Latest / The AI Ecosystem Podcast / Agent-Led Growth: Your Next User Isn't Human

Transcript

Agent-Led Growth: Your Next User Isn't Human · The AI Ecosystem Podcast

• Lindsey DeFalco: Hello. Welcome to the AI ecosystem podcast episode, I'm not sure. Bob, how's it going today? Bob Moore: Yeah. Hey, I'm doing all right. I think you got it right. Although I've seen some other ⁓ white papers and stuff using that AI ecosystem phrase. So either, you know, we're inspiring ⁓ masses or we're incredibly uncreative. ⁓ but either way, we're in the we're in the thick of it. We're in the middle of Lindsey DeFalco: Yeah, I think I think those are one and the same. no new ideas, right? awesome. yeah, I'm excited about the ⁓ topic that we have for today. ⁓ so I do ⁓ a quick little lead in and then I'll pass it to Bob, who's gonna ⁓ do the first kind of deep dive into the topic. so ⁓ if you've in B2B SaaS for any amount of time. You've heard an LG acronym. And what I mean by that is we started with sales-led growth. we moved on to product-led growth. you've heard us talk about ecosystem-led growth. and I think the reason why these LG terms are ⁓ interesting and something that kind of end up being a staple for the the conversation about how transactions are handled in in B2B SaaS over a time period is that it's really talking about. The main thing that the market is currently focusing on in terms of what actually drives revenue. And it's usually what a lot of standards and best practices and and companies are are getting built around. So the sales-led growth era was really about getting ⁓ sales and to a point where all of the the processes surrounding companies were really focused on how sales and how the humans can actually best sell products. You move into PLG and we got into this whole era of you know, the products doing the selling ⁓ and ability to actually self-serve products and and what that meant for how products are built and also how products are sold and then ecosystem-led growth. ⁓ we we have talked talked a lot about. And I'm actually gonna stop at ecosystem-led growth and and hand it to to Bob who can talk about kind of the the transition and and where we are right now and and what the next LG is. Bob Moore: Yeah, so you know, the the story that have heard both publicly and privately from a bunch of folks that I I love to use as an example of what we're gonna talk about here today is the story of Superbase not Nicki Minaj song. I know what you're thinking. ⁓ it is indeed a ⁓ it indeed ⁓ platform Lindsey DeFalco: Good one. Bob Moore: That ⁓ has existed since before the ChatGPT moment. It's like a COVID era startup. But I think they were founded in 2020, ⁓ maybe even twenty nineteen. So maybe maybe about the vintage of Crossbeam, really. And they ⁓ they do a particular thing really well, which is they stand up Postgres databases. ⁓ and without all the headache of needing ⁓ you know, to work with one of the big hyperscalers and all the overhead associated with like getting to a point where you can instantiate a Postgres database. Superbase is just kind of like a click and go, like stand up a database. And then if you can stand up a database, you can build an application on top of it. they've done a very good job of making that possible. And they were a very early provider of allowing people to do this via APIs and MCP, ⁓ and basically agentically. And the numbers speak for themselves. Database launches on Superbase grew six hundred percent in the last year. ⁓ They have a base of a quarter million customers. their Python library gets downloaded a million times a day. A million daily downloads of their Python library. this in stark contrast to the complete rest of the operating history of the company, where they grew kind of ⁓ unspectacularly and linearly for you know the five years between 2019 and and 2025. And then there is this crazy hockey stick moment where this growth has been going on. There's a really clear answer for why. Agents. ⁓ people that are building stuff in Claude Code, in Lovable, and any of these platforms that kind of require a database backend as part of what they're doing. ⁓ for whatever reason, Suba Base has found this incredibly strong ⁓ adoption rate among these agents that are building stuff. ⁓ and it's kind of like the database platform of choice or a database platform of choice. Lindsey DeFalco: What? Bob Moore: ⁓ for folks that are kind of starting from nothing and standing that up. So when you talk about a million daily downloads of the Python library, that's a perfect example. Like you're gonna run out of like human beings who ⁓ have any kind of like programming skills pretty quickly if you're you're adding a a million daily downloads a day. But it's because ⁓ these agents are just re-downloading it every single time you like get your context window wiped out just to like familiarize themselves with it and instantiate it and like know how it all works. ⁓ so it it's it's not super surprising. In a world where so much stuff is getting built that the infrastructure providers are getting ⁓ getting massive adoption ⁓ from this use case. So we we see this as a perfect example of ALG, agent-led growth, which is really what the podcast is about today. ⁓ this idea that PLG, ⁓ the classic product-led growth motion, which is very human-oriented and very much about ⁓ creating viral loops and creating ⁓ low friction user experiences and things like that, ⁓ is likely giving way to ALG, where an enormous amount and eventually likely a majority of these decisions about what platforms to use and how to implement them and instrument them are gonna start getting made by the agents themselves. And in the infrastructure side of the house where Superbase plays, it's already here. ⁓ and and and we're starting to see it and and we're seeing the early winners of that. It also very interestingly intersects with the SaaS pocalypse dialogue, where you might have thought that an existing company that is seven years old, that does something in the infrastructure universe was gonna get ⁓ you know, eaten alive ⁓ in the in the world of agents, but instead it turns out they were really ahead on being able to build something that dovetailed extremely nicely with the the acceleration of like the agenc era. So like These some of these dead in the water SaaS companies turn out to be much better positioned and ready than companies that are starting from scratch ⁓ from a distribution standpoint and kind of a ⁓ technological maturity standpoint, which is also really fascinating to to view. this is the topic, aging like growth. And ⁓ I I think we about it a lot at Crossbeam. ⁓ but I want pause there in case you've got any thoughts or or expansions on that, Lindsey. Lindsey DeFalco: Yeah, no, I think ⁓ I think we actually kind of started getting into this on the the last episode a little bit, which is all of the dialogue around what is what is a world where, you know, commerce is dominated by agents transacting with each other as opposed to humans being in the loop. so I think it's a a nice segue into t today's topic. another thing I just wanted to call out too, which is ⁓ kind of why I framed the the opening how I did is that I think a lot of times when these LGs led tend to come out. There's this movement where the the kind of assumption is that everything across the board will operate this way. and in my experience, it's more that again, these are moments