AI Infrastructure & the Evolution of RAG | Weaviate's Bob van Luijt

Show Notes

"This is the time. This is the time to start building... I can't say that often enough. This is the time." - Bob van Luijt

 Join Bob van Luijt, CEO and co-founder of Weaviate as he sits down with our host Conor Bronson for the Season 2 premiere of Chain of Thought. Together, they explore the ever-evolving world of AI infrastructure and the evolution of Retrieval-Augmented Generation (RAG) architecture.

Bob's journey with Weaviate offers a compelling example of how to adapt to rapid changes in the AI landscape. He discusses the importance of understanding developer needs and building AI-native solutions, emphasizing the potential of generative feedback loops and agent architectures to revolutionize data management.

Chapters:00:00 Welcome to Season 2

1:43 The Evolution of AI Infrastructure

04:13 Navigating Rapid Changes in AI

07:39 Generative Feedback Loops and AI Native Databases

13:26 Challenges and Opportunities in AI Production

19:03 The Importance of Documentation and Developer Experience

27:13 Future Predictions and Paradigm Shifts in AI

31:17 Final Thoughts and Encouragement to Build

Follow:

Conor Bronsdon: ⁠https://www.linkedin.com/in/conorbronsdon/⁠

Bob van Luijt: ⁠https://www.linkedin.com/in/bobvanluijt/

Weaviate: https://www.linkedin.com/company/weaviate-io/

Show notes:Learn more about Weaviate: https://weaviate.io/

What is Chain of Thought | AI Agents, Infrastructure & Engineering?

AI is reshaping infrastructure, strategy, and entire industries. Host Conor Bronsdon talks to the engineers, founders, and researchers building breakthrough AI systems about what it actually takes to ship AI in production, where the opportunities lie, and how leaders should think about the strategic bets ahead.

Chain of Thought translates technical depth into actionable insights for builders and decision-makers. New episodes bi-weekly.

Conor Bronsdon is an angel investor in AI and dev tools, Head of Technical Ecosystem at Modular, and previously led growth at AI startups Galileo and LinearB.

