback was another go-to-market team member who built like a thing that shows up in his inbox every Monday morning that's like a huge rundown on everything that happened the last week with his team and all the deals moving forward and everything, which ⁓ we've seen in various forums, but it was cool to... Kind of seeing pieced together some some extra insights from new sources there I think we're coming up on it letter rip remember what that is Lindsey DeFalco: ⁓ that was like a sleeper project. So this was a team ⁓ that took on improvements for ⁓ accessing Crossbeam data programmatically. So getting data out of our API, ⁓ understanding. Yeah. Yeah. So it covered API documentation. It covered an easier way to authenticate with the API. Bob Moore: ⁓ this one was great. ⁓ my god, yeah. How do we not talk about this one? Yeah. Lindsey DeFalco: It covered adding additional data to endpoints in the API. And it was so casually presented. I was absolutely dying the whole time they were presenting it because somebody would come on and be like, we completely changed how API authentication works. And I'm like, this is huge. This is another one that's really way, way closer to, I think some of it's actually already in production. Bob Moore: Ha ha ha ha. Yeah. Lindsey DeFalco: and the rest of it's like really close to being in production. There was way more progress than we would have previously seen. Bob Moore: Such a game changer too. Like the amount of people trying to build things against our APIs at this point is just like really, really radically growing. And I'll name another project out of order here that's connected to this same class of problems, which is Crossbeam sandboxes, ⁓ which it's annoying and challenging to create like developer sandboxes of Crossbeam with real data in it because unlike other like... Lindsey DeFalco: ⁓ Bob Moore: tech tools or SaaS products where you can just like stand up an account and put some fake data in it. Crossbeam, it's not just about your account, it's about the partner accounts you're connected to. And like you need data and they need data and a subset of each of those data sets has to overlap and what overlaps has to be meaningful and interesting. like, so standing up like a sandbox environment is actually standing up like five or six distinct sandbox environments and having this connective tissue between them. And there has to be qualities to the data. in each of the sandboxes that have them kind of rhyme with each other enough to make the Crossbeam instance interesting enough that you can build stuff on top of it. And it's a massive pain to stand these up for our customers and somebody build a hackathon project that like does it with a click. Like you click a button and it just kind of like spins these puppies up. Yeah, that was a great one. Lindsey DeFalco: Yeah, and that set of projects. mean, if you talk about, we talked about last time a little bit how like just two companies having MCPs means that there's all of sudden a use case and this kind of like integration that didn't exist before. But all the projects we just mentioned completely changed the game for what's possible with partners spinning up integrations really quickly. I think this is something that as everyone's investing in building an MCP, it's also like, don't sleep on your API and don't sleep on the fact that companies will want to spin up an environment really quickly so that they can test MCP, API, whatever. And so I think from a strategic standpoint, this set of work is really interesting because the partner developer experience will be more real than ever, the more integrations are possible. And once people see what's possible at the MCP, then there's going to be more people wanting to go deeper and see what's possible with API or more native integrations. So from like a thematic standpoint of what I think is going to happen in the market too with integrations, it's a great set of ⁓ projects. Bob Moore: Yeah, if we measure our success in this upcoming era by how much of our data gets consumed, you might think that we want all of our data flowing out through our MCP. I don't care. ⁓ If it's somebody else's MCP to some other product, but our data becomes essential to that decision-making, that MCP layer might choose to ingest our data through our APIs. And like, great. So if we see 10x the volume of data flying out of our APIs because somebody else's MCP needs it, more power to them. And I think like, yeah, don't sleep on your APIs because that might actually be the real power where the real power sits. ⁓ Lindsey DeFalco: Yeah. Or your data warehouse integrations for that matter. Bob Moore: Totally, totally. And last up before our ⁓ de resistance here ⁓ was Ecosystem Intelligence, which was a project about creating a visually beautiful, ⁓ ROI case for using Crossbeam. So ⁓ able create almost ⁓ like a magazine style, newspaper style read in a single page webpage for