Loading summary
Transcript67 lines
- [00:00]
Jason
It is the compounding knowledge of an entire organization, living, breathing organization, that is growing all the time. That is what AI is in the future. That is all it is. It is not OpenAI's LLM. It is not anything else. Without this knowledge, AI cannot perform for your business. There is not going to be a magic agent that comes in on one part and solves your business. What you're going to do is you're going to add a digital version of a human and create pain and entropy because that is not going to be connected to the rest of your business. So having a hundred agents in a company is something that. That is all they could figure out today, and that's what they're selling. But if you fast forward a few years, that is not the solution. The solution is can the company first have all of its data integrated into a single place so that everything can be understood. What, by an agent?
- [01:00]
Host
All right, so I wanted to do a little bit of housekeeping before we dive into the episode. This is one of the best episodes I've done with Jason so far on AI. We really talk about one, just how things have progressed and really from the customer's perspective. We are going to talk a lot about how the best executives and CEOs are thinking about AI implementation and adoption inside their companies. We talk about, you know, how we see this evolving. We talk about what's not working, we talk about what, you know, the common problems are for people. And some of it's just they don't even know. They hear I want to do AI in my business, or I want to put my business on AI, but they don't really know what that means. And so in this episode, you'll learn how we think the best CEOs and executives are one thinking about AI, but also implementing it into their company. But before we do that, I just wanted to let you guys know of a Slack community that we're launching. If you're a CEO or an executive looking to bring AI into your business, it's called the AI CEO Network, and it's being organized by the team at Foster as a space specifically for CEOs who are leading AI adoption inside their companies. The purpose of the community is to create a focused environment for sharing what's working, learning from others, and, and talking through the kinds of decisions leaders are making as they bring AI into operations, product, customer experience, team structure, et cetera. The community is hosted on Slack to make it easy to connect directly with other founders and executives. Inside, you'll find discussions on topics like Strategic planning, building your company context, expert insights, and what it actually looks like to turn AI from an idea into a system that runs inside a company. If you want to join, reach out to ceonetworkosterai.com to get on the notification list for when the community launches. That's ceonetwork.com Also, in the next few weeks, we're going to be launching Foster to the world. And if you want to get on the wait list for becoming a customer of Foster, go to FosterAI.com to request early access. I really hope you enjoy this episode. So I thought the best place really to start would just be to those listening. We did an episode back in April, episode 381, if you want to go listen to that. It's been about four or five months since then, and I feel like every time we sit down and talk, just what you come back with, you can just see how 1. I don't know if it's how quickly things are moving, but how people are starting to think about AI maybe differently than they even were five months ago. I know that's definitely the case for you. So I figured we would just kind of start there. Where have things evolved, maybe in your mind at Foster, just in general over the last five months?
- [04:00]
Jason
Yeah, I mean, obviously we've gotten much, much smarter just in terms of being so laser focused and doing nothing but evolve sort of what Foster is how we think about it, obviously implementing it into companies and realizing the use cases, how people are thinking about it. Because, you know, we came from a place where we had been using Foster or our version of Foster for almost two years at that point, you know, when we, when we, when we decided to launch it and, you know, you, you start to build sort of a predisposition around how other people are going to use it or a preconceived notion around how other people are going to use it, which isn't, isn't wrong, but it is, it, it's. It can't take into account everything that you're going to encounter. Um, and so what we learned, really, I think if I just wrote, you know, wrapped that up into how I really think about it today, is every person that's using AI has a view on how they're going to use it based on what little information we've all learned about AI right now. Right. And so if you're a CEO in a company, you started using ChatGPT and you started having ideas and then you started reading and you started learning that the noise is so strong right now. There's so much fire hose of information coming at people that are actually trying to use AI that you think the whole world is evolving daily, so it's impossible to keep up. And so you either give up or you go so deep down rabbit holes that as a CEO you just don't know how to implement the thing right. So you choose somebody that's smart, that you rely on and say, figure this out for me. And so what we, what we realized is that a lot of those companies have sort of just like one problem they're trying to solve, mostly like to start because they're like, if we could just get AI do this, if we could just get it to do this, then we would be good.
- [05:55]
Interviewer
Right?
- [05:55]
Jason
And they're still thinking about it mostly, and not everybody. I'm just saying in general, mostly like at the very front, like fraction of 1% of what AI is going to do in a company.
- [06:07]
Interviewer
Right.
- [06:07]
Jason
Just, they're just barely scratching the surface. And so I was thinking deeply about why, why are all CEOs not seeing the full potential yet? And it's really based around this idea that, or this like obvious thing that has happened where one massive company launched to the world, everyone knows what that company is and everyone uses AI through that lens, which is just the chat interface that OpenAI put out, which, it's great, we all use it and it is phenomenal. World changing without a doubt. But when that's all you know and that's how you're trying to figure out AI, then every decision you make is a comparison to how you do it in ChatGPT, not what is possible in my company and how do I take advantage of AI, right? So you start to figure out like, how do I take this thing? Like how do I take Google, right? Or how do I take anything on the Internet and how do I use it to my advantage for my business as opposed to that's a tool that was put into the market and people chat in there and it is amazing. But you see how much competition has come up around that and grok and clot and all these other options which are, which are all great as well, but people know it through that lens. And so what I realized is that CEOs really need to, and this is just an evolution that's going to happen. CEOs of companies need to reframe how they think about AI mostly and understand what is actually possible in these companies, not through the lens of these large language models. And what OpenAI put out that understand that is a huge component of it. That's part of the infrastructure of AI, but that is not the solution for businesses and how they're going to take advantage of AI inside their company. And so those two paradigms are existing right now. And what we see is that there is no incentive for the open AIs of the world to, to teach CEOs that, because they would realize really quickly they don't necessarily need OpenAI. Right. They just need a model.