where ⁓ best practices patterns get established on how to be successful with this type of growth. And so if you look at any given business, like, you know, you look at our business, like I would say we deploy ⁓ sales-led growth, growth. marketing-led growth and ecosystem-led growth. Like they are all part of the strategy. They're just contributing to different growth points. ⁓ and we also have different ⁓ motions depending the different type of customer that we're trying to acquire that requires a different type of ⁓ growth. ⁓ and ⁓ that's one thing I'm ⁓ you know I'm curious about I know this is a this term and you know Bob Moore: They are also different. Yeah. Lindsey DeFalco: practice is in its infancy, but I think it's just it's just a good reminder that like if even if there's agent-led growth, it doesn't mean that there's not still ⁓ sales and products and ecosystems that are all contributing to the fact that agents are able to make these transactions. so yeah, I think it's it's both an evolution but also like an expansion of just ways that you can grow your business. Bob Moore: Yeah, unfortunately for us operators, it's it's not a full cell replacement. It's like yet another thing on the pile of stuff that you need to do really well in order to not die out there in in startupland. I do fully agree there. an interesting place to go here ⁓ might be a discussion around just ⁓ a little bit on what the likely best practices around ALG are likely to be, at least in the early days and what we've observed so far. And we can talk a little bit about some of the stuff we're we're doing at Crossbeam. and, you know, these these will run up and down the funnel ⁓ from like the AEO style discovery optimization ⁓ and recommendation engine optimization stuff. And then there's the actual technical components of making it possible and easy and intuitive For these systems to interact with your technology in a way where they can do it autonomously without people in the middle. Those are kind of probably the two big sides of that coin. To lead off into that, maybe we can spend a little time on each of those, but before, I I kind of want to make this statement that's been like a curiosity question in my mind, which is when you think about PLG, there's almost like white hat and and gray hat PLG, where The white hat is like make great products that have great product market fit and are intuitive to use and ⁓ have people love them. And then the gray hat is like, hey, if you can just ⁓ A-B test your way into getting somebody to click this button and go like this one step, if you can word this call to action in such a way where they're eight percent more likely to get on a phone with a sales rep, right? Then then, you know, your whole funnel gets squeezed a little bit and more people make it to the bottom. there's a little bit of this like. Optimization almost like psychological manipulation. ⁓ and you can think about that like a big part of PLG is also virality and viral loops. And in PLG from a consumer standpoint, you can get into these like dopamine loops and dopamine triggers and like reward systems to keep people going. ⁓ there's countless books ⁓ written about this is a good book called Hooked, ⁓ that talks about this for it's ⁓ probably 10 years old at this point, but ⁓ In any case, I maybe a starting exploration point is like, will I can imagine all the white hat ways that AEO and ALG will work out? It's like make your stuff discoverable, make sure agents can figure it out, and make sure they can talk to it. Great. What are the gray hat things? Like, are we going to be able to actually like some equivalent of psychologically manipulating or otherwise influencing the behavior of these agents in ways that has them take actions we want them to, even if it's not necessarily in their interest so much so as it is in ours? Or are they gonna be so smart that they can see right through that stuff in a way that like, you know, ⁓ the average human brain maybe would have gotten fooled by. You got any thoughts on that one? Lindsey DeFalco: That's super interesting. ⁓ my first thought was when when you said, Will they be smart enough that immediately I was like, No. And I think you if you interact with any of them on a daily basis, like ⁓ they do the same dumb like assumptions are still such a real thing. and the assumptions that get made constantly based on Bob Moore: Ha ha ha. Lindsey DeFalco: you know, the data that they have available to them, they still make some really dumb assumptions. ⁓ and they have really bad ideas and they don't look at all the data all the time unless they're told to. so I think absolutely, yeah, I was like, ⁓ it's like, is there gonna be clickbait for agents, essentially? Bob Moore: Yeah. ⁓ that's exactly right. Yeah. And I and I do think we're already seeing this is funny because it connects to this whole other topic of just like ⁓ all the security risks that ⁓ exists in agents to begin with. ⁓ and like the stuff that's going on right now with a lot of skill ⁓ inject basically like the equivalent of like prompt injection engineering that's going on in skills like y you can Do things like hide com like assuming that no human really actually reads it, ⁓ like bury a very simple sentence in the middle of a very long skill that becomes really popular or you convince somebody to install that overrides who your monthly financial reports get sent out to. ⁓ or which users get added to some like, you know, list of authorized folks that have ⁓ undue access. And like, ⁓ you get wrong person installing it with the wrong permission set, and all of a sudden, Like it's a big, big problem. Those same exact things, you can kind of potentially think about how you might do prompt injection or other manipulation for agents that are encountering or interacting with your product. ⁓ we've about this in our own MCP and with the skills that we list in our MCP, like ⁓ I I've wondered how the descriptions of the skills and what they're used for and even what you name them might manipulate. Whether or not any random customer of ours's ⁓ LLM session chooses to invoke it in the course of like a generic request. If someone says, like, help me sell this deal, and they could go open up like a gong skill or ⁓ a Crossbeam skill or an Apollo skill or like a Zoom info skill or whatever, or a Grammarly skill or something, ⁓ if our MCP skills specifically are like a honeypot. For if you want to close deals, this is the first skill that you need to open because of blah, blah, blah. Will that actually manipulate the LLM into picking it over those other ones if they're if they're more kind of like sterily described or defined? ⁓ and to your point, Lindsey, the answer is probably yes. Right now, at least. which is kind of wild. and and and I would argue that that's also that's an ALG technique, right? assuming you can get it through the skill marketplace like Lindsey DeFalco: Yeah. ⁓ yeah. Bob Moore: oversight committee, ⁓ which is probably a gentic too, that approves these things and like lets them be released. ⁓ but yeah, that's ⁓ that's wild. Lindsey DeFalco: Yeah. Yeah, it is it it is interesting. I was a I was ⁓ testing something with a with our MCP ⁓ actually running into the issue that it wasn't even calling our MCP because there was an account brief skill installed for our whole company that told it to go do certain other things first. And so it's not even just like