any customer at any time about like the ROI. Crossbeam which what we could do in Crossbeam if they're not a customer yet which is also cool ⁓ really neat stuff all right whoo let's talk about ⁓ turn up to 11 and how I'm still undergoing existential crisis because of it Lindsey DeFalco: ⁓ yeah, okay, so we saved this project for last. It was my project ⁓ along with, it was partly my project. I was on the team ⁓ along with ⁓ an engineer on our team and then also somebody from the go-to-market team. But we wanted to play with ⁓ voice and audio. And so we used 11 labs and there's more of an idea here, which is a bit meta to this podcast, which I'll get into as well. ⁓ But the first step of all of this was we want ⁓ to see how good this voice imitation has gotten. So we ⁓ Bob Moore's voice ⁓ I will play some clips for you, don't you worry. But the ⁓ thing we wanted to test here is that especially with how easy it is to turn audio into ⁓ text. and audio transcripts into text that reads like a normal blog post or whatever format you want it to be in. The same is true about turning text into audio now. And so, I don't know, if you ask everyone at any given company, I think maybe half people prefer to read things and half of people prefer to listen to things. And I ran into this because I think I mentioned. Lenny's probably too much. This is not an ad for Lenny's, but great, great newsletter, great podcast. And he actually has this thing where all of his blog posts are read by his AI voice and are published as a podcast on Spotify. And I was like, that's so freaking cool. It's like a step further than when you go to, you know, like NPR and news website. And on the top, it's like listen to article and it reads you the article. This is like, hey, I can actually subscribe to a podcast that is this blog. So I actually listened to one when I was driving and not the same as listening to someone having a conversation because they're reading a blog post. But I was like, hey, we should do this with our blog. We should turn our blog into a podcast. And what better voice to use than Bob Moore, who was the only person we could consent to use his voice. ⁓ Bob Moore: I figured you asked me this question and there's so many, frankly, like these podcasts and, you know, old YouTube videos and like interview appearances and all kinds of stuff. What you needed from me in order to do this was alarmingly close to zero because I think the majority of what you use was publicly available and anybody could use it. Lindsey DeFalco: Yep. Yeah, so I think the only thing we actually needed you for is that ⁓ 11 Labs does make you verify your voice with your voice when you're creating a level of ⁓ imitation voice that we created. So you can do like a quick, like I could upload 30 seconds of me going, you know, blah, blah, blah, blah, blah. probably saying more than that, but, and it could imitate my voice, but then there's a professional voice, which is what we did for Bob, and you need minimum 30 minutes of speaking, and then you actually have to verify it. So I think that's all we needed from you because we had this podcast audio, and you just had to go in and I don't even know what you had to say to it, but you had to say something to convince it that it's you and you're not stealing someone's voice. Bob Moore: It was some randomly generated sentence like, zebras eat lots of mangoes on Tuesday in Zimbabwe or something like that that you couldn't have had prerecorded. ⁓ But yeah, whoo, okay, we'll feed. Lindsey DeFalco: Yeah. yeah. All right. So I'm going to open the, I think the most popular part of this project and the reason that it landed in the top five was probably just this thing I'm about to show you. So let me go ahead and open this up. is, if you are audio only right now, I am currently showing a soundboard and every tile is Bob's voice. What I want to do, because I want to demonstrate how close this is to Bob's actual voice, is I'm going to ask you to say something, Bob, and then after you say it, I'm going to play the tile. Okay? Bob Moore: ⁓ okay. Can I yes and your idea? Let's A, test it, all right? And either I'm gonna say it or you're gonna click the button and the audience has to guess to themselves whether this is me saying it or whether this is the sound. Lindsey DeFalco: Yeah, yeah. OK, I hope the audio is clear on this, because otherwise it's going to be really easy. OK, so I'll say the word first. And then, oh no, we need to see. Bob Moore: We'll do a, we're in the, we're in the, do you see the studio chat? Lindsey DeFalco: yeah, let's get this. Yes, I do. Bob Moore: All right, we'll do a little secret typing here about who's gonna do it. All right. Lindsey DeFalco: Okay, great, All right, ready? I'm gonna say the word first. Okay, Venn diagram. Bob Moore: Venn diagram. Remix! Lindsey DeFalco: Well. Bob Moore: So that was me, real me, and then play the fake me again. Lindsey DeFalco: V-V-Venn diagram. Bob Moore: Venn diagram. ⁓ wow. All right. Lindsey DeFalco: Pretty crazy. OK. All right. ⁓ Let's do another one. Ready? This is going to be two by two matrix. Bob Moore: Two by two matrix. Yeah, that was... Whoa! It's too, ⁓ it's too good. All right, this game ⁓ has served its purpose. It sounds exactly like me. And this is not like a recording of me being played back. I never said any of these words to this system. ⁓ Yeah, you wanna run through some, everything you're about to hear is not me. I'm going on mute. Lindsey DeFalco: I got. ⁓ All right. So the first thing I'll say is that I have actually been keeping track of Bob's most popular phrases over the past, what, almost eight years in a running Notion page. And so this is really just a cumulative effort of ⁓ everything Bob likes to repeat over and over again. So you're actually learning some things about Bob here, too. ⁓ And so I'm going to play some of these and it is wild. And then I'll play this fun one down here, which is that you can have Bob speaking at a different language. ⁓ And so I'm actually giving 11 Labs the text that was translated into French and I've set his voice to French and not English. And he apparently still sounds like an American speaking French. asked some people on the team, but ⁓ better than he could do by himself. So I'm going to click some of these. Bob Moore: I came up with musical evidence. Lindsey DeFalco: ⁓ I like this one was a fun one because of all the spamming texts that the team gets from Bob. Bob Moore: I love the tone of the gift card. ⁓ So, ⁓ yeah, do the French one. This French one blows my mind. Lindsey DeFalco: Yeah, I you a gift card. Yeah, and it's Robert because, you know, that's your name in French. Bob Moore: My greatest hopes and dreams on Duolingo have now been squashed. Can you show the thing where it's longer form too? Maybe you're gonna get to that, while you're loading that up, this can render things in as long form as you want. It could do an audiobook reading of my book. It can read a full blog post. It can do whatever in French. So there's this thing that... Lindsey DeFalco: Yeah, yeah. Bob Moore: I think is very like immediate term possible with very little delay where ⁓ like real time translation, I can be participating in a meeting in French with our French team and speaking English. Like they could speak, it could be heard, it could be translated back to English and presented to me in text or read to me in English. And then I could speak back in English. And like this technology has existed for a bit. but what has not existed is that it will be said back to them in my voice. So like I could be on Zoom with camera off and it would just seem like I'm participating in the meeting speaking like me but able to speak a language that I don't fluently speak or even sub-fluently speak. And it's kind of cool and exciting until you think about all the terrifying ways that it could be used against myself and humanity, yeah. Lindsey DeFalco: Yeah, and I don't know if you remember this, but I feel like this was like the pre-AI equivalent. But there was ⁓ this meeting we were in like years ago, and I'm just about to out you, by the way. Because I remember slacking you and saying, does your background image also have you in it? Because you walked into the screen and you were sitting at your computer and you were like, ⁓ ha ha. And you'd ⁓ take a picture of yourself and put it as your Zoom background. ⁓ Bob Moore: Yeah, yeah. Lindsey DeFalco: And it's funny because that was like, and there all these videos about that during COVID, right? People like pretending to be at their desk during meetings. I don't think you did it again after I messaged you about it. But ⁓ point being, there's now this like extreme version ⁓ where you can actually be fully participating in a meeting, ⁓ which is, yeah, terrifying for, you know, phishing and scamming reasons and ⁓ maybe interesting in non-phishing ways. ⁓ Bob Moore: Yeah. Lindsey DeFalco: OK, so yeah, this is I made this whole this whole app is kind of like some A.I. slop. But the idea here was, like I mentioned, to be able to turn these podcasts. And if you're if you're just listening, I'm just showing an A.I. slop app and you can turn these podcasts into blog posts and pull out quotes from from the podcasts. And then the inverse of that is that we have a blog. It's our insider blog. There's really great content that comes out, but it's only available for you to read with your eyeballs. And so what if we take that content either as is or turn it into a script for a podcast and then generate the audio? So I'm just going to play you a small clip of this, which is ⁓ a blog post that we have published that I turned into a podcast script and ⁓ used Bob's voice for. So give it a go. I mean, that's crazy. Bob Moore: Yeah, it's got something, we're through the uncanny valley. Like there used to always be some, I don't know, just, I