[00:00:00]
Welcome to Season Two of Chain of Thought
Bon Van Luijt: If you're a founder listening to this, this is the time to, verticalize around these kind of solutions. So what I just described, If you do that around, finance, insurance, and God knows what, this is the time to build it. And it's not a silver bullet, but it's also not the time to be too skeptical about it, because people are building, people are successful.
No. No, I'm talking actual dollars coming in right? it's it's unbelievable. It's going very fast So this is the time to do it. So silver bullet no, but optimistic about the use case because a lot of serendipity will, will happen. you will add a lot of value to companies buying from you or companies where you're bringing it, and maybe not a hundred percent, but if you solve 80% of this problem, you're gonna make a lot of people very happy.
Conor Bronsdon: Welcome back to season two of Chain of [00:01:00] Thought. I'm your host, Connor Bronson, and today we're diving deep into the evolution of rag architecture and AI infrastructure with Bob van Luijt, CEO, and co-founder of Weaviate
Bob, welcome back to the show.
Bon Van Luijt: Well, thanks for having me. It's great to be here.
Conor Bronsdon: Yeah, it's always fantastic having these conversations with you because not only are you someone who has a deep background on software engineering, has now spent years in the industry, you obviously shared some amazing thoughts with us, in season one of Chain of Thought around AI agents back in October, for those who haven't listened, check out episode two of Chain of Thought for that conversation.
But even since then, it feels like there's been so much of this change in just a couple of months. And I won't posture.
The Evolution of AI Infrastructure
Conor Bronsdon: I think there are many of us, myself included, that feel kind of constantly behind in our learning with the rapid pace of capability change that we're seeing within AI yet, since co founding Weaviate almost, I guess, a little over five years ago now, focusing initially on [00:02:00] embeddings, you and Weaviate both have managed to constantly adapt and grow.
And today Weaviate is an entire AI native database for the new generation of AI applications. Empowering developers around the world to build AI applications. How have you successfully navigated this rapid pace of change within AI?
Bon Van Luijt: It's a, it's funny when you say five years, it's like, that's true. It's five years already. I think what the most important thing that we've learned when, when you look at a new paradigm, like in this case, AI, is you need to answer the question, how do the developers work with the technology? Right. So let me give you a very simple example, right? So, for example, in the, during the web, the, when the web came up and a piece of infrastructure that were developed around that, then. We saw a lot of JavaScript. Now, you might say, well, that's obvious, right? Well, in hindsight, that's obvious.
Because in the early days, on the back end side, there was a lot of Ruby, and there was a lot of JavaScript. So, what we see today is that [00:03:00] there's a lot of Python for developers. So then the question is, how does the, technology evolve around that? How do people adopt it? How do they use it?
How do they build? Those kind of questions. And that is just one. Of many. so what you constantly try to do is like, how do people build, what is the language that they use or in a more fancy way, It's like, what's the nomenclature that they adopt, right? So the fact that what we today call RAG, we had that already in Weaviate in a way that we call it generative search. And then we saw that people adopted RAG, because it was in one of the research papers. so then, you know, you just adopt to that language. So in the product itself, nothing changed. It was just how we talked about it and how we positioned it.
so to answer your question, it's mostly about like constantly having the dialogue with users, developers, customers, and so on and so forth. It's like, how do they use it? And how do you lean into that? And, that's [00:04:00] basically my answer.
Conor Bronsdon: Has your perspective on how to approach these problems shifted as you've grown in and kind of learned and done everything with Weaviate over these last five years?
Navigating Rapid Changes in AI
Bon Van Luijt: So the first thing that I still cannot wrap my head around is how big it is. AI. It's like, to give you a simple example is I was at AWS reinvent, the fact that you can walk down, a, trade show with so many companies somehow talking about AI or doing something with that, that I can walk down the trade floor and see other companies flashing our logo as they integrate. That is something that is unbelievable how big it is. So I'm still trying to get used to that. The thing that I, that I also learned that is because it's so big and so massive you really can figure out who your, USP is, right?
So who do you want to help and who do you want to serve? And In the early days, [00:05:00] like in the really early, early days, everybody who was interested in machine learning, we were like, Hey, you know, we can help you that has changed, right? So it's more, targeted to what we like to call the AI native builder.
What does that mean? That means somebody who wants to build an AI app rather than just adding some AI features to it because you believe that AI is going to be everywhere. And that's very different. So I, you know, when I grew also in my role as a CEO, I sort of become more comfortable, guiding the team in that direction. And that comes to the benefit of the customers and the users that we have.
So that is the biggest thing that has changed, I think.
Conor Bronsdon: And obviously, we're going to see more of these waves of change. AI, within the last five years, to your point, has seen major waves. Whether it's, moving from ML to Gen AI is the kind of key thing we're talking about. To the rise of Reg, and now Agents. And you've been along for that ride. Starting with entity recognition and graphics, which evolved into what we now call Graph [00:06:00] Reg.
