At Dreamforce 2026, Savinay Berry of Salesforce, executive vice president and general manager for Data Management, discusses enterprise artificial intelligence, or AI, agent governance and enterprise data management. Berry highlights the Enterprise AI Harness and explains how task-level policies, explainability and a unified data catalog ground agents with context. They emphasize the importance of connecting agent development with data governance to prevent agent sprawl and ensure reliable outcomes.
The discussion examines composable integration across systems such as Snowflake, Databricks, MuleSoft and Informatica and the role of open, end-to-end platforms in enabling faster, measurable time to value. Analysts from theCUBE note that this approach helps chief information officers and chief information security officers, or CIOs and CISOs, govern at scale while maintaining compliance and delivering measurable business outcomes.
Topics covered include enterprise AI governance, agent governance, data catalog strategies, composable integration, platform openness, explainability and AI compliance for enterprise data platforms.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Dreamforce 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for Dreamforce 2026
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for Dreamforce 2026.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Dreamforce 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to Dreamforce 2026
Please sign in with LinkedIn to continue to Dreamforce 2026. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Savinay Berry, Salesforce
At Dreamforce 2026, Savinay Berry of Salesforce, executive vice president and general manager for Data Management, discusses enterprise artificial intelligence, or AI, agent governance and enterprise data management. Berry highlights the Enterprise AI Harness and explains how task-level policies, explainability and a unified data catalog ground agents with context. They emphasize the importance of connecting agent development with data governance to prevent agent sprawl and ensure reliable outcomes.
The discussion examines composable integration across systems such as Snowflake, Databricks, MuleSoft and Informatica and the role of open, end-to-end platforms in enabling faster, measurable time to value. Analysts from theCUBE note that this approach helps chief information officers and chief information security officers, or CIOs and CISOs, govern at scale while maintaining compliance and delivering measurable business outcomes.
Topics covered include enterprise AI governance, agent governance, data catalog strategies, composable integration, platform openness, explainability and AI compliance for enterprise data platforms.
In this interview from Dreamforce 2026, Savinay Berry, executive vice president and general manager of data management at Salesforce, joins theCUBE + NYSE Wired's Alison Kosik and George Gilbert to discuss how the company's new Enterprise AI Harness aims to unify fragmented agent development under one governed, enterprise-wide platform. Berry recounts a conversation with a CIO at a major aerospace company whose teams built agents in isolation from data governance efforts, leaving the enterprise with no unified visibility. That disconnect, he explains, is exac...Read more
exploreKeep Exploring
What problem does the Enterprise AI Harness aim to solve?add
What is an "Enterprise AI Harness," and why will it become the dominant end-to-end platform for companies to run, operate, scale, and govern AI agents across the enterprise?add
If someone wanted to run a proof‑of‑concept requiring this AI harness, how can they quantify and set expectations for the platform, its deployment, and the first agent built on top of it?add
(INTRO)>> Welcome to theCUBE NYSE Wired program and community coverage of Dreamforce 2026. We're coming to you live from San Francisco. I'm Alison Kosik alongside George Gilbert, and we're about to dig into how AI agents are becoming more and more powerful, more autonomous, but they're only as reliable as the data and rules behind them. Right.>> You know, it's, it's one of those lessons that we kind of had to reteach ourselves and customers how to relearn, which is AI is programmed with data, and we've got more agents for the last few years. We've had more agent development toolkits than you can shake a stick at. I've got a more colorful term than just that. But it comes back to who has the data and has modeled it and then who has the actions or the workflows that operationalize that. Yeah. And that's— we're back to Salesforce.>> Yeah. All right. Well, let's dig into the topic and more. I want to bring in our guest today. Savinay Berry. He's the Executive Vice President and General Manager of Data Management with Salesforce. Thanks for stopping by theCUBE.>> Excited to be here.>> So let's talk about how customers already had Data 360, MuleSoft, Agentforce separately. So what's genuinely new in the Enterprise AI Harness versus a repackaging of existing capabilities?>> Yeah, so I'll talk about a very interesting anecdote that just happened a few days ago. In preparation for a lot of the announcements we were going to be doing, I was talking to a lot of customers, right?>> Of course.