mean, there's ⁓ some uncanny valley, right? In which these things would sit and not sound like they're real. ⁓ I really do sound ⁓ like that. ⁓ It's got a speed, it's got a cadence, it's got inflections. Yeah, I... I don't know, I don't know. There's stuff about like, I get like a little choked up even hearing this, because like, okay, there's all this like identity fraud and stuff, but like, there's a part of me that is like, I, one day when I'm not here on this earth anymore, my kids are gonna be able to like hear me talk to them. there's some things that just like, Lindsey DeFalco: And if... Mm-hmm. Bob Moore: really bend the rules about what you think about like identity and permanence and ⁓ you know what it means to exist in someone's mind and like what someone's memories of you are and the idea that ⁓ yeah like your voice can be so reliably recreated and like if this was around a hundred years ago like I could hear my great grandfather talk about his opinions on AI and like what's going on in the world today and or reading the news. Like I could have my, you know, my grandmother reading me the news headlines or telling me a recipe out loud that I would have just had to read on an old recipe card before. And like there's something like beautiful and sweet, but also weirdly a little sinister about the whole thing. Lindsey DeFalco: Mm-hmm. Yeah. Bob Moore: It's hard to process all of that at once. Lindsey DeFalco: Yeah. And I didn't even really think about that. But I know so many people who save voicemails. Like they save like a voicemail of all of the important people in their lives. And it's to have their voice. And it is interesting because this does, I mean, you could presumably have conversations with the voice of somebody who is not around, is just wild from like a psychological, ethical, emotional standpoint, all of them. ⁓ So OK, have, again, if you're audio only, I've got 11 Labs, which is the tool that we actually created Bob's professional voice in. And there's some interesting things here you can do to modify it and make it sound a lot more like him as well, which I thought was interesting. So there's like a speed ⁓ where you can be a slower, faster cadence, a stability, which is the consistency between generating different clips and then similarity, ⁓ which is really how much you want to target being really similar to ⁓ the voice. And then ⁓ style exaggeration, which when I made your soundboard or no, when I was playing with this before, I had your style exaggeration turned all the way up. But I think that I think the soundboard is actually not doing that. Bob Moore: Ha! ⁓ that's so funny. Lindsey DeFalco: ⁓ but yeah, then you have this language override too, which is where if you put it in French, it's like, here, let's actually see. ⁓ hello, this is Bob. How are you doing? I'm generating the speed. OK, so it did nothing because we have to write it in French. Yeah, yeah. Let me just pop over to Google. I don't even know if that was mildly. OK, let me copy this, get the French version because I speak no French. Pop this back in here. Bob Moore: think it's a, yeah, you've got to write it in French. I said English in a mildly French accent there. Lindsey DeFalco: There you go. Yeah, so it just like I think it speaks it. Actually, let's see what this does if I put it back on English. Bob Moore: Bob. Yeah, I want, can you give me like Arabic or something too or like? Lindsey DeFalco: ⁓ just has a worse accent. Bob Moore: I'm so interested in like, or like mandarin, like a, yeah, like a non-romance language, like. Lindsey DeFalco: Let me, and let's see, we'll do this, the English version first. OK, so then I need to actually translate it. simplified Chinese drop this back Bob Moore: I'm shocked that this hasn't already hit the world harder. Lindsey DeFalco: Yeah, so think it just does better at speaking that language, but it's not like you even have a perfect accent. Bob Moore: Yeah, I got this anecdote. I one of the founders of Eleven Labs at an event back in October. And I believe Labs like a marketplace where if you want, once you've done this and your voice exists, you can put your voice up in the marketplace so other people can render stuff in your voice and you get paid a royalty for it. Yeah, there you go. ⁓ And apparently, ⁓ Lindsey DeFalco: Here we go. Bob Moore: you can ⁓ make quite a bit of money doing this if your voice gets really, really widely used. And there's some guy in Spain who's a native Spanish speaker who doesn't speak English who has become the most popular English language voice. And there's just something about this dude who's got this very particularly compelling sound in his voice where it's really good for voiceovers on like social media videos or like... ⁓ Lindsey DeFalco: Yeah. Bob Moore: just like there have been enough viral videos that have been produced that have his voice layered on top that, ⁓ he makes a significant, a, you