- [08:14]
Interviewer
Right.
- [08:15]
Jason
And so since models now are becoming a utility and they're improving every day and that's going to be a race, you know, Elon Musk is not going to stop improving Grok. And so no matter how good Chad GPT17 is, Grok is going to be right there.
- [08:31]
Interviewer
Right.
- [08:31]
Jason
And so there's always going to be a choice when it comes to those things. So the end consumer, the user, the selling access to large language models as a user is going to be a business and they're going to crush it. But that's not the business that's going to solve your company's issue with using AI. So that's what I've been thinking about. And we have a lot of ways to teach CEOs that. But also what's been most powerful in all that is that it just has highlighted to us so much that there are are some companies out there, like what we're doing at Foster, that really understand that focusing on the end user today in terms of like selling them access to a thing like ChatGPT will fade very quickly. Because what has to happen for people to truly leverage AI is the infrastructure has to be built. And that infrastructure is where we play. And that the ability to build the rails of how AI is going to just flow through everything in the future is where all the value is. And that's where we have focus and we just believe, and we know a lot of other people that are smart and we believe that are smart, that feel the same way. And so we got a lot of evidence behind that and a lot of data to prove it. And so that's where we're focused. And that's what Foster does. It's really focused on the infrastructure of how you take advantage of AI, not the flashy new toy that just came out.
- [09:56]
Host
Well, I think that tees up the next question pretty well, which would basically be even if, like, let's just say for a second I'm a CEO that's come to Foster or that's come to you in general, just to talk about AI and kind of you said, I Think I have this one problem I want to solve, but maybe by the end of a conversation you've convinced me. Wait, I'm thinking about it totally different. You mentioned foster being infrastructure. Maybe just talk about where do we play in that infrastructure stack? Where are you actually describing as you say that, you know, this is the stuff that usually fascinates me once you start breaking it down.
- [10:32]
Jason
Yeah, I mean the, the problem that they want to solve is actually real. It's a real good starting point. Almost all the, all, all the time. So almost in everything, is there a.
- [10:41]
Host
Common thread of the, Is there a common thread of the problem or is it always different?
- [10:45]
Jason
No, it's the same problem. It's literally the same problem. I mean, I, I, at least so far we have not, we have not ran across a separate problem. We have not ran across somebody that wants to deploy an automatic agent that runs accounting.
- [11:03]
Interviewer
Right?
- [11:04]
Jason
That, that nobody's. Even though online all you hear is agents. There is not one company that's trying to go straight to. What they're saying is how do I, I have this data over here and I have this information over here. I would love to be able to use AI to figure out how to take advantage of that data or combine that data with this data. Right? So essentially the way most companies are just trying, which is smart and what they're realizing, the reason they're doing that is because every company that is trying to contemplate using AI right now, these CEOs, this goes back to kind of what I was saying is they're trying to envision how do I get all this information in a chat GPT. And they're not wrong. But the problem isn't how do you get it into somewhere. The problem is your data is messy today. Your data is not structured correctly today. And this isn't a knock on any company. This is all companies. We have an infrastructure problem, right? We have a problem where everyone over the last decade was taking advantage of technology in the way that we knew and basically creating folder systems that no one keeps up with that get so messy. They just drop stuff into files and they think, well, I've got it saved, it'll always be there. And it is. But there's no way to take advantage of that from an AI perspective. Not, not in a, not in an efficient way or not in a valuable way from where it sits today.
- [12:29]
Interviewer
Right.
- [12:30]
Jason
And so all this data, and that's in a good case, right? That's somebody who actually has like built a structured database and like has a good organization that, that manages documents. Well, that, that's probably still only about 20% of the companies. The other 80% are just the data. You can just, you know what it is, right? The data is just a mess. And so when you have companies that are starting from that point, it's hard for them to envision. They're looking for somebody to come in and say, how do I get all my information or my data from where it is today to here? And so they, they actually don't even think that's possible. So they move right to how do I start taking advantage of AI? And obviously the solution is get in ChatGPT, take one document, drop it in there, ask a question, wow, I'm using AI, right? They're not for, they're forgetting about the 10 years of knowledge and the 10 years of things that they've compiled and that value that's setting that right? Because they're just trying to take advantage of AI. And then every day they do that. What they don't realize is that every time they're using AI going forward, without the compounding of the knowledge or the integrations of the knowledge that's going to keep coming into one place, they're actually making the problem worse. So you're taking disorganized or unorganized or just chaos, right, of data, and now you're, you're basically escalating it by saying, I'm going to individually use AI to help me with all these individual pieces of data and I'm going to try to cram some of them together into places and use AI to help me solve a problem. But that is not creating a cohesive place where your data is being aggregated. So back to the question is the same problem exists and it is data, right? Companies need a way to quickly and efficiently suck all their data into a unified place, have that unified place structure and organize that data in a very specific way so that it can be used to interface with AI, I. E. A large language model. But the key there is, is when you do that, that data has to be interrogated, right? So you have, you can integrate data, you can take a database, you can connect that database. So this is kind of what we're doing. I'll just walk you through like a use case where a company comes to us and says, I have this legacy database and it has all of this information in it. We need to get that into foster so that we can ask these specific questions around this specific data. Because it's not just a little bit of data, it's millions and