What do you see as kind of the next stage for AI?
Bon Van Luijt: one thing that we started to see when RAC emerged, and that people started to adopt it, not as an intellectual concept, but the people we started to build with it was that it was, it was one directional. So it is like you have a query, you retrieve some potential, documents, you send them to the generative model and you present it to the end user.
That's one directional. Out of that, what we started to see were these agents, right? So agents started to emerge and one way to think about agents is that they kind of create a feedback loop. So now the model itself says like, Hey, I need more information from Weaviate or I'm going to put something back into Weaviate and where we are right now. in these agentic architectures, or agentic reg, or how you want to call it, is that, the agent can request more information. But the thing that I'm personally very excited about is what we call these feedback loops, where the agents can also put stuff [00:07:00] back inside the database. So the biggest challenge that exists today in the world of data management is master data management, right?
If you put in the wrong information in the database, it's very hard to get the right insights out. And we believe that these, we call them generative feedback loops, as part of these agentic architectures, can help to solve that problem. So we're really looking at this complete new paradigm, how we're dealing with data, combining the vector with the models, and that's actually where the name Weaviate comes from.
It's like weaving the models and the database together. So we're still doubling down on that.
Generative Feedback Loops and AI Native Databases
Conor Bronsdon: Can you talk a bit more about these generative feedback loops and how you see them really being impactful over the next couple of years as people continue to build more AI apps and integrate AI into applications that already exist in today?
Bon Van Luijt: Yeah, that's easiest explained based on an example. So, many years ago, way [00:08:00] before I started, Weaviate, I was working asfreelance software consultant, and, I was hired by a company. it was a candy manufacturer and they had factories all over the world, right?
So, and they made candy and they were selling that all over the world, big, big candy manufacturer. And one of the challenges that they had was in their master data management. They stored, data that was coming from all these different factories. And yes, they had all their standards and those kinds of things, but guess what? You know, not everything was aligned with the standards. So, I don't know, could be. You know, the way, uh, temperatures where, you know, Fahrenheit is the default, started in Fahrenheit, and where Celsius is the default in Celsius. The British wrote certain words different than American English, well, and so on and so forth, right?
So a lot of stuff. the solution to that problem was people. So we just have people to go through our data, master data management and just try to solve it, right? Just change it and update it and improve it, [00:09:00] which is undoable because data is coming in faster than these people could handle it.
And so that's stuck in the back of my mind. And when these, fast forward a couple of years, when these REC, architectures started to emerge and out generative feedback loops, they don't say, what if we ask the agent to solve this problem? So we tell the agent, this is our master data management plan. And, now you solve it if there's something wrong in the database, which is fascinating because the models are now at a quality that if you have 10 data objects is if you talk, for example, about cell temperatures starting Celsius and one in Fahrenheit. It actually tells you it's probably wrong. It probably should be Celsius and, and stored in Fahrenheit. when it comes to languages, translations, and those kind of things. So, that is the, where that idea comes from and how I got excited about it. It's a completely different way of dealing with, master data management. And, native database where, you know, the models and the database are brought together. And the way that the [00:10:00] developers interact with it is completely different. optimally suited for that, right? So that's a, that is why I got so excited about it. It's, it's RAG, just the first step, and then agents the second, and then these feedback loops the third, whereas just the, you're not prompting the model, you're basically prompting the database.
It's like, this is what I want from my data, and it figures out itself which models it needs, and how to update these kind of things, what to search for. So that's that's why I'm so excited about it.
Conor Bronsdon: I really like this example you're using because it highlights something which we maybe don't always think about, which is it can feel like we have these thematic waves that are coming, you know, where we've talked about reg and agents and feedback loops, gen AI more broadly, ML before that, there are more micro waves we could talk through.
Often, if you're kind of taking an outsider perspective, or if you're new to the AI space, it can be hard to understand, okay, why is this new hotness now that the topic of conversation and the real answer here is that it's it's [00:11:00] solving the next problem as each stage of AI thought or problem solving gets taken on, we go, Oh, we have this new problem to solve.
We've kind of taken this new layer, and it feels like that's the mental model. That you're approaching both Weaviate and the industry with is like, okay, we've solved one layer. What's the next layer down? We can solve.
Bon Van Luijt: Yeah, exactly. there is this knee jerk reaction, especially when the company grows, to become more conservative, right? So it's okay, now we have this AI thing and now we look back and we're gonna try to be part of something traditional because, you know, but then with AI or something. Whereas I keep telling the team, no, no, no, no, no, this is just the first step, right? Where we are right now. And now we need to hit the pedal to the metal to really just get into this new paradigm. And the paradigm is these are these agent architectures where just, if you think about it, it's very logical, right?