know, ⁓ a meaningful salaries worth of income every year because of the use of his voice through this marketplace in a language that he does not speak. So like, I Lindsey DeFalco: Mm-hmm. Bob Moore: how it is that he gets discovered and matched up and bubbles up to the top of this. I couldn't answer any of that for you. How you too can get rich on your voice, I can't tell you, but it's a thing. ⁓ yeah, as you look at this marketplace here, like they do have John Wayne and Richard Feynman and Michael Caine. ⁓ I don't know. Yeah. Lindsey DeFalco: Yeah. You want to hear one of these? I also, this is funny. It just says that a little pop. Bob Moore: Your account was flagged. ⁓ that's so funny. I wonder why. Lindsey DeFalco: Yeah, no wonder why. pretty crazy. We got John Wayne here. Yeah. ⁓ Bob Moore: Yeah, it's like ⁓ a too good impression. A lot of these are your most standard stereotypical impressions that like a hack stand of comedian would do, ⁓ but they're like so good that they're not even funny. ⁓ That's an alternate team name for you. So good, it's not even funny anymore. But yeah, I did it. Go ahead. Lindsey DeFalco: Yeah. Well, it's actually, I was gonna say this whole voice generation thing. It was funny to me because ⁓ my wife, who does not work in tech, I was working on this hackathon project and I was playing with it. ⁓ I did, Bob, I made your voice ⁓ say, Lindsey's doing a great job at work. And I played it out loud, the living room, ⁓ and I got a funny look. ⁓ But what was funny is when I was telling her about this project and I was like working on it, Bob Moore: Hahaha! Lindsey DeFalco: And she's like, ⁓ yeah, TikTok's been doing that for years. And I was like, they have, they have those popular voiceovers and you can just put what voice you want and it plays it back. And I just thought of that, cause I'm looking at this 11 Labs page and it says popular TikTok voices. Bob Moore: Wow, yeah, I bet that's where that Spanish guy's voice lives, inside of that library, those popular TikTok voices. That's so wild. ⁓ No, ⁓ I think I'm actually gonna reach out to, or sit down with, or at a minimum compose an email to all of my family members, ⁓ just being like, hey, I'm never going to call you. Lindsey DeFalco: Probably. Yeah. Bob Moore: ⁓ Not my usual annual note to all my family members telling them that I'm never gonna call them, but this is a special one. ⁓ It's basically like if I call you at 2 a.m. and say, because this is a scam that's like as old as time, like say, hey, I got arrested and I'm scared and I need you to wire me, you know, this many Bitcoin to pay my bail or something like that. And it sounds just like me. It's not me. And like some two or three factor authentication to validate that it's actually me before you go like ever doing anything in the realm of ⁓ yeah, like buying Google gift cards and sending them somewhere. ⁓ You know, the days of the random text message like, Hey, it's Bob. ⁓ Got a second to hop on the phone. And then somebody who sounds nothing like me is on the phone. It's about to be, you get an inbound call from a number that is like ghosted to be my number and you say hello and it sounds exactly like me. Lindsey DeFalco: Yeah. Bob Moore: And I don't know where we go from there except, you know, things like pass phrases and yeah, multi-factor verification kind of stuff is going to need to be like standard. Lindsey DeFalco: Yeah. Yep, especially as video is getting there too. I mean, we're not even talking about the AI versions of yourself you can create from a video standpoint. And I know like the test for that now is it's like put three fingers in front of your face, ⁓ they're going to solve that. yeah, ⁓ all need passcodes that when you start talking to us, you got to say the passcode before you can ⁓ keep going. Bob Moore: Yeah. Well, ⁓ that's a great note to wrap on. a great rundown of projects. And yeah, if company's not doing this stuff, I'd suggest doing ⁓ It was a pretty eye-opening thing. ⁓ And I do think that having... Claude stood up and having some of the like foundational pieces of this in place before we did it, I think was also really essential. think even like three or four months ago, this would have been a lot more chaotic, but I think we... Lindsey DeFalco: Well, the same is true for three to four months from now. I'm like, we're going to have even more in place to get. Bob Moore: ⁓ yeah, yeah, once we have that, like a lot of this stuff that was kind of experimental is going to be the way we're running the company within three to four months. And then what do you build on top of that? ⁓ Yeah, we should put that next hackathon on the books now, because it's coming. Lindsey DeFalco: Soon the hackathons are going to be build an entire company in 24 hours. Bob Moore: Yeah. Yup, yup. I think a few of them did. From what I just saw. ⁓ All right, till next time everybody. See ya. Lindsey DeFalco: All right, bye.