>> And one of those customers happens to be a CIO of a very large aerospace company in the country. We would all know that. And he started talking about how they had started to look at all these agents that were being deployed in many different parts of the company. Yet he has very little visibility into what was going on. So he was like, well, that's a different part of the team that's doing the agent's work. And then there was another part of the team which was looking at what you would consider as traditional data management, which is going off and looking at data quality and data governance and those types of things. And they were not talking to each other, literally in this large, massive company, they were not talking to each other actually. And he realized that that's a problem. We need to find a way to culturally change ourselves, but more importantly, change ourselves with the right tool, with the right solutions. And when he started talking about how he needs to connect the dots between how the agents are built, governed, scaled, operated with the right context that needs to be given to the agents for the right data with the right data governance and then be able to measure all the costs associated with that, I said, that's it. That's the exact problem that we're looking to solve for companies and industries. Now multiply that times 100,000, 500,000, across every single company that needs that type of a solution. That's what the Enterprise AI Harness is. It truly is going to become one of the most dominant enterprise AI platforms end to end, which becomes a way for any company to be able to run, operate, scale their agents, as well as be able to ground them with the right data to get the best outcomes.>> Perfect.>> So is it fair to say that, to use a movie analogy, I start with The Hunger Games where all the contestants start, they're gonna run to the center where they get the weapons. Every vendor, whether it's Anthropic or OpenAI or Palantir or Microsoft or Salesforce, at the center is that harness.>> Yeah.>> And it's the data and the data model and the tools, but tools that are not just read and write to the file system, it's to invoke a workflow. And so we had a few vendors who started off with, hey, I got this quick and dirty harness. But if you want real work done and you want it governed, you need the real end-to-end platform.>> That's exactly right. Harness as a word sometimes gets overused as well. Buzzwords in the industry these days. So harness in this, there's a lot of micro or point harnesses, a coding harness, for example, that's different. You have Claude as a coding harness. What we're talking about here is truly an end-to-end enterprise AI harness, which covers all the capabilities I just mentioned. So it's almost like the world of cloud. You remember the world of cloud when we started like 10 years ago, 15 years ago, every company, every person, any developer in the company was creating their own cloud applications on different clouds. Accessing different services, but then they run into trouble with the CIO.>> Yeah.>> CIO's like, no, no, no, timeout. We got to get this all governed into a single cloud platform. And that's what happened. You ended up having 3 major cloud platforms that started to become the dominant ones over time. And we have them today. We think that same thing's happening in the AI world as well. And so yes, it connects the dots. And what's really interesting is it's open and composable. So it's not, because we heard this from customers, Hey, you are Salesforce. If you do something, make sure you do it with the entire enterprise in mind, not just CRM, not just marketing, not just service, but the entire enterprise in mind. So it's truly open and composable and it's across the entire enterprise.>> So I want to unpack that some more, the composable. So let's say a customer has Snowflake or Databricks or both. Exactly. How do you engage with them? Do you say, we have the Customer 360 model that sits on top of that data and harmonizes it. And then we can bring in, MuleSoft and Informatica and then, so.>> That's right. So bring your own data. If you want to bring your own data, you can bring your own data. We have a unified catalog. We have a unified set of semantics. We have a unified layer which can connect this and then normalize that and be able to provide then the right context to the agents. If you have a different coding harness, you can bring that coding harness as well. If you have a different way to do data integration apart from Informatica, which is what we would have as the default, bring that as well if you'd like. So it's truly one of those open and composable platforms which starts off with a default set of all the capabilities we have from Salesforce. Yeah, for all Salesforce and non-Salesforce solutions. But then if you need to bring one, you can certainly bring one. It becomes the full enterprise platform.>> Let's talk about trust and guardrails for a second. Can you give me maybe an example of guardrails stopping an agent from doing something it shouldn't do?>> Great question. And I think this is where the trust and compliance becomes critical. So we had a situation recently actually where there was an agent at a customer, it was a contracts management agent. The contracts management agent was trained with the right documents for procurement, supply chain, et cetera. And it was supposed to take a look at all the contracts, decide which ones get deflected to human, which ones get automatically approved. And in some cases we started to see that there was an approval that was not supposed to be done. So it should have been deflected. And the reason we found out is because we have explainability, which is a term, an actual term in the agent world, that explains what the agents were doing. And we were able to see that ahead of time. And so that governance was able to stop that kind of approval from happening on its own. It had to be deflected to an agent, not to a human. So that governance has to happen at the task level for agents. It cannot happen at the application level. So this micro granular approval process is part of what the AI Harness will also have.>> So I just want a quick clarification on that last answer.>> Sure.>> When you're governing at the task level, is that part of the observability?>> It's actually more than just observability. So you shouldn't think of the AI Harness only as the collection of all the different products. Yeah, the collection of the products, that's a starting point. What we're doing is we're going way ahead of that and we are now integrating things like the catalog, the governance layer, the semantic layer, and the policies. So the policies and the agent identity, the agent identity now is part of the scoping exercise. So imagine If you have a human identity, if I create an agent for myself, it's not going to inherit all the policies I have as a human. It's going to inherit only the task-level policies based on the scope that I've defined for the agent. And that granular policy is so critical in this world.