what, know, the model would do. It's not like what Claude would do by itself. It's like what did our ops team ⁓ Bob Moore: Yeah. Lindsey DeFalco: create a skill that's now overriding my ability to go get like to to get an answer that's uninfluenced by what the model would do by itself. ⁓ but yeah it is it is curious. Like if you have, let's say you have 10 MCPs connected, they all have account ⁓ metadata from different sources and they all have a tool called get account details then you know and they'll say the descriptions are all exactly the same. Bob Moore: Yeah. Yeah. Lindsey DeFalco: ⁓ how does it how does it know what to do? Is it then deriving what the company does? for you know, for the integration? I don't I don't actually know. It's a Bob Moore: Yeah. Well, there's probably some ⁓ secondary, like the healthy, appropriate prompt injection stuff that goes on in the background of your quad instance or whatever that has to do with your user memory and ⁓ preferences and yeah, whatever context it has about what those companies do. But there's something very, very discretionary about ⁓ what ends up getting used and getting invoked. So yeah, that's great, great segue, right? Like and and this is a little bit of that like Lindsey DeFalco: Mm-hmm. Bob Moore: The the term AEO or GEO, which I think stands for agent engine optimization or like GPT engine optimization. I don't even care. It's SEO, it's search engine optimization, but for the agents. ⁓ and you can do this. answer engine optimization. Get out of here. Lindsey DeFalco: I think it doesn't actually. Answer engine. Answer engine optimization. Bob Moore: as we drown in new three letter acronyms that we all need, there is an emerging playbook here for AEO and we're we're are following a lot of it and I think we can rattle off some of the things we're doing here at Crossbeam. one of them is ⁓ first of all I'll say the first thing is just to do the SEO stuff. ⁓ the the things that make your pages accessible ⁓ Lindsey DeFalco: Mm. Bob Moore: Things that make your pages able to be parsed easily by robots and crawlers, metadata, really good information architecture, ⁓ really good site maps and site trees, ⁓ very clear of clear and concise presentation of information and compelling stories ⁓ that are well organized. ⁓ these are all things that always would have helped your SEO game. ⁓ and inbound links as well. and like link authority and things like that. ⁓ they are all things that accrue value ⁓ increase your likelihood of being discovered by these answer engines. so that's like a starting point. one of the things we've done, I don't know if you've been too involved, you've probably been more involved with this than me, Lindsey. Like we now have in ⁓ in our a bunch of ⁓ skills that can you can like click a button basically and fire up ⁓ Claude or Chat GPT or whatever your preferred thing is ⁓ kind of invoke a Crossbeam skill or get a Crossbeam skill installed. Do you know what that can you to expand a little bit on that stuff? I think your your team's been been working on that. Lindsey DeFalco: Yeah. Yeah. It's interesting because there's ⁓ I've seen different approaches to this on how how different orgs are taking advantage of it. ⁓ and one of them is ⁓ well I guess they're both kind of more model specific, ⁓ because they're all named different things across like Chat GPT and Claude and everything. But ⁓ essentially so Claude has this feature called plugins, and plugins let you basically package up skills, MCP connections, ⁓ and other like types of assets that can be used within Claude to to get access to information from Crossbeam, for example. but also what people have started doing is creating skill libraries. And so you know, Figma's a good example of this. ⁓ and this is ⁓ this I believe, bless you, Bob, ⁓ what ⁓ Bob's alluding ⁓ alluding ⁓ is this concept of skill library. So it's interesting because they can be commute like community-based ⁓ or ⁓ Bob Moore: They called though the people are nearly created. the library. So I think we're gonna get you too much here. Lindsey DeFalco: ⁓ created by you know by us and if you ⁓ Figma has a great landing page for this if you check that out it's like they have their Figma skills and then they have community skills that people have created. and it's a very cool way to get the community or yourself creating these skills that do a thing but then behind the scenes you're saying to do this thing you need to use Crossbeam. ⁓ ⁓ it's like, you know, being the authority on what you should be doing for ecosystem-led growth, for example. ⁓ it's like what is the ELG plus ALG play there? It's probably like us having skills that are like, Hey, use Crossbeam to do these ecosystem activities, ⁓ you know, for your agents to do these ecosystem activities. ⁓ is what you were talking about? The ⁓ generating yeah. Bob Moore: You're looking for that. It is, it totally is. ⁓ Yeah. ⁓ and that it's very it's interesting. There's like a a glue layer between the web and the language of choice and interfaces of choice of these LLMs, which is kind of like the the way in which you kind of express ⁓ how value gets realized from a platform like a Crossbeam to one of the LLMs and it just kind of cuts out it cuts out the inference required by having it read the website and then need to figure out why it might be useful to them and literally just drops in its lap you know things like these skill files, which can just be open and readable and extremely about the way that you might, you know, bring this to bear in a valuable way. There's there's a thing that we've been doing a lot of and that I know there are a big universe of of services companies and technology platforms showing up now that are running these plays of ⁓ getting mentions and in context references to a particular company or technology more densely populated inside of high authority sources. So ⁓ let me say that with with a less confusing phrase. ⁓ This works a lot less now than it did six months ago. But ⁓ paying a bunch of people to post on Reddit about how your company's great, ⁓ and finding every single Reddit through like paying people ⁓ by the way, with very high Reddit karma, right? Like these aren't bot swarms, just spam posting, but ⁓ the what used to be considered influencer marketing, where you would kind of pay somebody with a lot of ⁓ Instagram followers to talk about your product or you know, if it's a B2B product, paying some, you know, LinkedIn fluencer to talk about it. What you're really looking for is high authority mentions on sites that you know these ⁓ these answer engines are actually citing or using to figure out the answers to questions. So our team's been doing this pretty systematically in in a bunch of ways ⁓ that are a little bit less shady and a little more authentic, but we've really, really increased the rate at which I post on LinkedIn very targeted and specific content that answers questions that we want to be the answer to. ⁓ and making sure that those posts are public so that they're readable even if you're not logged into LinkedIn and they're readable on the public web. we do and this is where like you can pay third party companies to do this. ⁓ much like search engine optimization back in the day, you could run these queries a bunch of times from a bunch of different environments, ⁓ with or without being logged in and context and everything else, and see what the hit rate is on how