It's like you say, okay, we now have a model that has an opinion on our [00:12:00] data. How does it get that opinion? Well, we prompt it and it interacts at the data that we store in the database. But we should not forget that a lot of people, are very jaded, right? Because it is like the garbage in, garbage out thing is, is like. So old and oh boy, now we have another solution for it. but people see, they feel it, right? And that's why that ChatGPT moment was so important. Because that was like an eye opener. They were like, Oh, so this is the stuff you're doing. And now integrating on the infra level, it's The database with the data and the models to create these feedback loops is just the right time for it. People, they, they, they see it, they understand it. So it's the right time to really go into this new paradigm. So that's also why I'm saying like, it's not a, you could look at like vector database and that kind of stuff as part of the NoSQL paradigm.
You could do that, right? But the exciting thing is the new thing that AI native paradigm, that's the new thing that comes after that, and that is people are ready to build [00:13:00] with it. And, that's why I'm still so excited about, working on this, uh, on this product in this category.
Conor Bronsdon: I appreciate how iterative your mindset to solving these problems is and broadly, I think it aligns so closely with how the industry is thinking about. Each layer and step of AI is like, we have to keep iterating on this because, Each problem we solve opens up these new possibilities, which I think is exactly how you're describing it here.
Challenges and Opportunities in AI Production
Conor Bronsdon: And a couple of weeks ago, You had a linkedin post that went very viral, highlighting seven different popular RAG architectures. Why do you think that rapid architectural variation struck such a chord with all these AI builders?
Bon Van Luijt: Yeah, so the question we get the most is, I can see it, I get it. Can you help me do it? How do we do this? And these kind of posts are an invitation to developers like, hey, we can help you. Look, these are the seven methods [00:14:00] and here is like follow up documentation, blog posts videos, god knows what to help you do this. I find it very important that we keep this this in open hand to help the folks like we're here to help you build these kinds of applications. And I believe that that's the reason why these posts do so well, because people need help. And this shows like, hey, this is a way how to do it.
These are the seven ways how to do it. Which one fits your use case, right? And because we should not forget that, Yes, there's a lot of innovation has been happening in the past years. We have a lot of early adopters, but it's now the time for the majority, right? So the early majority is starting to adopt. Just, you know, random developer at, Enterprise X gets an assignment. I think, hey, wait a second, I can solve this with RAG.. And we're here to help these people to build these kinds of applications. And that's the reason why I believe these posts do so [00:15:00] well. Because it's just an open invitation to start building.
Conor Bronsdon: And it feels like we're also seeing this transition from experimentation to actual productionization of applications. How are you seeing these different architectures actually being implemented as you help AI builders to execute on their ideas?
Bon Van Luijt: Yeah. So for us as a company, the last year, 2024, What's the year of going to production? How do we notice that? Just, you know, based on the skill we see at customers and the questions that they ask. So one of the things that we've learned last year because of people going to production is that if your use case stays relatively small, you can do anything you want. With the database, right? You don't really notice it, in your cloud bill or that kind of stuff, these are relatively small use cases. But with the use cases growing, there's this whole operational effort around the database that starts to, [00:16:00] emerge.
So one thing that we've learned is that a lot of the indexes, right? So the indices that we have in the database, they sit in memory. Right. And that sometimes you need that for e commerce use case and that kind of stuff. But, we also got more and more customers who said like, well, for our use case, we have so many data objects.
If we store that in memory, it's just, we can't make a business case. It's like we can't, we can do the POC and prove that it works. But now if you want to scale it to production, we can't do it anymore. So last year we introduced the storage tiers. So now you can tell Weaviate keep it in memory. Or get a little bit of a penalty in speed, but store it in disk or even in S3. And that's a fraction of the price. And the penalty is lower speed. So what you can do in a handful of milliseconds, now you, it's a little bit, slower. It's still in the milliseconds, but it's significantly cheaper because we saw a lot of these rec use cases have a lot of tenants, right?
Think about like document search, email clients, those kind of customers that we [00:17:00] have that needed that. So that is a sign of production, right? That people ask these kind of questions. So are a lot of our, innovation last year Was, based on our bigger customers and asks they had of things that they had in, that they were noticing production.
To give you one more example, we've implemented something called acorn that comes right off the scientific press, if you will. And that is a way to deal with filtering in vector search. And now you might say, well, okay, yeah, whatever.
Right. But the thing is we have customers with billions of e commerce products and all of a sudden, if they had low results. The queries became significantly slower because we didn't optimize that filtered search. If you have a million products, you're not going to notice that. But if you go up to a billion, you start to notice it.