millions and millions of pieces of data. And that data is so unorganized, they're trying to manage it through spreadsheets and filters and things like that. And you just don't get any real insights. You get a lot of manual work. So you can take that data and you can suck it into a new database. But what happens if you suck that data into a new database in the exact same form that it exists today? You get the same shit that you have, right? It's no. There's no. It's like, you know, shit in, shit out. That's what you know. Everybody knows that when it comes to the data world. So you have to first take that data and you have to do something with it. So that is where Foster comes in. That infrastructure layer, we are converting that data automatically into a series of things that create relationships between all those pieces of data. How they exist in that current form and how they connect to the company and what they mean to the company, and where they fit into the company, including the people, the time, the locations, the what, the why, the how, all that stuff that is in that data, the metadata, gets connected to each other as that data is coming in. What you get when you do that is a graph, right? As opposed to a list of information. You get nodes of information that are all connected to each other. That is critical when you want to interface with AI, because you're trying to create context for the large language model to understand how the dots are connected, not to search through your document. So this is the biggest difference between what a system like Foster is doing versus a chat GPT. And I don't like. We're not comparing ourselves to ChatGPT because we utilize these tools, right? We, as Foster, utilize large language models like Chat GPT. We're not, we're not, we're not trying to compare. But there is an infrastructure layer that doesn't exist inside those that you must have as a company, and that is the metadata that is inside your data has to be organized, structured, and collected inside your system first. Before you ever ask AI a question, if you don't do that, all you're doing is taking a piece of your data, sticking it in a place where the LLM essentially searches it based on the question you ask, right? So that's why they say prompting is so important, because what is prompting? Prompting is the only way that you actually give an LLM context. It's the only way. So what, every time you prompt and that's how they. People are making millions of Dollars right now, being prompt engineers because they can ask better questions to give the LLM better context, which then wows the person on the other side, right? It's like, holy crap, it knows everything. No, it knows really well of the great question that was asked, how to find the words that would answer that question, right? So it is a search mechanism. But when you take that same idea and you structure relational data in a graph that is already structured with knowledge and information, connections, people, everything that you would want to know about a piece of data and why it exists in a company that already has context. And if you can retain that context inside of an organization and then ask a question through the lens of that context before you leverage an LLM, what you get is basically the pattern that we have as humans, right? When we have. When humans, we have a thought and we, we think. But what are we thinking to? We're thinking to our brain, which stored all these connections, emotional senses, memory, right? It's our memory, but it's tied to other things. Like in, in a human, we have, we remember how it looked, we remember how it sounded, we remember how it smelled, remember how it felt. But in a relation relational database, if you structure this graph correctly, this knowledge graph, what it has is a memory based on who, what, when, how, right? The time, the location. But if you structure that with also, say, the company's history, its goals, all those things, right? It knows that piece of data and how it is structurally related to everything inside that organization. Who said what, when, how, why does it matter? And what project is it related to and when is it due and all those things, right? So we create this graph, knowledge graph out of all that data. So that company that said, hey, we need to ingest this data, that is the first thing we do. So at Foster, what we had to solve was, okay, we know that we're going to essentially be, as a part of the company, company's responsibility to get people onboarded to AI efficiently and quickly. We're going to have to be really good at data integration. Well, for us, that doesn't just mean connecting Slack to Foster, because that's what OpenAI is working on right now, right? They're saying, let us just connect your tool to ChatGPT. That's great. But if you then ask ChatGPT a question about your data, what does it know? It's doing the same thing. And if you dropped a document in ChatGPT, it's going to take your prompt and search that data and try to give back to you the most relevant Question answer based on your question. But it is not taking that data, structuring it with all the relationships and retaining that knowledge. So what we believe the future is, is that we have to create these structured knowledge graphs inside companies. And so that's what we've solved. And this is what we talked about last time of creating the digital twin. We talked a lot about that on the last episode. This is the same thing. This is just the evolution of that. This is saying what happens after you first start the company and collect that data and you start creating that digital digital twin. You've got to keep that going forever. So what happens to that knowledge? That knowledge is being piped in. And when I say knowledge, all your integrations, Slack, Zoom, Notion, Airtable, Google Drive, any email, right. Calendar, all your information is connecting to foster. But instead of just connecting every second of every day, it's transforming into these relationships. And these relationships create a knowledge graph. And that knowledge graph is where all the value is for all companies in the future. And we'll talk deeper about that. I'll get into it more hardcore here in a little bit.
- [21:47]
Host
Let's get it. Let's, let's. We're going to get right back to that real quick. The question would be, real quick, you said, you know, 20% of their companies kind of have their data in a good spot. Maybe 80% don't, whatever. What are the 80% of companies need to do? Or like, when you're saying like, what makes a company kind of ready to do this? Like, what's the minimum viable structure of data that a company must have to even start here? Because I think when I talk to people and kind of what you said, they get so stuck. Just they're like, oh, we don't have good data. We've got like a Dropbox link and, you know, some QuickBooks and some Excel. Like I don't even know what to do.
