Interview #96 George Fraser, CEO of Fivetran + dbt Labs

SPEAKER_1 (0:01)

This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required compatibility and availability varies 18 plus.

SPEAKER_2 (0:26)

So good, so good, so good.

SPEAKER_3 (0:28)

Cashmere, outerwear and boots are up to 60% off at Nordstrom Rack. Save on brands like Ugg, Vince and Cole Hawn. Join the Nordy Club for free to unlock exclusive discounts. Great brands, great prices. That's why you rack.

Tony Hoang (0:50)

Welcome to The Artificial Intelligence Podcast, ranked in the top 2% of over 7 million podcasts on Spotify.

I'm your host, Dr. Tony Huang. Today we're with a special guest, George. Can you give us a brief description of your background?

George Fraser (1:04)

Sure. I am co-founder and CEO of Fivetran and dbt Labs, that recently merged. And before that, I was a scientist.

Before that, that was it. I was in college.

Tony Hoang (1:20)

Cool.

George Fraser (1:20)

I've been mostly doing this my whole career.

Tony Hoang (1:23)

Oh, wow.

So the Labs building the most capable, like AI models right now are also customers of like data plumbing companies like yours. What do they do with their own company data that a typical business doesn't?

George Fraser (1:40)

Well, both OpenAI and Anthropic are a long time Fivetran customers. They set up Fivetran when they set up their very first data stack. And we've been happy to be on that journey.

We've seen some of the crazy scaling through that lens. And in terms of what they do differently than most companies, they have really fully connected their AI agents to their company's data stack. They've both written about it on their blogs, about their data aid, their internal data agents that people who work there can use to answer any question about their company's data. And underneath that is the same kind of data stack that most companies use to run business intelligence.

Tony Hoang (2:29)

And so if you take like an agent that has been given like a real job inside of a company that can't see that company's data, can you like walk me through what goes wrong on its first day?

George Fraser (2:42)

Well, most tasks, if you just go in and you look at a bunch of, you know, GitHub issues, pull requests by the engineering team or the analytics team or the Salesforce administration team, or you look at other forms of tickets that represent a unit of work, what you will discover is that even if they were done by humans, the humans had to gather a lot of information along the journey to solving that problem. Most problems cannot be solved in a closed world like mathematical theorems. And so they need to find out details like, you know, what was the error message that happened when this customer's pipeline stopped working? Or what is the recent history of performance for this customer who is complaining in a support ticket? And in order to answer those questions, they have to get the data from somewhere. And what we in those lab companies you mentioned, and many more companies every month now are discovering, is that the best way to surface that information to AI agents is not to connect them directly to all of the systems of record.

Because it is simply too slow and too limiting to go through the APIs and MCP servers of every individual system of record separately. Instead, what you want to do is get your data all in one place, construct a simplified model of that data in a database as a set of tables, and connect your agent to that. And it turns out that the ideal way to do that, the ideal form of a central datastore for agents, is very much like the ideal form of a central datastore for human beings, who have been using these things forever for doing business intelligence and reporting.

Database is still great, great for agents in many of the same ways as they're great for people.

Tony Hoang (5:05)

Now, would like a smarter model fix all these issues, or like, is it only like better data, in your opinion?

George Fraser (5:12)

Well, smarter models are great, but as I said, outside the world of pure mathematics, there are a few problems that can be solved in a closed world. So you have to have data to solve the problem. It's not a question of how smart the model is. It's simply no model nor human being could solve a typical workplace problem without access to data about what's happening out in the world and inside the company.