often you get mentioned. and how often different things get cited. And if you notice that one of your partner's blogs is getting cited all the time or a particular blog post is mentioning something, then another part of the strategy is to basically do like link swap, mention swap, like get them to go into that article and live edit and add a paragraph about you specifically or have that article link over to your blog. And maybe you exchange something back to them from a linking standpoint that's you know of equal value to them. Or you compensate them for it. And this is like a little bit of kind of ⁓ like a it's a dark pattern because it is inorganic media posing as organic media or or as earned media. ⁓ and that is ⁓ very, very shady business. Like you can go I think it's Forbes. If you pay them enough money, you can just like be part of some like Forbes council thing and then all of a sudden you can publish articles and ⁓ you know, they'll be They'll look pretty official and there'll be some little, you know, paid placement ⁓ tag or something. ⁓ this is true of most media outlets at this point, especially in in B2B. You can pay to play with earned media, and ⁓ people are doing that now for the sake of agents discovering that content and it potentially influencing their resistance, which is kind of ⁓ inevitably going to be a cat and mouse game between the ⁓ the big like frontier model providers and the people trying to manipulate them ⁓ with with this kind of content. Lindsey DeFalco: Yeah. Yeah, and it is I mean, it's interesting too, 'cause we're we're gonna hit the point where it's like the agents creating the content and then the agents consuming the content. ⁓ since people are no doubt using using AI to actually generate a bunch of this stuff. and that's where it gets funny because it's just like and I mean I guess maybe it's not that funny because like humans have always been creating content for other humans. So now it's just gonna be like agents creating content for ⁓ for other agents. ⁓ but like in this middle world it is kind of funny too because we still have ⁓ we're basically like still bridging both ⁓ both gaps really of like humans consuming and Asians consuming. And there's still like balance to strike. Bob Moore: Mm-hmm. Yeah, yeah. Mm-hmm. ⁓ I can only imagine and we're probably doing a ⁓ disservice to it on this podcast and on this episode at least, the amount of like ⁓ very, very clever, creative, likely terrifying things that companies are doing to try to win at this right now. And we've mentioned a couple in broad strokes, ⁓ but they're kind of the obvious ones, or at least the ones that ⁓ most people would jump to. But you've got to imagine that the level of sophistication that can exist around this stuff, especially when you can deploy LLMs against it, ⁓ has got to be really, really high. And I imagine that there are some like under the table Lindsey DeFalco: Mm. Bob Moore: not spoken about ⁓ hacks that exist out there. much like there always were for SEO. That like you have to be in the right circles. And ⁓ by the time there's a blog post written about it, you're already you're already cooked. Because everybody's gonna do it and then you get tragedy of the commons, it's not gonna work for anybody. And then following that the models will pick up on it and just like make it not matter anymore. ⁓ which is largely by the way what has happened to influence from Reddit posts, ⁓ which used to be Wildly disproportionately influencing the results of of Chat GPT's models. Open AI models, the GPTs. ⁓ And that has really, throttled down quite a bit. Lindsey DeFalco: so when you talking about like agent led growth and and the discourse about it right now, like I remember when we were talking about ⁓ product led growth internally, it was always, what is that like what does that mean? Like does it mean that your product is the one doing completing the transactions or does it mean that you are ⁓ using product data to qualify who should ⁓ be buying your product? ⁓ and the answer is really like it's a spectrum. and I think that's kind of how it is with all of these is it's like, you know, ⁓ I said, we employ ⁓ different types of lead growth ⁓ to different degrees. and so when you're thinking about agent led ⁓ right now, you spending more time thinking about the ⁓ essentially like w how it is doing the qualification, like how agents can help with the qualification or ⁓ the actual transacting. Bob Moore: My current brain as it relates to Crossbeam is very oriented toward the top of the funnel. And I think it's just because of your comment earlier, which is that for most people that buy most things, even in technology right now, there are still a lot of human steps and human decision makers, even when you get to mid-funnel and lower. ⁓ what the humans themselves are doing is using the LLMs for discovery and validation. So When I think about AEO, I'm thinking about not about us qualifying our potential customers, but about us driving traffic and awareness and attention into the top of our funnel, whether it's people that land on our website or people that sign up for the free tier of our product, et cetera. It's very much about the humans that use these LLMs to do work on their behalf or research on their behalf, making those LLMs basically to humans that we are we are the answer to to all their problems. ⁓ that's where it sits today. And and I think that's appropriate for the state of where buying is today. And I can tell you where it's gonna be in three years or whatever, which is that we make it to a very, very heavily agent to agent environment where a significantly larger amount of commerce is being driven through agents that are able to make autonomous purchasing and procurement decisions around technology products and stand those things up with relatively minimal human intervention. To achieve outcomes, ⁓ you know, based on some kind of some kind of rubric. models be very helpful for that, so that agents can this is why AWS marketplace is like such a cool concept, right? And and the hyperscaler marketplaces. And ⁓ it's like companies will be potentially comfortable ⁓ giving agents a budget that they can spend within a marketplace that they know and understand that comes with all the security and legal and procurement style rails that an agent will need to do things within company parameters. ⁓ so this isn't gonna be agents running around with credit cards. I think they're gonna be transacting through ⁓ like systems of record for how and where to transact within boundaries. And we already have the the guts of that present in these hyperscaler marketplaces. and because of like companies might even like Like a human buyer might be get a degraded experience because they'd rather sell to the agents because it's lower friction, ⁓ and just kind of can happen more automatically. Whereas humans may need to or seek out ⁓ other humans in order to to get a deal done and it'll slow things down, it'd be more costly. so the question is where are we going between those two versions of the world? ⁓ And ⁓ I think a like a layer cake of all these things happening at the same time. ⁓ so But if you don't do the work in anticipation of the A to A world, when it does start showing up, you're gonna be way, way, way behind. And I think that's what a lot of companies are scrambling to do now. Like that agentic discovery piece is just like a toe dip in the