So, What changed from the year before that to last year was that a lot of innovation again was customer driven, [00:18:00] where people say, Hey, we're going to go to production. It's too slow. Improve this, right? They didn't notice that with a million, but now with a billion, they notice it. So those kind of things we started to see, and that was proof for us like. People are ready to production. They're scaling up, they're bringing it to their own customers. They start to talk about business cases, right? So like, you know, what's the dollar, how many dollars do we need to invest to make one of our customers, that kind of stuff. And in 2023, that was not the case.
Conor Bronsdon: Absolutely. And I'll say on our end at Galileo, we're certainly seeing that on the evaluation side too, where
if you're just in experimentation, you're not really feeling the pain. And then as you move to production, you realize, Oh, like I should have had this all along the way so that I didn't magnify my problems.
That's so that I could build trust my customers when I actually go to production. And I mean, obviously we're seeing a huge shift in the market as people are saying, evals need to be a base for Part of how we build AI applications. and it seems like there's a lot of these AI infrastructure decisions that are being made now where we're realizing as we go to production, Oh, we actually have to go back and adjust how [00:19:00] we're building this in the first place.
Bon Van Luijt: Yeah, exactly.
The Importance of Documentation and Developer Experience
Bon Van Luijt: And that goes back to your first question, because one thing that I find fascinating is that. The point of entry for developers, of course, you guys know this as well, is through documentation, right? So that's the point of entry. And if you make something very lean and very simple, right, you have like very minimalistic documentation, and it's very easy for people. to get started. Then, when they start it, and then you get this move to production, it's like, Oh, wow, this is great, but we need this feature and we need that feature. Great, sure, we can do that. So you have these features. And now, your documentation grows a little bit, right? So, because now you need to also explain these features. So, by the time you get to production, like a production quality database in our case, Your documentation has grown with
that and we have a whole team that is just concerned with managing that, right? So making sure that you keep all the goodness of this [00:20:00] easy to get started in five minutes thing. But that you're also showing people like, Hey, and by the way, when you go to production, you get all these cool features like multi tenancy features, data offloading features and that kind of stuff.
And balancing that is part of what we've learned last year. that's hard. you constantly have to catch 22, like, are we going to show them all the stuff we can do at the risk that it might come across a lot of information, but that they. Get the trust as a developer like oh, we can do that or are we gonna do the other the other way around and then later Show them.
Oh, by the way, we can also do all that other stuff and That's a fascinating Catch 22 because I believe so often developer experience is people put that equate that with Lean documentation, like just one, two pages of documentation. And I'm not so sure about that. Because if I use a product and I see that there's [00:21:00] detailed documentation that I know when I scale this up, they can handle this. That gives me a good experience. And I'm willing to just learn a little bit more from before I run it. So, those kind of things are also things we've learned this year. And I find it a fascinating experience. process because as the database evolves, documentation evolves with that, clients evolves and evolve and that kind of stuff.
Conor Bronsdon: I agree with you. I think it's crucial to have good summarization because I don't want to jump into a product and feel confused as I start out, Whereas if it's like, Hey, here's a clear route in. Let me get to that.
Wow. Moment rapidly. And then you give me depth to explore. I'm much more likely to be successful with the product.
Bon Van Luijt: Yeah, but the opposite is also true. So if you have that wow moment, but there's nowhere to show them, Oh, and by the way, we got your back when you scale up. That also has a negative DX effect because people are like, Okay, this is a great thing to get started with or to play around with, but I'm not going to use the technology to go to production.
And I [00:22:00] think that has not changed. With AI, is that the way we make money as a business is still over volume,
right? We Production people, so if you put, if you add something to a database people pay you to make sure that a day later It's still there Right? And, and that they can retrieve it at a convenient speed for their use case that's what they pay you for.
So you need all those tools and functionalities to do that. So it, that's a fascinating, a catch 22.
Conor: Unlock evaluation intelligence for your AI team with Galileo, the industry leading platform for evaluating Gen AI applications.
Trusted by AI leaders like HP, Databricks, Comcast, Twilio, Headspace, CollegeVine, and many more, and powered by our proprietary Luna Evaluation Suite, Galileo enables evaluation driven development and observability for your AI systems. Learn more at Galileo. ai
Conor Bronsdon: As [00:23:00] this shift from. experimentation 2023 to production use cases 2024 and increasingly even more productionization 2025 has occurred. Are there particular themes that you're seeing around which use cases are going to production first among this early adopter set?