- [22:27]
Jason
Yeah, I mean, the truth is my opinion on this has transformed a little bit or it's evolved a little bit over the last few months. And I think what's more important than anything is a CEO or a person that is the decision maker saying we have to be on AI or we have to utilize AI in a real way in our company. And we know that we are going to have to be an AI first company. That mentality of that CEO, it is the most important thing. It's like any, anything in a company, if that person isn't bought in and fully ready to do it, that has to be. So then the secondary piece is how ready are they as a company from a data infrastructure, that sort of thing. The truth is it doesn't matter as much if it's messy or clean as it did before. And I'll explain why that is because of a major thing that we solved which we actually believe might be as valuable as anything we've done. Just this one component that we built. Built. But what's really valuable there is that the data doesn't have to be clean or messy, but if they don't have enough of it, right, they're just not going to get enough value for the effort to, to, to have a big impact today. Now, that doesn't mean they shouldn't do it, because where we see the biggest opportunity is say, a new company, I'm saying we see the biggest opportunity for an individual, right, Is a young company that is, say it's got a say. It's you and I, right? Let's just rewind and let's say 10 years ago or 12 years ago, you and I, we were already like this, right? We were already digitally native. We wanted to use apps, we wanted to use whatever tool was out there. Let's say AI was there, then we were the candidates, right? To start this way so you can start and have nothing. And if you make your connections and start going, you're way ahead of the curve than having to clean shit up, right? So the difference is, isn't really so much how much data they have, it's if you have a really good company with really clean data, you're going to have an impact pretty quick. But you still got to get all that data transformed and then change your habits, right? If you're a company that has bad habits and not a lot of good clean data, you have to, it's. It's kind of worse, right? You have to, you don't have a lot of good data to start and you have to change how you've been doing it, right? So it doesn't mean you can't do it, but it's, it's less efficient, right? So the ideal candidates are companies that are digitally native, they're tech savvy, they have good IT infrastructure, they have good databases, and they have the mentality that they, they understand they've got to take advantage of AI and actually understand it. They just don't have the tools to actually make it happen. And that's where Foster can plug in. And then they will adapt quickly because they already have the mentality. And then you have these young digitally native companies that are Just starting. And they will skyrocket past most companies because they won't know any different, right? And this is just how they'll be built. And they will understand that having an infrastructure layer that takes advantage of AI from day one is way better than any traditional operating system that's on the market today. And they won't even question it because as soon as they get their data in there and they can ask those questions and build those reports and they won't start need to hire so and so to tell them that they won't need to hire somebody to create the next marketing flyer. They won't need to hire somebody to tell them the best ads to run on their next ad campaign, Right? It'll just be there. And so they won't think about teams the same way we do. They will think about teams as who can take advantage of AI, right? Who is really capable of taking this AI marketing thing that we do and running it using AI, not printing flyers.
- [26:20]
Host
And before we get to Knowledge Graph, just real quick, what's the simplest way to describe clean versus messy data?
- [26:27]
Jason
Structured database. So a lot of companies have structured databases where their data, at least the critical data that's happening internally is being organized correctly, right? So think like a SQL Server, right? And then they, and next to that, they'll have like a cloud storage document management system, right? So you have these two components that are usually inside each company and that's, that's where most of the critical data lies, right. Then you've got other companies. I mean, you know, messy. We know messy data, right? Everybody does. You got other companies that every person has a folder on their computer where they save all their Excel files, right? And so, and you could expand that. All their PDFs, all their things, right? And then they've got, they've got people that some people use this system. Like, let's say One department uses QuickBooks. And that department keeps most of the things in a relatively clean place. But they're working on so many projects. Half of the things that are happening in QuickBooks are downloaded PDFs or spreadsheets that they're using on their own computer. And that's actually a separate department from marketing. And marketing uses another tool called Figma and something else, right? And Adobe and their files are mostly in a file system, but most of them are also on their own computer. And then you compound that times. Spy departments, right? That is messy. You would have to literally go to every individual and say, please start sharing with me all your files. You know, if you really want to have valuable data.
- [28:03]
Host
Okay, you said the knowledge graph, one of the most important components of all this. So why.
- [28:10]
Jason
I think I'll touch on why. But I think what's important is to paint the picture a little bit of how we get to the knowledge graph. But, but I'll just, I'm going to leap forward because I do think it's valuable for people to understand right off the gate what, what it is and why is it important. But what we talked about, of that data getting structured over time as that data's piping in. So if you're a CEO and you're imagining I've got my different systems I use and we'll just use three, for example, I've got email, I've got a Google Drive and I've got Slack. Let's just. And there could be 20 other, there could be a hundred others, it doesn't matter, right? And then in addition to that you've got your internal server where your company data is flowing in. But you've got. Or we'll use Zoom. Zoom's a good one. So we'll use Google Drive, Zoom, Slack and your internal database. So you've got these three layers of data that are happening every day. So you've got communications happening in Slack, short term, right? Back and forth, problem solving, cast. Excuse me, anything that happens like that. Zoom meetings, critical meetings, right? Critical things being said, critical check ins, critical follow ups, critical lots of things, right? Tons of amazing data. Then you got Google Drive, where and if you use a different system, it doesn't matter. This is for example. But let's say you got Google Drive, all these documents are getting stored in there, they're getting shared with other people. There's critical information in there and they've got your internal database, right? So all that data is right now living in those places. If you use a system like Foster, that is pulling that data in, converting it, transforming it, all that data becomes one meshed knowledge graph. So let's now envision you have a knowledge graph that is there as opposed to being static, where that data keeps going into those places. Every day it's feeding that knowledge graph. Every day that knowledge graph is building more and more relationships between everything that is happening, what is being said, who is saying it, what is happening, what got executed, what got recorded, what project got completed, what, what box got checked, right? Everything that happens in a company that is in flowing into one of those databases or into one of those systems today gets aggregated into this knowledge graph. So what is that knowledge Graph then that knowledge graph, when we talked about digital twin before, it is more than that, right? What it is, is, it is the compounding knowledge of an entire organization, living breathing organization, that is growing all the time. That is what AI is in the future. That is all it is. It is not OpenAI's LLM, it is not anything else. Without this knowledge, AI cannot perform for your business. There is not going to be a magic agent that comes in on one part and solves your business. What you're going to do is you're going to add a digital version of a human and create pain and entropy because that is not going to be connected to the rest of your business. So having 100 agents in a company is something that, that is all they could figure out today. And that's what they're selling. But if you fast forward a few years, that is not the solution. The solution is can the company first have all of its data integrated into a single place so that everything can be understood by an agent.
