How Google Is Building Enterprise Trust in Agentic AI

**Andi Gutmans** (0:00)
It's going to take time to earn the trust, to understand what quality outcomes we can drive. There are going to be certain things that we feel very comfortable making autonomous today. As an example, if you look at customer support today, there's a huge amount of that that's really autonomous today.
But as you mentioned, inventory replenishment for $10 million, probably companies still want a human in the loop.

**Cindi Howson** (0:34)
Welcome to The Data & AI Chief. I'm your host, Cindi Howson. Agentic AI sounds miraculous, doesn't it? Schedule a meeting for me, replenish inventory, grant credit limits.
Maybe this sounds a little scary too.
Is our data foundation even ready for this? My guest today is Andi Gutmans, Vice President and General Manager of Data Cloud at Google, where he leads a unified platform of databases and lake house technology built to put enterprise data to work through AI agents. In this episode, you'll learn how much autonomy AI agents are really ready for, what's driving the rising token costs, and how to modernize despite the technical debt holding many enterprise organizations back. Andi, welcome to The Data & AI Chief.

**Andi Gutmans** (1:35)
Thank you so much for having me.

**Cindi Howson** (1:37)
Where in the world are you joining us from today, Andi?

**Andi Gutmans** (1:41)
I'm in the San Francisco Bay Area.

**Cindi Howson** (1:43)
Okay, the Bay Area. And yet, you travel the world. You've lived around the world. I think you have such a fascinating background, living in multiple countries, and working for two big behemoths in the industry.
So describe a little bit that background for our listeners, and maybe reflect on how much that international background has shaped you for the needs of enterprises today.

**Andi Gutmans** (2:15)
If we had AI back in the day, I would say translation and everything would be far easier. But yeah, I actually was born and grew up in Switzerland until I was 10 years old, to a British mother. So I started being Swiss and British. Then I moved to Israel at the age of 10 At about 20 years ago, I moved to the US. So I'm actually quad citizen. I also went to an American high school.
But that's really kind of shaped how I think about technology and how technology is actually able to flatten, grow old and just help people communicate. That's something I've been really excited about with AI. I mean, you see some of these advertisements.
Grandparents in China are speaking to their kids in the US, and they're speaking Chinese, and their kids are speaking English. That's not science fiction anymore, it works. So I think it's just been really exciting to see how AI is democratizing communications, and just the opportunity of technology to bring the globe together.

**Cindi Howson** (3:17)
Yeah, for sure. You've hit on some of my favorite countries to visit, and also having lived in Switzerland, married to a Brit, a lot of cultural experiences there.
As I think about AI for communication, it also is trust. We have different laws in each of these countries and world regions. And if you think about some of the headlines of late, some of it's a little scary. An agent deleting a database, a production database nonetheless, or a rogue agent granting credit for millions of dollars. What are your thoughts? Are we really ready for this?

**Andi Gutmans** (4:03)
That's a great question. I get the question all the time. I think there's a few different parts to that. First of all, there's a perception versus reality. So let me give you an example. If you take a Waymo, you're actually 80 percent less likely to get into a hurtful accident than if you take a service with a human driver. Still, most people prefer to take the Uber versus the Waymo because there's a sense of security that they feel.
I actually send my kids with Waymo because I know it's safer. So I think there is one element where it's going to take time to earn the trust to understand what quality outcomes we can drive. There are going to be certain things that we feel very comfortable making autonomous today. As an example, if you look at customer support today, there's a huge amount of debt that's really autonomous today. But as you mentioned, inventory replenishment for $10 million, probably companies still want the human in the loop. I think there's this huge spectrum of use cases.
What we're going to find over time as quality goes up, as we understand what controls to integrate, that we will increasingly feel comfortable having autonomous agents acting on our behalf.
Given that also humans make mistakes, my guesstimate is we're probably going to get to a higher level of accuracy over time than having an average human practitioner in many, many use cases. But that will take time and it takes time to build a trust. But we're definitely seeing customers already today deploy these autonomous agents in production and having them take action on the business list behalf. I do want to just touch on one thing you said, which is deleting a database, which of course I manage a large amount of databases for customers and analytics engines and storage. And I think this is regardless of whether it's AI or a human error, this is why using a hyperscaler like Google actually is beneficial to customers because we do have controls in place for soft deletes and other things where we can actually deal with some both human and AI-based errors. Of course, you shouldn't count on it. You should be careful and put your own controls in place. But there's a lot that we do as part of managing your data to make sure that we can try and keep you as secure as possible.