Bon Van Luijt: the first use cases that we saw, go to production was just more large scale, just search, right? So, and then mostly hybrid search. But the thing that we saw immediately after that, that I was very excited about, was what we like to refer to as these multi tenancy use cases. So for example, think about an email client where people say like, well, we have, I don't know, a billion emails, but distributed over a hundred thousand customers. We don't want them to sit in the same vector index. They need to all be separate, right? Multiple tenants and in an ideal world we can also offload them if somebody's not using it So that it's cheaper for us to [00:24:00] operate this large scale, operation and that's something I became very very excited about because That was also part Of new stuff that people are building with the models, right?
Thanks to the models, right? And better search and better recommendation. That's great. That's nice. And we have some amazing customers there and we're happy that we can serve them. But I'm also very excited about that new stuff. And turns out that people who build RAG applications for their customers, because in the end we're like a B2B solution, Was that they said, we, that, that my customers went like, we have a lot of customers and we want to efficiently store all the data and turns out that for their individual customers, they might not have a billion data objects, so they have a billion in total, but it's distributed differently and we want to manage that efficiently. So that's, it goes back to the storage tiers. That's why we've introduced that. And that's now how the database is evolving in again, a really truly unique. piece of technology that's not [00:25:00] a replication of old stuff with something new on it, but something really new. And that's also why we start to refer more and more to it as AI native database, because we really want to help these AI builders and not just do some search add on or something.
Conor Bronsdon: As we see this new group of AI builders coming in, these early majority folks who have seen the experimentation, seen these first use cases and say, Oh, Hey, I need to take this on. I need to learn this. Are there new challenges that are emerging around how we get them into widespread production adoption?
Bon Van Luijt: Well, there, so what, something that starts to emerge is that the way that people build. so let me give you a very simple example here. So a simple example is chunking, right? So somebody has a use case where they say, Hey, want people to search through PDFs? Okay, great. How do you do that? How do you chunk a PDF in such a way that if you retrieve it from the database, Pipe it through the [00:26:00] model, for example, direct to show the right, the relevant answers. That's actually turns out that's hard to do. Let's say that I have a public document about, Apple's revenue, right? And I say like, what were, was the revenue in Q2, 2024? And I say like, what was the revenue in Q2, 2024? Then if I look at that single PDF, the only way that I might able to recognize that it's Apple is that it has the little Apple logo in the top. Then it only talks about revenue numbers, but in nowhere in the text says, Oh, this is Apple's revenue in Q2 of 2023. So what we need to do is we deal with that. we need to represent the pages with multiple embeddings, right? and we need to be able to store all of them. We need to be able to retrieve them in a way that we, see fit for that specific use case.
Becomes pretty elaborate pretty quickly. So that's a more complex chunking strategy. So, again, that goes back to that viral post that we talked about. We start to post more and more of those kind of things, again to help these developers to build these kind of things [00:27:00] and now knowledge is starting to, develop inside these developers who become knowledgeable about building these AI native, cases.
So that's just one pragmatic example how we see that in practice.
Future Predictions and Paradigm Shifts in AI
Conor Bronsdon: Are there particular opportunities that you're seeing around this next stage of AI infrastructure, or predictions that you have for what you expect to see around AI infrastructure in 2025 to help solve some of these novel challenges?
Bon Van Luijt: One way to think about it is that you can, everything that people are building, you can chuck it up into, you know, put it in two buckets. So one bucket is just doing stuff better, existing stuff better. And it often evolves around search and recommendation, right? A lot of value created.
It's great, right? So it's like, okay, you use vector embeddings to eat more, you know, better retrieve information and data, that kind of stuff. But we have a lot of customers there. the thing that I'm very interested in is the other bucket. And the other bucket is the new stuff, right? The AI native stuff, if you will. And that [00:28:00] example of the agentic architectures. For example, through the generative feedback loops, where we basically have our master data management, documents to a model. It's like, hey, you're hooked up to this database. Every data object that gets added to this database, check it, validate it, throw a warning, throw an error, fix it for me, whatever you want to do.
So rather than having humans in the loop to fix that kind of stuff, we now have these models to do our master data management. And I, I, again, I would like to highlight the master data management The market is huge there, right? It's like if the search market is tiny compared to that. And that's what I believe that these AI native databases like Greenfield enable, right?
So yes, a lot of customers, a lot of users in the first bucket, which makes sense because that's where the market is. But I believe that we need to lean forward and show the market that this is all great, this stuff here in this bucket, but look what we can do here. and I hope that this is the year where these agentic architectures really take a foothold and that we [00:29:00] see the production case for those now as well.
Conor Bronsdon: Are there any particular gaps that you see where you're like, Hey, I want, I want to hear that folks are working on this problem, or this is something that needs to get solved to unlock this next layer.