- [31:41]
Interviewer
Right?
- [31:42]
Jason
The agent has to understand what's going on. So there has to be context. So what that knowledge graph is, is the context of your business, but in a static state and a forever state, right? So that the difference of what we're talking about is the difference of having a context engine and not having a context engine. So what is a context engine? That is how you get to a knowledge graph. A context engine is the pipes, right? So if you think what is the fuel of an engine in our world? It's data. We've got to connect those fuel lines to the engine. So we have to first tap into Slack, Google your database, right? We convert that fuel like we talked about, we convert it into the form that can be collected into that knowledge graph. That context engine is running all day, every day. So when all those actions happen, we're just fueling that knowledge graph with that context engine all day, every day. That is how companies are going to be able to interface with AI and deploy agents and, and have 100% assurity that it knows what the hell it's doing and how it should be doing it and what is happening. And that's, that's not even where all the value is. When I explain what the actual value is at the end, I think most CEOs will understand that. They'll, everybody will get it. But if you don't start with this understanding that your data can flow into one place, it has to flow into one place. And building a knowledge graph in a company is the only way to truly Leverage AI inside your organization, period. And if somebody can challenge that, I would love to hear the point on it. Because no matter what AI systems you build, you are going to have to tell it something new that happened, otherwise it will not know what just happened.
- [33:47]
Interviewer
Right.
- [33:48]
Jason
And so how are you going to do that? You're either going to have to tell it individually with a prompt or you give it the context automatically.
- [33:56]
Interviewer
Right.
- [33:57]
Jason
And so what we're talking about is building the context flow of every organization, building that knowledge graph by using the context engine to just pump that data in. And so that's, that's sort of the, the gist of the knowledge graph. And then I'll explain a little bit more and I'll let, let you keep going, but I'll get back to one of the critical things that I think people are missing on where this leads. When you have a knowledge graph, keep, keep going. So a knowledge graph, you can say, wow, okay, it's collecting all of our data and it's all cramming it into one place and cool. It's structuring it. Well, how's it struct? We have solved a very simplistic way, although extremely complicated, to get to, to build automated data pipelines. So what used to take months can now be done in or weeks can now be done in minutes, which is insane. Where we can leverage foster to map to where the data is today and understand it and build the relationships itself. It creates a schema, right? It maps it. This is all like technically how a data pipeline is built. It breaks it up, it chunks it, it understands it. That can happen now very rapidly. So you, you do that, you're pulling all this data in in a very structured way that then deals this knowledge graph. But then you say, okay, now I'm converting my data quickly. I'm a CEO, I can convert my data quickly. Great, let's clean it up. I can chat with AI and it's going to pump it all in there. Then what? I've got a knowledge graph. Okay, great. What does that do? Well, it's not what the knowledge graph does, but now you have a place that you can interface with AI. You can have it take automated actions, right? These are what people are saying, agents, right? It'll take the action for you, it will send the email, it will do the follow up, it'll create the thing, right? It'll start to act like an employee. That's great. But where is the real value coming from? This knowledge graph is not going to grow just because of your data Meaning the data you're pushing in you at that point are going to be using AI. You're going to be asking questions every day inside Foster. More importantly, you're going to be getting an answer using AI and you want all those answers to be coming from the same place so that you get unified knowledge share, right? So if you have a team, it's all happen, happening in one place that's also super powerful. But what people are missing is that the value of compounding your knowledge in this way doesn't come from the data you're pushing in, it comes from the question and the answer that you get back every single time you do it. Think about how many times you, we as a human have a great question in our mind and we get an answer back and then we take an action that none of that is recorded, right? So imagine if now AI, which is what is happening. Foster, every time you ask a question and you get an answer, the AI itself, that knowledge graph just gained context around, oh, that person had a question, oh, and this was the response, every single one. So it isn't the value of a knowledge grab isn't in the data that's coming in, it's in the questions that get asked of that data and the responses that get. You will get 10 times more productivity out of the data and out of your ability to understand it. Because every time you ask a question, your knowledge graph is going to get smarter and have better relationships, not only of the data that's in there, the questions that people ask and the best responses that come out of that, that drive results. And we're talking about in a, let's say it's one day, take that same concept and do it over the course of a year. What is happening is that knowledge graph is going to grow so smart because of the interactions that happen. That knowledge graph becomes a proprietary model of the company that no one has, except that company. How did we get here? What decisions did we make? What knowledge did we gain? How did our team think? How do we think? How do we make decisions? And if you're a company that's doing that and you have success, this is what is going to happen with knowledge graphs. Knowledge graphs themselves are going to become the new gold of the world that this is what, what is value? Because this is one of the little secrets that they're talking about. But they don't try to highlight this right now. Although Nvidia did just put out a really good, Jensen did just put out a really good paper on how the micro models are actually outperforming the giant LLMs and you go, well, that's crazy. No, it's not crazy at all. It's logic. More knowledge of the entire world is not going to help your business, right? What your business needs is specific knowledge. It needs specific action, it needs specific questions and specific