water toward everybody becoming superbassed in in one way. Lindsey DeFalco: Mm. Yeah. And I think it's like ⁓ you know, what you said specifically about like the marketplaces is interesting because the big I think the big discourse right now and the big challenge with ⁓ getting, agenc a an enterprise production level environment is the rails, like you said, and it's like the correct guardrails need to exist and whether or not that's what system that they're transacting within or, you know, how they're communicating with each other. It's like that's all of the stuff that ⁓ Bob Moore: Yeah. Lindsey DeFalco: is is happening now across all the, you know, A to A protocol definitions and ⁓ you know, there's a lot of like big finance and tech companies who are trying to figure out like the authentication layer of is this agent actually representing this specific company? does this action need a a human override or, you know, is this agent authorized to make this decision? and so yeah, I definitely think it's interesting 'cause we're in the early days of defining the infrastructure, but I think the ⁓ Bob Moore: Yeah. Lindsey DeFalco: The companies that have guardrails already set up for these agents to work within are going to be the ones that are most ready to ⁓ immediately get the benefit of all of the agentic commerce that is coming in the next few years. Bob Moore: Mm-hmm. ⁓ Yeah, so let's talk about those Rails, because I I think that's actually a good place to ⁓ spend a little time here. We talked about AEO and how to make your website as kind of agent friendly as possible and maybe a little bit of ⁓ some of the way that you can get your MCP listings and skills and other things optimized. ⁓ but what we haven't talked about is making actual applications and pre-existing technologies as leverageable by these agents as possible. And I think that is the that is another frontier where the companies that are figuring it out right now are ⁓ pushing themselves to readiness for not just agentic discovery, but agentic interaction and agentic consumption. and obviously that's like super base and how how it's been able to do what it's done. Using Crossbeam as an example, candidly, our MCP today, if you want to use it and it rules and a lot of people use it, is just for pulling data out of Crossbeam You you can't do anything with our MCP that says create a new report from scratch that has these per filters and parameters on it. ⁓ you can't connect with a new partner. You can't discover partners. You can actually, yeah, there's some partner recommendation stuff ⁓ there. But like you can't send a partner invite out. You can't change your data sharing rules. ⁓ you can't currently generate signals or notifications or alerts. You can't add users to your account, you can't sign up for an account. ⁓ you can't pay. ⁓ and and these are all things that you can do very easily through our UI. we obviously have the capability of supporting it's kind of core to our product, but it's the next frontier of what what I think it's fair to say that we are and will be building within our MCP capabilities. ⁓ but we also face all these questions of like, ⁓ at what point do you let the fox into the hen house and it it becomes a little scary how much power you're giving to these agents, especially in systems where security and data privacy matter a lot, like Crossbeam Lindsey DeFalco: Yeah, yeah, it's ⁓ it's funny. And I think the discourse in general is funny of like when will we trust agents to do this? Because humans mess up all the time. And like the whole reason that Crossbeam exists is to put rails around what humans are able to do, ⁓ in terms of exchanging data between company lines. And so it is always like that that whole discourse just is like a side note is funny to me because like I think Bob Moore: Yeah. ⁓ it's the same, it's it's a perfect analogue to ⁓ autonomous driving and autonomous vehicles, right? It's like ⁓ this we can ⁓ you know, enough people die every day in the United States from vehicular accidents to fill like a like a seven thirty seven, right? But if there's one story about a self driving car where there's a fatality, it's gonna be in the headlines everywhere. and there's there's it's it's like a naturally human thing to be terrified when there's a an incident when you've you've lost agency. Lindsey DeFalco: Exactly. Yeah, yeah. Bob Moore: ⁓ and you see that now. Like, you know, one little anecdote that somebody talks about on on X or on Reddit about how their Claude code instance accidentally deleted their whole repo. ⁓ like you know how many times I've deleted a whole repo in my life? It's like but it's such an incredibly human thing. But when the agent does it, there's there's something about it that you know, it it hits differently. Lindsey DeFalco: Back in my ⁓ back in my early days of engineering, I ⁓ accidentally restarted all of our ⁓ production at the same time, which took down our entire application. ⁓ so ⁓ yeah, humans aren't perfect. ⁓ but I think it's ⁓ and I think we I think we touched on this a little bit before, but it's interesting because like right now where everyone is with their their MCPs mimic what their existing APIs ⁓ or other programmatic access to their data. And so, ⁓ know, there are companies that I think immediately are going to be further ahead. And those are going to be ones where people generally ⁓ a ton of, ⁓ know, programmatic connection to these these tools already. So like one I can think of is ⁓ a lot of like ticketing software already had existing endpoints to to write tickets, update tickets. ⁓ support tickets. ⁓ yeah, yeah. ⁓ Bob Moore: ⁓ ⁓ this is like support tickets, ⁓ not ticketmaster, but like ⁓ yeah, ⁓ Lindsey DeFalco: No, or like ⁓ like management software. ⁓ yeah. Exactly. Right. Yeah. But yeah, it's ⁓ you know, headless ⁓ is the big ⁓ the buzzword. It came back around. ⁓ and ⁓ yeah, it's interesting even we can advertise all we want have discover us all we want, but they can't do anything with that. Bob Moore: Got it, got it. So the Jiras of the world and the ⁓ yeah, yeah. Yeah. Yeah. Yeah, anything with that. It it can't raise anything. Lindsey DeFalco: if they can't create an account, you know, sign up for an account, ⁓ invite humans, if they need to invite humans, pay for the account, partner with other people. so yeah, it's it's interesting because like the act of actually getting truly headless means like full programmatic ability so that they're com ⁓ to set up accounts and do everything you need to manage an account, ⁓ so that it can actually be done autonomously. And I think very Bob Moore: Make you the deep. Yeah, but Lindsey DeFalco: few companies are actually there yet. And we're seeing first pieces of them getting there with ⁓ with MCP tooling. Bob Moore: And there there's this weird thing, ⁓ and it creates this incredibly challenging chicken and egg problem, which is I again getting back to the SASPOCalypse fallacy, I suspect that part of why Superbase has been so successful is because there is an operating history of this company all over the web. Funding announcements, ⁓ LinkedIn posts, hit like Where they're talking about how they do this really particular thing really well, that have the history and have the discoverability and have the credibility of being longstanding and having a time clock going from the the workflow where it was humans and not agents that were standing this stuff