Bon Van Luijt: So. Funnily enough, I wouldn't give a technological or like a tech answer to that question. It's more a almost a psychological answer because it's paradigm shift. Since we've been storing data in what I believe IBM had the first commercial database way before Oracle, right? Since we've been storing in the data in these databases until today, we have been dealing with these challenges of like garbage in garbage out. Proposing something thanks to the models to fix that or potentially fix that is a huge paradigm shift. And having people get a heads up what's possible and having the model solve that rather than people is a [00:30:00] huge paradigm shift. So it's more getting people to accept it and to adopt it.
That's a bigger thing than the actual. Technological challenges, right? So because we can do it, we know we can do it. Can we, is it a silver bullet? No, of course not. But it brings us. Damn far, right? You know, it's we can do a lot with it, but then it's often people in these businesses to accept it. This is how we're going to do it. And that's new. So I think this year is also a lot off more, you know, evangelizing, showing people the value. We have the first customers who are daring to take that step, right? So if we look at our full transparency, 95 percent of our customers sit in the first bucket, But now that's, that's growing rapidly.
Now we're not where we wanted to be yet. So, that's hopefully for this year that we're going to grow. That it's truly becoming AI native. That really people use the models in combination with the database to build new applications or [00:31:00] enhance mass data management challenges that they might have. In a completely new way. And that's a paradigm shift. And that's not a technological challenge anymore, but it's now psychological challenge that people are willing to take that bet.
Conor Bronsdon: To your point about the psychological challenge of this paradigm shift.
Final Thoughts and Encouragement to Build
Conor Bronsdon: We've talked a lot in this conversation about iteration and the need to just keep building and keep layering these improvements on, but I think a lot of the expectation from outside of the industry or for folks just getting in is, Hey, this is a silver bullet and this is just going to solve my problem.
But the reality is there's a lot of work to be done to set you up for success. It takes work to get from, Hey, this is an experiment to production. We're not at a point where you can simply say, Hey, here's the agent, turn it loose, it's done. There's work to do to make sure the right guardrails are in place, that you are ensuring that data is accurate.
you have to avoid that garbage and garbage out problem. As you mentioned. how do you think AI builders should be. thinking through this next stage [00:32:00] of this rapid evolution. What's the mindset they should bring to avoid over focusing on silver bullet problems and instead actually find problems that they can solve?
Bon Van Luijt: I think because of course I don't know the answer because if I would know the answer, then, we wouldn't be having this conversation. Right. So, but what I do think is that this is the time. if you're a developer listening to this Podcast and you work at a large bank or something This is the time to take that AI approach, right? Not, next year, but now you can do it, right? So now you can start building, build that expertise and bring that to your company. If you're a founder listening to this, this is the time to, verticalize around these kind of solutions. So what I just described, If you do that around, finance, insurance, and God knows what, this is the time to build it. And indeed, it's not a silver bullet, [00:33:00] but it's also not the time to be too skeptical about it, because people are building, people are successful. I'm an angel investor in like two dozen companies or something. And what I see people, and most, most of uh, uh, uh, Weaviate users. So. How I see them be successful and I'm not talking about successful in Open source downloads.
No. No, I'm talking actual dollars coming in right? it's it's unbelievable. It's going very fast So this is the time to do it. So silver bullet no, but optimistic about the use case because a lot of serendipity will, will, will happen. you will add a lot of value to companies buying from you or companies where you're bringing it, and maybe not a hundred percent, but if you solve 80% of this problem, you're gonna make a lot of people very happy.
Conor Bronsdon: Yeah, absolutely. And you learn a lot more by jumping in now versus
waiting. The sooner you start
Bon Van Luijt: this is the time. This is the time to start building. It's a, I cannot, I, I [00:34:00] can't say that often enough. This is the time. It's like you don't have to wait anymore. The tooling is there. well, you guys know this of course. It's like, this is a time to build. so that's why I'm excited about it.
Conor Bronsdon: I love it.
Conclusion
Conor Bronsdon: Bob, thanks for sitting down for this fascinating conversation and for sharing your perspective. Where can folks learn more about Weaviate and keep up with your team's work?
Bon Van Luijt: the, just the website. weaviate.io. We do a lot of stuff on socials and then especially LinkedIn. So if you follow our company or reach out to me on LinkedIn You post us like almost daily about this kind of stuff So if you follow us on our journey, you can sign up to our newsletter Slack channel that kind of stuff Database open source so you can start playing around with it. So, just, you know, you can start if you are traditionally through a Google search or more modern to a, uh, search through a model, uh, about Weaviate but, we're everywhere.
Conor Bronsdon: Fantastic. Bob,
thanks so much for coming back.
Bon Van Luijt: you. Bye.
Conor Bronsdon: Awesome. And that is all for [00:35:00] this episode of Chain of Thought. Thanks for listening, everyone. And don't forget to like, subscribe, and leave us a review wherever you get your podcasts. And as Bob said, check us out on LinkedIn. Check out Weaviate on LinkedIn.
We'd love to hear from you.