answers. So if you've been doing this in a company for say a year or two years, and you have this knowledge graph on specific things related to your industry, what you have is a specific knowledge graph that is primed and ready to train a private model, a local LLM. You don't have to do that, but you could, right? So as you're, as that knowledge graph is growing, you could just be piping that information straight into your own proprietary LLM and you go, well, where do you get an LLM? They're open source now. That's when Deepseek put out their open source model. That opened the world to. Now there's plenty of open source models. You can download them, you can host them yourself. There's a lot of companies that are going to go this way. It makes it very secure, very private and you can control your whole world. But let's say you don't want to train a model. You're just like, okay, great, but I'll, I'm happy having my knowledge graph and tapping into one of the more powerful models like Grok or OpenAI to basically feed back to me the answers. But I still want my knowledge graph. Well, now you could license that knowledge graph. You could say, I want to take components out of this that we do really well and I will sell that or license that to the company that is providing specific models to companies, because that is where all this is going. Every company in the future will be buying a tuned specific model for their industry, for their business. And that will keep getting fine tuned until there are specific models for like say a real estate company. It'll just be a perfectly tuned model. What type of real estate do you do? Well, we just do brokerage. Here's the model that does that, right? It won't be throw all your documents into open AI. It'll be, here's a model. And so you go, what does that mean? If you have a model that runs a business. If, if you have an. If, let's say you had a small private version of a chat GBT that was tuned for your business, you would already have the environment and the infrastructure to automate anything, right? You would be able to ask any question, you would be able to constantly Train that model to get smarter and smarter and smarter. But what is still missing? If you had that model, you still have to have that context engine and you still have to have that knowledge graph growing, you still have to have that connection. So this layer that we're building is not a one and done thing. If you ever want to have a model that's learning and getting smarter at your business every day, you have to be building a knowledge knowledge graph. And the value of that knowledge graph is going to come through the questions and the responses that you get from the LLM over time. And that compounding of that is going to be infinitely valuable. And I compare it to. It's not really any different than how a company like Google was built. Google needed aggregated data into a place so that it could take advantage of it and give responses back, right? And what Google did in early days was it went out to the market and it bought website search, web search from companies or websites that had search bars, right? So if you were a company that had a lot of traffic and people were coming there and searching and asking questions or, or getting an answer or searching for products, anything on the Internet back then Google would pay those companies to buy the search, buy the search results, right? They needed to buy that compounding knowledge. And once they got that compounding knowledge, look what they did with it, right? It's not that different. What we're saying is that you are compounding your knowledge, but you don't have to go out and buy it, you already have it, right? You just need to compound it. You need to compound all those searches into one place of exactly how you do it. You need to compound all that how and why and what you do in a company into one place. And the value of that is what is going to power AI in the future. That that black box that people would be creating that no one can see but them is where all the value is. That is how one, you protect your company from being crushed by, by AI. That is how you, you basically build a fortress. And you know that at the very least you have a way to do it that others can't see. And because you're compounding that knowledge, AI is only going to get smarter. Your company's only going to get smarter and you're going to have an advantage. And so you'll be able to leverage it one way or another. Either it's keep it private and crush the competition or have something very valuable and license it to other people or sell it to OpenAI because they'll all be producing private local models. All these companies that are right now selling access to their giant model in three to five years, they're all going to be saying, here's a model for this, here's a model for this, here's a model for this. And they're going to be selling them to companies. But you say, well, how are they going to build those models? They're going to need people like us that collected the data correctly for that specific industry and then hand it to them.
- [43:59]
Host
Oh, man. Okay. And so if you said real estate brokerage. So let's just. I'm kind of going with you for a second. So there's been a knowledge graph of a real estate brokerage created, and now I'm wanting to start a new real estate brokerage. I go look at some LLMs that are available. I say, oh, there's one. It's a real estate brokerage. One show up to the office, day one. What have I actually bought? Like, because on one end you could say, why would I want to give my knowledge graph up? That's all my secrets and that's why I'm winning. Why would I want to share that with the world? Maybe the answer is because you could get paid a lot of money to do that.
- [44:38]
Jason
Yep. But.
- [44:39]
Host
But let's just say somebody did. They said, I'm willing to share my knowledge graph and take in royalties or however it's paid. What do I.
- [44:46]
Jason
It's a great concept.
- [44:48]
Host
What do I as the customer get the first day I show up with my new real estate brokerage LLM?
- [44:56]
Jason
So just think like Chad gbt.
- [44:58]
Interviewer
Right.
- [44:58]
Jason
If you get the same thing you get with Chad gbt, Except with Chad gbt, you have general knowledge, general that. What if you had a, a place that you could go chat and you knew it was the ultimate guide for business success for brokerage, what would you ask? I mean, what would you ask? That's. It's a hard question, but that's.
- [45:22]
Host
Yeah, I would say, like, how do I hire the best agents in my market? Right.
- [45:27]
Jason
And so then. Or you would say, what. What is like, give me a, give me a list of steps to complete.
- [45:32]
Interviewer
Right.
