Why AI Models Are Only as Good as Your Data | Dev Ittycheria

With agentic workloads, it's very hard to predict, you know, what's going to happen, how quickly these workloads are going to grow. They don't exactly take weekends off. They don't, you know, take coffee breaks. And then with, you know, these agent swarms, you just don't know how much traffic you're going to see. So you need a system and a platform that can scale rapidly and also scale down rapidly. These models speak JSON. JSON is now the lingua franca of how people work with data, and we were the pioneers of building a JSON database.

It's great to be here. Thank you.

Yeah, obviously, we didn't anticipate this, to be totally clear. But, you know, my predecessor and successor, Dev Ittycheria, decided he wanted to do something else. Got a lucrative offer, and the board asked me, would you come back? And I said, absolutely. You know, I love this company. As you said, I was, you know, have helped build it for 11 years. And I think this, the other thing I would say is the company's on great, you know, great footing. We had a lot of momentum when I left, momentum's only got stronger. It's going after a big market. It's very relevant in the world of AI. And the people here are amazing. So I'm really here to help support them and continue to drive the momentum.

Yeah, it's interesting. You know, even I've been through so many technology changes. You know, I started my first company in 1998, during the boom of the internet. Now, now I'm like this, you know, gray-haired person in tech. And when I talk to these young talented founders, they're dealing with the same issues I dealt with in terms of how can I recruit the right people? How can I create a culture that, you know, really I want and sustain that culture as I add a lot more people very, very quickly? How do I hold people accountable? That's something people really struggle with. How do I manage different kinds of people? A lot of the founders today come from a very technical background. Have they ever hired a strong finance person? Have they ever hired a strong go-to-market person? What do I look for? And yes, you know, they can look at the pedigree of someone who worked at a great company, but that in itself is not always a great signal. Sometimes, for example, a great salesperson at a second-tier company who did really well probably was a great salesperson because they were selling an inferior product, which is so counterintuitive, where like a great engineer at a great product company, you know, that they kind of were schooled and worked with some talented people. So there's a lot of nuances in terms of how you scale a business. And then as a business scales, you find that not everyone scales at the same rate. And so people say, I've got someone really good, but I'm not sure they can get from where we are, you know, by the end of this year to where we want to be by the following year. And what do I do with that? Can I develop them and hope they'll grow into the role, or at some point do I need to bring someone in more senior? And these are classic management challenges that any founder faces.

Technology has been such a boon for productivity. It's been such an equalizer for people from anywhere in the world. Even I think with AI, that to me is the biggest misnomer. AI is such a democratic technology. You can be, you know, a talented person anywhere in the world and use it to create value for yourself and value for the people around you. But I think there's so much fear around what's going on. And I think what's also interesting is I think the fear is driven by the change is so fast. I mean, even the last month we've seen so many new models come out, so many new capabilities. You know, now we have even like personal agents like Instinct and Muse that people are starting to use. And at one level people see the power of the technology, but at another level, like if I don't, you know, if people aren't comfortable with technology, they start feeling like this is a world that's so foreign to me, and their first reaction is to slow down or stop it.

And that's, I think, something we have to address as an industry, is like we really have to make it feel like technology is here for the good and it's a tool. Yes, obviously some people can use it for bad reasons, but generally most people want to use technology to drive good outcomes. And I think that's something that we really have to focus on and emphasize more.

Yeah. So, you know, we definitely believe like the, you know, we're moving very, very quickly into an agentic world where agents, you know, you're going from essentially automating or, you know, accelerating what you were doing individually. Now all of a sudden we're seeing this transformation where now, you know, this technology can actually do the work for you, which is, you know, a very new concept for most people. So it's not like we're building a faster horse now, we actually have a car. And so the way I think we play in this world is that one of the things that people don't fully appreciate is that for agents to reason, agents to, you know, drive good outcomes, they have to have access to fresh data, right? If you're trying to deal with a customer service issue, you need to know what's going on with that customer right now. If you're trying to, you know, buy something, you need to know what the status of the inventory system is right now. And what people have realized is that they need to provide, you know, these systems with the most accurate and fresh data. So having real-time data is one super important. But then even if you have all that data, you need to know what data to find, right? So much like if you told someone, I want you to learn about something, go read the, go to the library and read all the books. That'll take a long time, be very painful. You need to know, you know, what part of the library, what shelf to go after, and what maybe book, and maybe what chapter in that book I should read. So there's techniques around like really precisely finding the right information for that context, for that outcome you're trying to drive. So we play a role because we have very sophisticated embedding models, very sophisticated reranking models, so that when you're searching for something, you get not only just the fresh data but the most relevant data as well. That's super important. The third thing we see is that with agentic workloads, it's very hard to predict, you know, what's going to happen, how quickly these workloads are going to grow.

And so the unpredictable nature of workloads, like in human behavior, you know, like Cyber Monday, okay, we better ramp up. A lot of people are going to be shopping on Cyber Monday, and you can understand the seasonality of your business. With agentic workloads, they don't exactly take weekends off. They don't, you know, take coffee breaks. And then with, you know, these agent swarms, you just don't know how much traffic you're going to see. So you need a system and a platform that can scale rapidly and also scale down rapidly. So in some ways really manage the ephemeral nature of some of these workloads. You need to be able to branch data so you can do different things and more sophisticated things with data. And so we are providing a platform to enable customers to do that. And one of the reasons we feel we're well positioned is, you know, these models speak JSON. JSON is now the lingua franca of how people work with data. And we were the pioneers of building a JSON database, what people now call a document database. And so the fact that we can interact and work with these models without having to go through convoluted, you know, iterations, and we can also deal with very complex things like nested JSON data that a lot of other platforms cannot. We just make it much more easy to work with data, which makes those systems much more effective and scalable.