Eugen Alpeza, Edra CEO on AI Agents, Enterprise Workflows and Automation

Grace Gong (0:01)

Oh, we're live. Okay, one sec. Oh my god. Of course, there's always one platform that's going to give crazy reaction. One sec, let me just remove this and then add this again.

Okay, done. Hi, Eugen. Welcome to Venture with Grace.

Eugen Alpeza (0:20)

Thanks for having me.

Grace Gong (0:22)

I'm so excited about our podcast because I heard you guys created FDE.

Eugen Alpeza (0:27)

I mean, Palantir has had FDE for a very long time. Me and Yana specifically created the Forward Deployed AI Engineer role and that unled that team at Palantir.

Grace Gong (0:38)

I love that.

Okay, so before we start off our show, I want to give a quick shout out to our amazing sponsor. This episode is brought to you by Nebius, the ultimate cloud for AI innovators. Nebius provides AI infrastructure you can count on combining reliability and speed with flexibility and engineering support unmatched by hyperskillers. AI leaders like Meta, Shopify and Higgsfield already partnered with Nebius to run their AI workflows. Plus, venture-backed startup can save up to $150K on compute costs when they apply for access. Visit nebius.com or nebius.com/startup to learn more. Okay, so I want to start with your journey. So obviously, you guys have quite amazing career at Palantir. Before that, you were a data scientist at Upwork. And among others, maybe we'll go start with what were some core lessons that you've learned early out of your career kind of shaped you into who you are today?

Eugen Alpeza (1:26)

Yeah, exactly. So I mean, I actually did all my education with my co-founder, all of it focused on machine learning, right? And then I joined Palantir right out of school in 2017 I was truly forward deployed, meaning I got hired in London, I got shipped to the Middle East, spent a year there, spent a year in Brazil, spent a couple of years in Texas. So I was really going to the customer site. Most of my career was spent in the customer office in a city that Palantir might not even have had an office at that time.

And I think there's a lot of lessons now that are kind of, everyone is very fond of the DfD model, everyone is sort of trying to learn and copy from it as much as possible, but just being really, really close to the customer and caring about the outcome and the last mile is something that gets very deeply ingrained in you. And I think now with AI, which the agents have the ability to take on the last mile of the work in every single organization, that is why everyone is now trying to do DfD. Like it's no longer, the standard is no longer, hey, I'm giving you this fairly generic piece of software, use it, good luck, and I don't care if you get value out of it, but it's really about, okay, I'm here to make sure that you get to your outcome, you get to the value that the agents do what they need to be doing.

Grace Gong (2:46)

I love that. I feel like, okay, so I want to say, maybe let's start with what Edra does. What you guys do is automate work by teaching agents how their business actually operates. I feel like context layer and memory layers, right now it's the most important keywords of the year basically.

Maybe we can start from how does Edra work, and then we can talk about all the business use cases and everything.

Eugen Alpeza (3:16)

Yeah, perfect.

The Forward Deployed model is actually very closely linked to the context layer and the memory layer. A lot of the FD job is coming into an organization, and normally the way it would work, a customer would say, hey, can you help us automate this process with AI? We say, yes, of course, no problem. What would you like it to be? What is the procedure? Well, we don't really know. There's Alice, she's been here for 25 years. She deals with these types of exceptions. There's somebody else. We have 4,000 documents on SharePoint. We don't really know.

A lot of the FD job today is translating, capturing the tribal knowledge, speaking to the people, analyzing previous data to try to get to a clean set of instructions for the agent to go and execute. People have tried doing this in different ways. Like you're thinking about, there's context graphs. People are thinking about memory and all kinds of different things. But a very simple way to think about it is, the agents are super capable and they can do effectively anything you tell them to do, but you need to specify it very specifically to your company.