- [45:33]
Jason
And people go, oh, but you can get that from, from ChatGPT. You can't get the same thing that a brokerage that has been crushing it has actually been doing every day that they have not been giving to GPT. Right. So what you get is a playbook. So here's the difference is like every NFL team has a playbook, right. They all have playbooks why is there one that wins? And sometimes why is there one that wins for like five years? Like the New England Patriots or something, right? Like, because their playbook is better. And so you take that idea of the better playbook and then compound it with AI. And that's where I keep going back to. Some people aren't going to get this. So I'm going to, I'm going to keep reiterating it, because until you actually, like, process this in your brain a couple of times, a lot of people don't quite get the value of if you are compounding knowledge of what you're happening, is happening in your company into a knowledge graph and you ask questions there through to AI, that then gives you a response back. And right now people do that in ChatGPT and they go, oh, my God, Chad GPT gave me the best response and it said this and I used it and I built a business plan and I did this. And okay, what happened to that question and that answer? Who got, who got the benefit of that? You think you did or the person did because they asked the question and got the response. But who really got smart was Chad GPT. Because Chad GBT owns that question and it owns that response. So whose knowledge graph grew there? You got an answer in a silo. They got a knowledge graph, right? So when they go to train their next model, guess what? They got all your questions. So here's the way to simplify it. Every time you go on the Internet and you go to log into a website, what does it make you do for security? The pictures? No, the sec. You know, the. Am I human? Why do you think it does the am I human? I mean, you may know this. I'm asking you. Cocky. Don't. But you know, why does it ask you to do that? Am I human? Because we think, oh, it's a security check to make sure it's. I'm a human. No, People are really smart when it comes to technology. What they need to do is they need to get you, the user to take a step to train their data, right? So what you do is you go, bus, bus, bus, motorcycle, motorcycle, motorcycle, bicycle, bicycle, bicycle. What you're doing is training their visual optical AI to understand if it was correct in predicting if that was a bus or a bicycle, right? So behind the scenes, that program was already running, saying the AI was already trying to figure out if that was a bus or a bicycle. And the human just validated it for them. Right? And so that's what we are to OpenAI that's what every company is that chooses to use. And again, I hate to say I love ChatGPT and by the way, and I use it a lot for specific things that we do at Foster. But this isn't a comparison. But what their business model is right now is to get. Why do you think Sam Altman says let's just give chat GB55 to everybody? Because they need all the sheep to get in there and follow exactly what they're doing. Because every time those people ask a question, including me, and get that response, they are building the most powerful general model because all the humans are in there doing this every day. That's not going to stop, but that is not going to power your business. That is not going to give your business an advantage. And if you lump your business into that, what I would challenge people on, and this is, this is for any of the large language models from a business perspective, if you lump yourself into that, you are not creating advantage. You are getting sucked into a hole that you are never going to get out of. They are going to eventually own your business or your business model.
- [49:22]
Interviewer
Right?
- [49:23]
Jason
And there's a lot of people like that can't be. That's not true. Don't underestimate the power of them being able to collect your data very quickly just based on the basic things of you asking questions and getting responses.
- [49:35]
Host
So that would beg the question that a lot of people ask. If I sign up for a Foster account, why? What, what's the difference between Foster owning it or chatgpt owning it?
- [49:45]
Jason
Like what is we don't own it. That's the difference. So every Foster is a multi tenant platform and we did that specifically so that every single user has its own instance of Foster. Every single user has a place that they go do this. Every single user has a private server. So we have set this up for this exact reason, to give companies a place to go operate and own their world. What we want to do is provide the rails that get them there to that world. But we are not trying to own their data in their search or combine it with everybody else's data. This is to create these buckets of silos so that companies can do it, but somebody has to do it right? Somebody's got to create the place for them to go do it. Are we going to benefit? Absolutely. That we're going to benefit for them to take advantage of this for themselves, but for a to take to stay in control of their data and stay in a place where they can own their proprietary knowledge graph. So you go, and this would be a next logical question. It's like, okay, well, who owns the knowledge graph? It's their data, right? It's their knowledge. It's how they do it. If they want to go do something with that, that is where we come in. We are the path for them to take that and actually do something with it, right? So our goal, my goal at some point in the future will be to help companies take their knowledge graphs and do more with them, right? How do we get them from where it is today into where it can be? And that includes setting up, training local models, right? You'll have an knowledge graph. We can easily set up a local model and start training that. I think that in the next probably 24 months, that'll be a normal thing where you're piping in context through a context engine. You've got all this data flowing in. You've got a knowledge graph being built. And that knowledge graph is training a model. And that's just the cycle cycle that happens.
- [51:36]
Host
Will it show who built the knowledge graph that you are or the LLM that you are basically using? Meaning it's different if Ryan Sirhant's amazing brokerage that's nationwide builds a knowledge graph versus three guys, you know, in a local office here in Fort Worth that are doing a couple deals a year. Well, I know I'm buying. And maybe this is a question. No. How do I know how good the LLM is and where it came from?
- [52:08]
Jason
This is the best part there. So if you go look at training data on LLMs, right? So they put it out like, every. It seems like every week. Now, I don't know what the cadence is, but at least once a month, every LLM, so Grok, Claude, ChatGPT, all of them, there's a list of like, 15 of the big ones, right? They all do a test where they test the strength of the model. Meaning, like, what does it know across a series of categories? And because they can use AI now to test the AI, they basically can just test it and say, how does it do for medical? How does it do for science? How does it do for biology? How does it do for writing? How does it do for math?
- [52:49]
Interviewer
Right?
- [52:49]
Jason
They can do all this. So what will happen is those models will have to prove that they know what they know. And so you're exactly right, though. Let's just say somebody is known to be a proven operator. Or let's just use football again as an analogy. Let's just know that this model is the New England Patriots. Of whatever run they're on.
- [53:10]
Interviewer
Right.