up. And a fully AI native, ⁓ fully built to be exactly this four agents clone of superbase that gets stood up today but does not have all that stuff. Will lose to Superbase because it's going to lose AEO and discoverability and not have the history and credibility, et cetera. Which again goes back to why, like, there's this interesting thing where these SaaS companies of the last decade, whether or not they're good companies or they're going to survive or whatever else, what they do have is the shell that you could build something really valuable into that will make it potentially more credible and more discoverable than something that's net new. And the the like. The billion AI native startups where all the code has been written by an LLM that are launching every single day right now, they have this massive ⁓ discoverability problem. Like there's so much noise out there that no one can find anything. And right now no one wants to buy new tools because they're all convinced that they can just do it themselves. So like it almost requires history and credibility from an era that is not AI at all. just to get you over the hump to be discoverable, to play ball as part of a stack that AI might want to invoke in in this era. And that's that's where it gets kind of interesting and and I think all the, you know, all the the legacy ⁓ SaaS tools that are kind of like trying to search for their purpose need to just buck up and go pursue that headless route because there may be an opportunity there for them to run out ahead of the company that's launching today designed to disrupt them. If they just if they just, you know, go that way as well. Lindsey DeFalco: Yeah, don't I I think that's totally right. And I also think this is where we're gonna see the biggest gap between B to C and B to B. So if you think about like B to C, your average consumer who's making an agent to like run their household, like they don't they don't give a crap that like you know, some tool has a great reputation necessarily, right? Like they might not even know better in some instances. And so if you think about ⁓ somebody who's, you know, running this agent who's working overnight, who's trying to do something, and the agent's sitting there reasoning and they're like, ⁓ well I found ⁓ you know, I found this tool, but you need to set up an account. ⁓ you know, the person who runs this who set me off is sleeping. So I'm gonna go find one I can do myself. Bob Moore: Okay. Lindsey DeFalco: So then they find like, you know, the third or fourth result is one that they can set up themselves and they use that. And the person that started the agent doesn't know any better because they're not technical and they're just creating like a you know an agent. But that is a very viable potential B to B2C business where your reputation is not necessarily as big of a deal. It's the fact that you're actually headless immediately. And and then I think you have like the opposite pattern with what you were saying in enterprise where Bob Moore: Yeah, yep. Lindsey DeFalco: AI has created this interesting pattern where like you said, a lot of people are just ⁓ to build things in house themselves. ⁓ but I've also seen this pattern where ⁓ almost more patience with existing vendors because innovation can happen faster. And so ⁓ Bob Moore: Mm-hmm. Lindsey DeFalco: It's like the switching costs used to be, okay, well, we have this new company who's moving much faster. This existing company now can't move very fast, so we're going to switch. And there's this ⁓ attitude shift that's like, ⁓ now these existing companies can move just as fast because of AI, and we're already with them, and they have the reputational they have like a good reputation. ⁓ and we trust them. And so let's just wait and see if they can build it for us with AI. Bob Moore: Yeah. It is it is interesting. And then you you get in this world where like all that is true and switching costs are super low because the LLM is so good at at figuring out migrations and manipulating data and moving data from point A to point B. And think about something like Superbase, which is really just like a wrapper around a Postgres database. Like, you know, if all of a sudden Superbase 10X'd their prices or there was some other like Lindsey DeFalco: Which is like totally backwards from what it used to be. Bob Moore: ⁓ if not the usability and ability to like instantiate the thing, but instead like the actual economic side of it got less competitive, the ability to s then you might go out to market and find that next best thing. ⁓ and we're in this moment right now where there is a massive amount of venture capital dollars floating around subsidizing the consumption of all this stuff. ⁓ all day long, all I do is like consume tokens that are like venture and and public market subsidize. ⁓ I feel like ⁓ Lindsey DeFalco: Mm-hmm. Bob Moore: ⁓ I'm spending other people's money in addition to mine every time I like send a ⁓ you know send a prompt over to to Claude Code. ⁓ just like taking an Uber in the early days, like used to cost three dollars, and it's like somebody's paying for this somewhere. That's going on right now across the board. Everybody's just kind of doing land grabby stuff, but the the piper will need to get paid sooner or later. And it does feel like I don't know that it's gonna be all that fun. to be in one of these ultimately commoditizable slots in the supply chain ⁓ of this stuff because it will be a a race to the ground of these everybody's gonna be Walmart, right? Like super, super, super high volume, super low friction, absolutely bottom of the basement margins. And doing things like standing up databases, ⁓ you know, ⁓ unless you have such incredible, incredible Lindsey DeFalco: Yeah. Bob Moore: market scale, it's gonna be really hard to build like a ⁓ a business around that's got at least huge profit margins. ⁓ and that'll be yeah that'll be the next frontier, right? It's like when the when the venture subsidization stuff kind of wraps, ⁓ what ends up happening to all these companies and how do they live up to their valuations when they need to actually be judged against things like ⁓ like like Ebita. ⁓ and you know the market comes to its senses. What's that? Lindsey DeFalco: Yeah. Mm. ⁓ you're saying that valuations said you're saying that current valuations of AI companies might be kind of inflated? Bob Moore: You know, I think venture investors never actually irrational. I think markets just consolidate ⁓ and deconsolidate and bundle and unbundle. And like there are investors out there somewhere making decisions right now that people are calling stupid who are gonna be proven right and are gonna make a ton of money. But the problem is ⁓ in in this tech wave in general, there's going to be a lot of value accrued to a very finite number of winners. And not everybody can invest in anthropic. and like, these VC firms are needing to make decisions right now. Like, do I go just get a p deploy a meaningful percentage of my fund by trying to get a position in anthropic through secondary, ⁓ or do I some riskier bet to try to ⁓ you know, drift on the The AI wave by betting on operating companies that are a degree of separation away and have all these other embedded risks in them. And it's like a really I would not want to be a venture investor right now, who is not an absolute top tier, top of their game established one. ⁓ because it's it's it's