>> And is that defined now at the Agentforce level or at the Data 360?>> It's at the agent level. So Data 360 then becomes part of this. Agentforce Studio becomes a way to build the agents. So it truly becomes an end-to-end platform where the customers don't have to do the connections themselves. That's what was happening. They had to do all these connections themselves. Right.>> How has the conversation shifted over the past 12 months where the technology side of the house often stood up these little agent POCs really quickly, ones that didn't have system dependencies. But then they couldn't do a lot of work. How do you engage with the customer now where you need to get to the technology side of the house, which Salesforce hasn't traditionally been engaged with, but then the business side of the house owns the outcome, but you're now in the best position to bring those two together.>> That's right. So that's the beauty of it in some ways. Salesforce traditionally with all the clouds, Revenue Cloud, Sales Cloud, Service Cloud, very strong on the business side. Now you bring in Informatica, you bring in MuleSoft, very strong on the IT side of the house. So you bring those two things together, you start to get the respect from the CIO, the CISO, and the CDO in addition to the business users. And if you combine those two things together, you can get the data context, the grounding, and you reduce the agent sprawl. Agent sprawl is a word that gets used a lot these days. It's the same thing as SaaS sprawl in the world of cloud. So you reduce the agent sprawl, and you can now govern it. And that's what's changed since last year, where people, the CIOs are much more focused on the governance and compliance, less so about the bright shiny object of a cool demo.>> But just in terms of time to value, when you had those standalone ones and, well, the more colorful term I had was that there are more standalone agent development tools than there are fleas on the average camel. But you could get something up and running really quick.>> Right.>> But now that you have this harness.>> Yes.>> There's more of a lift in having that in place. How much work is it to get a minimum viable component of that? Maybe not an end-to-end deployment, enterprise deployment, but the minimum viable product of that platform, of what we now call the Harness. So that then you can put a governed agent that you can operationalize.>> Yeah, so one of the things we are listening very deeply to our customers on this. And one of the things that we are going to do is we won't require them to go out and procure new licenses or subscriptions for existing products they already have. There is no migration as such. if they have existing products, great. We are now leveraging the best capabilities of it. So things like Data Catalog, the Agent Context Engine, the security and the policies, and the agent building experience, pulling that together into what a basic capability might look like for the AI Harness. And then we keep adding that over time so customers don't have to think about, oh my God, now I need to migrate suddenly. No, no, no, no. If you have something, great, we'll inherit that. You now have the single place to manage and observe and scale and operate everything. But if you don't, great, now you can buy it. License for the platform and you can pay for what you consume.>> Okay. So is the, if someone wanted to do a POC, but it needs this AI Harness. Yeah. is there a way even to quantify how to set expectations for this as a platform, the deployment of that platform, and then that First agent on top of that?>> Yeah, great question. So think of it this way, if you are not a user of Salesforce at all today, you don't use any Salesforce product, you start off with that and your immediate quantification is how you can get discovery of all your agents that you have in your enterprise. Starting off with that, simple. Starting off with the data catalog, doing all your data audit. So audit, first checkpoint, simple, easy.>> So this catalog, it's not, that's separate from Fabric?>> Correct. So Agent Fabric is a capability that enables all the observability and monitoring of all the agents.>> Okay.>> The Data Catalog is the data underneath for all of the structured data in the enterprise.>> Oh, okay. Got it.>> Yeah. Which is critical for enabling the context for the agent.>> Okay.>> Okay.>> Let's look ahead. What's possible 2 or 3 years from now? Let's take out your crystal ball.>> Yeah. Yeah.>> Once companies truly trust AI agents to act on their behalf.>> So one of the stats that I heard was as humans over the last 40, 50 years, we have created about 10 to 15 zettabytes of data ourselves. And if you think about what the agents can create on their own, there's probably going to be maybe millions of agents, billions of agents, honestly, in the enterprise over the next few years. They're going to end up creating about 150 to 200 zettabytes of data. It's almost 10x.>> It grows the brain.>> Exactly. So think about that scale for a second.>> Yeah.>> How do you navigate that? What do you do with it? How do you make sure that it's compliant? If the CIOs and CISOs are already pulling their hair out on how to govern what they have today, imagine what happens in 2 or 3 years when that level of scale happens. So I think we'll be in the mode of ensuring that the surface area is contained, governed, while at the same time we give enough flexibility for the agents to do what they can. So it's going to be an exciting few years ahead.>> Savinay, great to have you on theCUBE. Thanks for stopping by.>> Absolutely. Great to be here.>> And you've been watching theCUBE, NYSE Wired program and community coverage of Dreamforce 2026. We'll be right back.
