Transcript: Alex Merced, Dremio, on Reliable AI — Interview with Alexy

My name is Alex Merced, head of Devril at Dreo, the Agentic Lakehouse. When it comes to my career, one of the most proudest moments or sort of the moments that sort of have been very impactful for me, um, one has been sort of being able to publish a couple really sort of books that people have really seemed to enjoy. One of them being Apache iceberg the definitive guide, Apache Polaris the definitive guide and right now putting the finishing touches on architecting an iceberg lakehouse. Now while these focus more on lakehouse architecture versus AI lake house architecture has become the foundational way you sort of put together the data for AI and this has allowed me to kind of participate a lot in AI conversations and because at the end of the day you do kind of really need to make sure you lay down that data foundation before you take those next steps. That's an interesting question. I mean honestly when we when I hear the word like AI stack oftentimes I feel like what is what stack versus another is sort of very much morphing where basically AI is becoming part of like all types of stacks. Um but in a sense as far as most of the time people's interactions are going to be AI it's going to become more and more abstracted away and that generally happens with all technologies where you know in the early stages you have the really technical users who can really kind of get in touch and work with like the lower levels. So in this case, you know, training foundational models and and building rag pipelines. But in the future, sort of like a lot of interactions people with AI, which you're already starting to see this like that's part of what we're doing over there at Dreo and trying to create a way where, you know, your average business user who's not technical can really feel accelerated by AI and feel like they're leveraging AI and building things um basically through uh things like MCP. Like I think MCP is a great sort of accelerator in this space by creating a standard for me to just be able to plugandplay functionality. So that way I don't have to be so technical that I need to actually build that functionality into my different agentic workflows. I can just plug different MCP servers with different functionality and I get that. But that's going to even become more abstracted and more abstracted over time where um you know basically it's going to be very accessible at the end of the day. You see this with most technologies where over time you get to a point where uh it's successful and then you know people like me from back in the day are become sort of the the AI hipsters who were like well I was into some of those things before it was cool but you know here I