Swapnil Bhartiya: Michel, as we all know that airbyte built its reputation around open source data integration and connectors. How does that foundation translate into infrastructure for AI agents to solve the problem that we just discussed?

Michel Tricot: The first thing is really open source is very much is a very strong distribution mechanism. And this is something that has really made us successful in the analytics space and even more when it comes to agency can platform shifts. If you think about it, every single team is tasked for how can I adopt an agency system, how can I build an agent internally? And most of the time that task falls on the engineers working on the project. And open source has always been the best way for enabling developers. They are always going to be looking for what already exists, what is the state of the art. And in general they fall on an open source project, they fall on an open source product. And that is why being open source, our core really helps on adoption and the development of the platform and just enabling all these users that are building this new system to just go as fast as possible and deliver value as fast as possible. I would say the other piece is the problem of data. Connectivity is a completely unbounded problem. The number of places where you might have data available data sitting somewhere is just infinite. And you have what you know today, but you don’t know what’s coming tomorrow. And this is a place where open source actually succeeds. It is meant for extensibility, it is meant for giving access and control over where the product is going to be connecting. And this is something that has always served us and our community really well. And it’s going to be the same when it comes to agency system. Like all these agents need to connect to thousands and thousands of different places. And they can only succeed if they have the ability to to build and connect to all these new systems that are appearing.

Swapnil Bhartiya: Let’s talk about this specific problem that we’re talking about today, which is about, you know, companies have ton of data, they have ton of models, but the connection between and also just because you have connected something doesn’t mean AI. I’m a heavy user, I know it hallucinate, it gets lazy. Also it will just read one line of the prompt and it will skip everything else. Can you talk a bit about, as you have seen, a lot of companies are building their AI agents today. What problem did you see in real world deployments, not hypothetically, where we talk about blocks that you connect mcp, everything will work magically. It doesn’t. What problem did you see in real world deployments that convince you that data infrastructure not models was the missing piece there?

Michel Tricot: I mean a very simple example is just going to any kind of sales team and tell a sales rep to just ask question about a specific customer. Very simple. What the agent is going to tell you is whatever it can find online. Now it’s great, like all these chat system have access to web search, et cetera, et cetera, but all of that is just public data. Like anyone has access to the same amount of information. So you don’t really have an edge compared to another company trying to sell a product. Now if you overlay on top of that all the internal data that you have about a customer of a prospect, it really changes the value because now you can actually inject your own knowledge about a customer. So if you, I mean I can tell you like even in, in my team we have people that go through sales serve, we have people who come from open source, we have people who come just through a talk to sales form. Like all this data is not available to the model. So the model is just going to give you the data that everyone has and it’s not going to make your salesperson more relevant or more lethal than another one at another company. But when you bring all that data together, then suddenly you can say oh, this user has been connecting 50 different systems with open source. Oh, and they’re using all these stacks of application that they are bringing and using airbyte for. Oh, and they actually try to sign up on the salesurf product and suddenly the conversation and the context that you can bring to the conversation really changes how the perception of the person in front of you, it changes the angles that you want to take into the conversation. And this is something that is becoming more and more important. And here we’re just talking about place where a human is involved in the loop. But imagine that you’re building a whole process for managing and this is also something that our customers have. Like we are, we’re working with a fin finance ops startup and well, they’re building agents to enable other companies. But if this agent doesn’t have access to NetSuite, doesn’t have access to Quickbook, doesn’t have access to Stripe, like the agents basically cannot do anything. It’s just going to give you and that’s why it hallucinates. It’s just going to say oh yes, this person had this charge on their account, but if there is no data to back it, it’s just hallucination. But when you start actually overlaying your data, your internal data on top of it, Then suddenly the agent is working out of real data and real information. And that is really like where the, the, the problem is when you want to productionize these agents, an agent that is just searching the web is kind of useful, but it doesn’t really help your business. And that’s what people want from agents. That’s the promise of AI is just, it’s going to give you more, is going to reduce latency, is going to improve your operational throughput, it’s going to make you more relevant in anything that you’re doing. And for that you just need the data that’s been sitting in your system. And we can make a parallel with analytics. Like if you don’t know what your business is doing, you’re blind. And it’s the same thing for agents.

Swapnil Bhartiya: And since you bring in analytics, and if I recall correctly, you folks historically have focused on data movement internization for analytics. What fundamentally changes when the same data is now being used by AI agents for making decisions or taking actions, sometimes in real time, sometimes autonomously.

Michel Tricot: If we do a little bit of history, you go for analytics. People don’t do analytics because they like doing analytics. They do analytics because they want to understand their business. They need to understand where there is gaps, where they need to double down in investments. And in general, the person who is making the decision is an actual person. And when it’s a person, then the rate at which you can make this decision is human time. So the latency is generally going to be pretty high. When it comes to agents, suddenly you have the promise of automation, meaning that you can have a CPU or GPU making the action for you. So operating at sub nanosecond type of latency. And this is what agents are looking to solve and why companies are so bullish in adopting this agency system, is to give them this ability to just streamline, reduce the latency, increase the throughput of any kind of operation that they have internally and, and gain massive speed from that.