- [53:11]
Jason
When they were good. I know it's not now, but back when they were Tom Brady. Let's just say that that model is known. Of course it's going to be more valuable. Not necessarily because it's better than some of these other ones. There might be a company that actually has a better one, but it hasn't been proven right. So somebody will pay more if you have a proven model. And what is a proven model? Proven model means there is an actual company on the other side of it with proven success. And so you want to do it like Blackstone. Here's how you do it. You want to do it like whoever. Pick. Pick whoever. I'm just throwing out like different people, but you know, pick any massive company. But the, the real results will come down to a test. These models will have to actually do what they say them do, you know, and I think based on what they say they do is where there'll be more value. Now this is another thing. I'm not saying every single company that builds a knowledge graph is going to be able to monetize it for outside, but what they will be doing is monetizing it inside. So you still will be taking advantage of it internally to build a moat that no one knows what you're doing. And if you are really successful and you compound that long enough, you will have an advantage. But at the very least, even if you just said we're never going to monetize an LLM or knowledge graph, we're never going to. We just want to keep it silo because we want to keep all the secret sauce and we want to do is still what is required to survive in where we are going with AI now, nobody knows if that's going to be five years or 10 years. That what you can guarantee is that the rate of change that has happened and how AI is learning and then the fact that on the back of this is going to be coming like robots that also know this stuff, is that if you don't start down this path, your business will not look like it does today in 10 years. And so the question is, are you willing to take the risk to not transition your company to the largest technological shift in human history and just bet that I can keep doing using my spreadsheets and QuickBooks or whatever you're using.
- [55:23]
Interviewer
Right?
- [55:25]
Jason
I don't worry about that because I'm not trying to convince companies that want to argue that point.
- [55:29]
Interviewer
Right?
- [55:30]
Jason
Companies that want to argue that point, they're not the customer Right, but those are the companies that'll get swallowed by the rest of this wave. So I'm not trying to predict if it's in a few years or 10 years. But what I do know is that this transition is happening and not just because things like Foster exist. We use this every day. We see what happens inside Foster when we can do something like build automated data pipelines. Right. So we can build automated data pipelines. And people that might listen to this, that are more technical on the engineering side, if there are any, will understand how crazy this is. But used to, you would have to pay a data engineer or a team of data engineers to map data from one system to another. If you wanted to build an integration or a pipeline of data, easy to connect an API. But if you want that data to flow through and do something, you have to first know what all that data is. And I mean every line of data. Well, what if that data is messy? What if it's not categorized correctly? What if, what if, what if?
- [56:34]
Interviewer
Right?
- [56:34]
Jason
There's a million questions. And if it's millions of pieces of data, that's where people are just like, I'm not. That's too, too complicated. That has been solved. AI has solved the ability to map that data and ingest it like that. So just that problem alone was solved by, in a matter of weeks, using AI to build the scaffolding and the schemas and the processes to automate that process. That would have been something where just six months ago we would have said, this is the limitation to all business AI. And it solved in a blink of an eye. And so. And it happens almost weekly now that some crazy new thing is solved. That used to take a lot of time. And so it's just getting eaten up very quickly. And so we believe the companies just need to. The companies that get it, they're going to need to shift and shift quickly. So they just need to get their data flowing.
- [57:31]
Host
Okay, well, let's say they want to do that. What did the next couple months look like at Foster? Like, where? I know we've been kind of behind the scenes in, wouldn't say, beta, but we now have customers, we've now had months of learning from them. What do the next 45, 60 days look like?
- [57:51]
Jason
Yep, it's been a ramp up, obviously to scale, and there's been a lot of thought around what is the right path, because everything we just talked about, it's like what you were saying about if a company wants to retain their knowledge or they want to sell it we think the same thing does, is Foster, should Foster be available to everybody or should Foster be available to a select few?
- [58:18]
Interviewer
Right.
- [58:18]
Jason
And I'm not saying that literally, but that concept, right, like should Foster be only for the companies that can really take advantage of it, or should we open it up publicly so that people can just start connecting all their data and interfacing with AI in quickly?
- [58:35]
Interviewer
Right.
- [58:35]
Jason
Should we do that? And there's arguments to both because, you know, there is a path where you could do either. But you want to make sure that you're providing the right value to the market and that we as a company can focus on the things that provide the most value, not only to the market, but to us as a company. And so we've spent a lot of time thinking about that and we spent a lot of time positioning and getting ready to launch. And so where we are is, and we've launched. So we've onboarded almost a dozen companies. We've had several in beta or 5 in beta now for almost four months where we've gone very deep building out entire platforms for them. And now over the past few weeks, we've started to onboard selectively companies that we've chosen to be a part of the first cohort that, that we've spoken with and, and understood the use case of, and believe that there's a potential there to build a really powerful knowledge graph at a pace of around two to three companies a week. That is still by design. So what we plan to do is continue that path over the next 30 days, teeing up for a larger rollout to where we are planning to open it up to the public. On we'll be launching the new website which allows people to go log in, sign up, go ahead and create an account and get a base user access to set up a company. And we plan to do that in the next 30 days. What that'll be though is it'll be a tiered system. So we're going to allow the rest of the people that are on the waiting list of which we have, I think it's close to 300 companies now waiting to get into the system, which is very positive. But we're going to give all of them access to go onboard themselves and get what we were calling like tier one access. So they'll get access to set up their base company, make their base integrations and get going. And then what we will do is we will come behind that initial onboarding wave and help them basically integrate their entire company.
- [60:39]
Interviewer
Right.
- [60:39]
Jason
Which depends on how deep each company wants to go, that next wave will be a little more intentional. And that is where we'll work with the companies that we think have the biggest opportunity and the biggest value for them and for Foster to basically move quickly through this space. And that will be companies like you asked earlier, who's who's most primed or ready to take advantage of it is those companies. It'll be obvious after they've onboarded themselves who's ready. Now that doesn't mean that the companies that onboard to a base level and use it that way won't get a tremendous value. They will. But there will be some companies that are ready to run full speed and those are the companies that will end up being more hands on with once they onboard. So we anticipate that in the next 30 days the website will be launched live, we'll open up self onboarding and we anticipate that in the Next, call it 60 to 75 days, we'll have hundreds of people in there using not just people, companies, thousands of employees in there basically interfacing with Foster inside their organization to get all their results of their data.