How To Secure Enterprise AI Agents With VMware Tanzu

Evan Kirstel (0:01)

And it's Evan here, and I'm at VMware Explorer Blockbuster Week. So much happening, and really excited to dive into the Tanzu platform with Pramina. How are you?

Purnima Padmanabhan (0:10)

I'm doing well. Good to chat with you again, Evan.

Evan Kirstel (0:12)

It's been one year almost exactly, and there is a lot going on. So much has happened before that. Perhaps introduce yourself and your team and your mission within VMware.

Purnima Padmanabhan (0:23)

Perfect. Purnima Padmanabhan, I'm the general manager for the Tanzu Division here at Broadcom.

And what does Tanzu stand for? Well, it is all about helping you build applications and agents along with the data that goes with it, right? End-to-end. How do you build it with good code, secure code? How do you build your applications and agents? How do you run them securely?

And then how do you connect it to the data that it needs, whether it is large data or whether it is in a database?

Evan Kirstel (0:55)

It may sound so simple, but there's so much complexity behind that, including developers and tools and tooling and platforms. Maybe talk about first what customers have been telling you over the past year about running enterprise AI agents on a private cloud. This has been all the rage, of course.

Purnima Padmanabhan (1:11)

Yes.

Evan Kirstel (1:11)

But what are they telling you personally and through their colleagues?

Purnima Padmanabhan (1:16)

Private AI is actually something that has now had a significant bump. And it's hard to talk about AI just in one swoop, because AI has impacted both in terms of creating new threats as well as creating new opportunities. So, I'll start with the threat side, because with the Frontier AIs really, in the last six to seven months, that has been the primary conversation. And the threat is that the Frontier AIs are now able to hack into any system, and that is what people are very, very scared about. And the conversation usually with Tanzu starts with my applications. Customers say, hey, I have got a lot of applications. Most of these applications are built, 90 percent of these applications are built on open source. And that open source could be either Spring or Java, very, very big in Spring and Java. It could also be Python, Node.js and others. And they have started coming to us, Broadcom, to say, can you help secure the open source for us? You already are the primary committers of Spring.

We already depend on you for the Spring and Java dependencies. Can you expand it to other libraries, maybe other open source databases? And that is where the first part of the puzzle comes in. And in response to that huge push of the AI generated threats, we have also announced TrueSource by Broadcom. And TrueSource by Broadcom gives you that secure supply chain of open source solutions and open source packages, libraries, images that enterprises can simply get from us, and we will stand behind it.

The second area is more around as customers start using AI more and more to address the opportunity, right? They are primarily trying to go beyond rag and chat and all that. They are primarily going to solve, trying to solve the business problem that they may have, usually business workflows.

And when you try to solve business for business workflows, there are a few problems. One is, I'm building all these agents, I'm giving a lot of autonomy to agent, I'm giving a lot of agency to these agents. How do I make sure the agent stays secure, right? The whole OpenAI leak that happened, or the Alibaba agent hacking that I talked about, they want agents to do just what it is being asked to do. So how do I sandbox an agent? How do I secure an agent? Has been a big conversation. But then as they try to do the workflows, the second conversation is an agent is only as useful as the outcomes it delivers, and the outcomes depends on what data you give it.

These data projects are what are slowing down these agent adoption, the ability to build workflows. So we are seeing customers ask us to say, help me give these agents quickly the data that it needs in a governed way, in a synthesized way that can scale out. And so those are some big problems that we see on the opportunity side that we are looking to solve. And I will say, along with that also comes in TCO. Also help me do all of this so that I'm not burning my tokens, burning my Cloud, burning in Cloud resources, or burning a lot of memory.

Evan Kirstel (4:41)