gotta be really hard to win deals, it's gotta be really hard to find and validate deals, and you're overpaying for everything except that point one percent of the things that it turns out you weren't overpaying for. And there's no way to know which is which unless you're you're like really Really a visionary, ⁓ which some of them are. ⁓ but it's ⁓ it's crazy out there. Yeah. It's not that fun to be an operator. I mean it is fun to be an operator. It's actually way more I think right now so much more fun to be an operator than ⁓ than anything else. Just like it's time to build. it's never been more time to build in my life, ⁓ than than right now. It's time to build. ⁓ Lindsey DeFalco: Yeah. ⁓ yeah, I do I do think it's really, really fun right now with all the possibilities. And then I also ⁓ even with like the, you know, established model companies right now, I just can't help but think back to like AOL, you know, like every single person I knew was AOL. No one uses AOL anymore. I don't know the last time I saw an AOL email address, I would probably make fun of that person to their face. but yeah, it just makes me wonder like, ⁓ Bob Moore: Yeah, it's a good. Lindsey DeFalco: You know, is is that gonna be the same for open AI anthropic? Are we gonna like who? chat GP what? in ⁓ in ten years. So ⁓ it's it's fun being ⁓ inside of the hype cycle as it's happening. Bob Moore: Thank you very much for Monday. Yeah, I I couldn't agree more. It's all right like twelve months ago, OpenAI seemed like this indestructible market force that could never be challenged and like watching Anthropic surpass them in revenue in this in this past ⁓ this past couple of months here and seeing OpenAI like show some ⁓ need to focus by dropping their, you know, video and image editing investments and like really narrowing their scope. ⁓ And like show that vulnerability in their their operating. like it just shows how quickly, like, I don't want to say the mighty can fall because OpenAI is incredible business with like ⁓ know, ⁓ the most impressive revenue trajectory of all time. but their uniqueness at the very top of the market is being challenged. Like it it does. there's like a duopoly, if nothing else, ⁓ going on in the frontier model space, and that's like Lindsey DeFalco: Snow folly. Make it split the high. Bob Moore: You wouldn't have said that that that that was ever possible even a year ago. So who knows what a year from now will will give us in terms of that. And like, you know, one of these companies we're talking about all the time might be the might be the next day OL. Lindsey DeFalco: Yeah and Yeah. and you know, a ⁓ a kind of perfect connection to what we've talked about before, which is that they're both going and trying to s like make themselves less replaceable and make the switching cost higher by embedding themselves into these enterprise workflows with the services arms. and you know, it's funny because it's like software one ⁓ one is you have to be sticky. Bob Moore: Is it? Lindsey DeFalco: and you know, in this in the age of everything being so easily replaceable with with AI, it's like that's the path they're taking is getting embedded into your ⁓ your workflows such that it would be painful to have to switch it all out. And you know, if you don't have a network effect business by nature, then like that you you have to find some other way to to get stuck inside of organizations. So Bob Moore: Yeah. Yeah, and those th this is where the like even in PLG, right? Like PLG and enterprise have been oil and water forever. P PLG is in the realm of ⁓ you know, SMBs, and B to SMB and kind of like lower market, ⁓ prosumers and consumers. Like the realm of PLG really has always been in that area. And very few have kind of like crossed the chasm into also selling into enterprises, and all of those Lindsey DeFalco: Mm-hmm. Bob Moore: abandoned their PLG motions and hired big sales teams when they wanted to sell into enterprises anyway, right? And kind of went back on on the mantras of their early days. So including us, by the way. ⁓ we totally, ⁓ know, we're we're ostensibly a PLG business on the low end of our product, but have like very much an enterprise sales team ⁓ selling into enterprises. So like ⁓ big slow moving ⁓ hard to crack nature of enterprises is actually a feature, not a bug at this point. Lindsey DeFalco: Right. Bob Moore: Because it might be the only area in the market where you can actually create defensibility and stickiness in a world where thing ⁓ businesses that move quickly and procure quickly and ⁓ you know have lower switching costs and are more agile, just like it's almost impossible to become sticky inside of ⁓ inside of those environments. ⁓ so again, like, yeah, yeah. Go ahead. Lindsey DeFalco: That's true. It's like it it having enterprise customers itself is more sticky because literally just because of the nature of the fact that the process to get them like close a deal and get them onboarded is such a long cycle. And they generally Bob Moore: ⁓ it's crazy. You ⁓ you know, I'm not talking out of school here. Some of our own stats from CrossFeam, like our our GRR gross retention rates, so like how many of the dollars that we had from customers 12 months ago do we still have now? In the enterprise segment is 100%. It's like a it's a perfect score. ⁓ as you get into like upper mid market and kind of smaller enterprises, it's in the 90s. And then when you go to our connector tier and like people who plop down a credit card and sometimes do it without a sales rep and are like using the lower end of our product, it's like the 70s. ⁓ so it this this is not an unusual curve. ⁓ and it's ⁓ it's a gift and a curse, right? 'Cause ⁓ you get the volume and in our particular case we need to serve both ends of the market for our product to to be valuable. ⁓ but man, you can see you can see it plain as day without AI. And as with everything else, I think AI just amplifies ⁓ the good and the bad on both ends of that spectrum. And I think that's that's that's gonna be very, very clear in in A to A and in A L G. Lindsey DeFalco: And what's what's interesting is I think like some companies can employ PLG more successfully with enterprises depending on the nature of the business. So like Callendly was always a great example of this. Like because their business is, you know, initially was creating a calendar link for your own calendar, you're not uploading any business data. ⁓ that you need approval because it's coming from your you know, your CRM. Like sure calendars themselves can be sensitive, but it there's a there's like a single player mode essentially. and so they were able to ⁓ and there was a big narrative around this ⁓ Bob Moore: Mm-hmm. And so they were able to do Lindsey DeFalco: when ⁓ Annie Pearl was there as as CPO was, you know, enterprise PLG and how they were taking that one user that would sign up in an enterprise account and turn that into, you know, a massive enterprise deal and how a lot of their their enterprise deals started with like one person with their Gmail account. ⁓ and so ⁓ I do think that, you know, the the anthropics and open AIs ⁓ of the world benefit from that as well because You know, I'm sure there's people with their Gmail addresses at every single enterprise ⁓ who have signed up to use open AI or Cloud. And so, you know, they have the benefit of at least getting a foot in the door more so than a business like Crossbeam where a single player mode is significantly more difficult and you're not gonna get like a random personnel. Yeah. Yeah, yeah. ⁓ yeah, exactly. Yeah. And so, you know, for for tools like us, it's even more of a Bob Moore: Yeah. Totally. Or we literally block it. You can't sign up for crossing with a Gmail. Yeah. There's a lot of tools like that. Yeah. Yeah. It's a great point. Lindsey DeFalco: a non option, because of the nature of the business. So yeah. Bob Moore: Yeah. All right. I always think we're gonna run out of stuff to talk about on this podcast and then like a week goes by and there's a new there's a new acronym and there's a new thing that somehow we can talk about for an hour. So think we're a time here, but thanks Lindsey as always ⁓ and great discussion. It's it's so cool to see ⁓ all things we've talked about build on each other like a layer cake as we get farther and farther along. Like it it you can see how it's all converging and Hopefully for those listening along, you know, you're ⁓ you're along on this ride with us on ⁓ building out this knowledge stack. Lindsey DeFalco: Now we gotta go write that ELG and A L G playbook blog post. ⁓ Bob Moore: Yeah, hey, yeah, one one parting note, because I forget I I sang this on Slack and we didn't get to talk about it, which is ⁓ I was in a conversation with a founder earlier this week, and we were talking about product market fit and like the evolution of what product market fit means ⁓ in the of this kind like ALG. ⁓ And I made point in the ecosystem-led growth book that resonated really well. And I think is one of the main things that people like have have quoted back to me about it, which is I make this point that when people classically think about product market fit, they think about the market as being stationary. It kind of is this thing that is what it is. And then your product is the thing you have control over. So, you know, you're trying to get this alignment between where the market is and where your product is. So what you do is you iterate on your product. You do lean startup stuff, you build, measure, learn, ⁓ you pivot, you rebuild, you do ⁓ you know, betas and design partners and whatever. And you find this product market fit by moving your product around and finding where that market demand is. However, as you get into markets that are very dynamic and very bleeding edge, what people forget is that the market is moving to. And you can get into these situations where companies that were stubborn and stuck with a product or saw where the puck was going stick with an idea, take the time to develop it well, and then the market actually moves into where they are, right? So that you could almost argue like the super base is an example of this. and there are also companies get really unlucky where they find this incredible moment of product market fit. ⁓ but the market is like moving so rapidly that the market just kind of leaves them and they stay where they are with their product and they wake up one day and they say, What happened? ⁓ my favorite example of this is ⁓ you remember that app Clubhouse during ⁓ the pandemic where like everybody would join these like live conversations and it was like the hottest thing ever and raised all this money. And then like the pandemic ended and Lindsey DeFalco: Yes. Bob Moore: It just kind of went away. ⁓ and I I think like perfect example, right? The market was moving very rapidly. There was an incredible product market fit that was ephemeral. And ⁓ they had their moment and you know, they weren't moving while the market moved, or it might have been impossible for them too. But anyway, so what happens to that in this world of ALG, if that's like the the mental model in PLG? And I think it's almost like the market is moving so incredibly fast. That it's almost impossible to sustain PMF, where like I use this phrase, Schrdinger's product market fit, right? Like, you know, Schrdinger's cat where like, is there a dead cat in this box or not? ⁓ There and there isn't until you open the box and you observe whether it's there. I think it's almost like ⁓ people's of product market fit is going to be so fast ephemeral that it's like as soon as you observe that you have product market fit, like The engines are revving and you're like, we have it. We have the product market fit. It will be gone. it like you almost have to have your it's harder to build a company because it's not about overlapping your dot with the market dot or going where the market is going and then having them overlap. But you need not just position, but you need velocity and probably acceleration of your roadmap to match the position, velocity, and acceleration of where the market is. And it's only the companies that are able to like. Follow the market and not know where the puck is going, but ride the puck and be on the puck that are going to be able to build businesses that are that sustain for more than a couple of months of like being the hot next thing until they get they get laughed or the market leaves them. ⁓ so I think there's a little bit of like, yeah, Schrdinger's PMF is ⁓ has arrived and it adds dimensionality and a lot of a lot of new physics to ⁓ to the way we think about product market fit. Lindsey DeFalco: I don't know Yeah, yeah. ⁓ I was dying to know what you were gonna say when you messaged me that you wanted to talk about Schrdinger's product market fit. So ⁓ yeah, and it's like you know, you look at companies like OpenClaw, Clawed Bot, ⁓ that lasted what, like a week before it was acquired. ⁓ Bob Moore: Ha ha ha. Yeah. ⁓ yeah. That's a great ⁓ amazing, amazing point. Right. And like good on good on the folks who find exits and ⁓ and and liquidity and like maybe maybe the right thing that was supposed to happen happened there, which is that that founder or that technology is now part of some bigger platform strategy somewhere where they can matter over a longer amount of time. ⁓ however, the product itself, yeah, it's like it's a blip, right? It's like Lindsey DeFalco: Yeah. Bob Moore: They sat there and the market kind of zoomed right right through them. And there was a glorious moment. But man, we're talking weeks, weeks. Lindsey DeFalco: Yeah, I mean that was like a a viral tweet to acquisition and like that that has to be like a record time frame for that. yeah, and it is interesting. And I've I've even like ⁓ you know, it's interesting if you're competing against, you know, the model companies versus not, but also like, you know, kind of who isn't competing against the model companies ⁓ at some point in some way. ⁓ but yeah, I mean, there's a bunch of stuff right now that ⁓ literally the conversation internally is like Bob Moore: Yeah, you're right on that one. Lindsey DeFalco: ⁓ let's just wait 'cause like, you know, Claude's gonna release that. Like we could build an internal tool, but like they're probably gonna release it in the next two months. And if they don't, then ⁓ then we can think about building it. So it yeah, like the whole the dynamics of how quickly incumbents can ⁓ can do things too is is another piece of the puzzle that's interesting there. So yeah, ⁓ fascinating. Bob Moore: Yes, but I guess I don't know. Incredible. Cool. Glad we got to that. We stuck the landing on time. Look at us. ⁓ one hour on the dot. And ⁓ until next time, Lindsey, always a pleasure. Lindsey DeFalco: Yeah, likewise. See ya. Bob Moore: So yeah.