Name: Taming AI Chaos: A Unified Approach to Agent Governance - AI Adoption with Microsoft Copilot
Uploaded: 2026-01-14T22:56:04.924Z
Duration: 12 h 40 min 8 s
Description: Taming AI Chaos: A Unified Approach to Agent Governance - AI Adoption with Microsoft Copilot

Transcript for "Taming AI Chaos: A Unified Approach to Agent Governance - AI Adoption with Microsoft Copilot": All right, thank you everybody for joining us today. Super excited to talk to you about how you can unleash the power of agents without all the risk and how to create a framework that allows you to accelerate your AI adoption. We just launched the Rubrik Agent Cloud, we'll be featuring that as part of our conversation. I'm gonna briefly introduce our speakers and we can get started. I'm joined here today by Deb Ricci, our GM of AI, Jackie Ho, one of our AI product leaders, and Adam Key, who's a engineering leader here as well at Rubrik, who are working actively with hundreds of customers on their agentic workflows in our new solution. And we talked today about agent adoption and how you can build safe, reliable agents in production. With that, I'll hand it over to Dev Rishi to take us through the conversation. Thanks, Michael. Good morning, everyone. Today, I'm really excited to announce the launch of a new product we call the Rubrik Agent Cloud. To introduce our new products, I'm joined today by a few of my colleagues on the product and engineering team, who've been working really hard to bring this launch to life. As a brief introduction, I'm Dov. I was one of the co founders and CEO of Predabase, a general AI infrastructure platform. And today, I work as one as a general manager for AI at Rubrik. Now, some of you may be aware, Rubrik joined forces with Predabase, a generative AI infrastructure startup that hosted production LLM workloads for both digital native tech companies, as well as large fortune 500 organizations. We were built on top of open source projects and foundations in machine learning infrastructure that we had authored ourselves, and built a commercial platform that was really looking to be a world class general AI infrastructure platform. In partnership with Rubrik, a company that's been building leading data and security products, we're now looking to innovate on the next generation of products that are going to help the enterprise AI transformation. The way I look at Rubrik is that our company has always been a company helping accelerate business transformations. This started off with the automation transformation, conventional backup and recovery. Then came the digital transformation, where we needed a secure transition towards the cloud for enterprise organizations and the advent of the Rubrik Security Cloud product. And now the next way for enterprise transformation is the AI era. And that's what we're building for with the latest set of capabilities in our product. Now, as you can likely tell from the name Rubrik Agent Cloud, the enterprise AI trend that we've decided to underwrite is the rise of Adjunctica AI. It feels like these days, AI agents are everywhere. In fact, I can't commute back to San Francisco without seeing at least a few dozen new agent tools or a number of different ways to make building custom agents even easier. And I think that today, most leading industry experts agree. Agents are going to be on the rise inside of enterprises. Driving everything from enhanced productivity, decision making. Now, at this point, I find it helpful to be able to define agents. Because it's something that a number of organizations have had different definitions for and every billboard might say something slightly different. We define agents as simply LLMs with access to tools. So you can think of agents really as models that are able to take actions on your behalf. Now we think agents with this definition have the capability for enhancing the vast majority of enterprise workflows. Driving workflow productivity and automations. But whereas the idea that agents are coming inside of the enterprise may start to feel like a foregone conclusion. It does say, Lurie, some critical points in terms of what it actually looks like to be able to deploy, to operate and to govern these agents really at scale. We started to know this really firsthand because at Rubrik we're both building agents directly ourselves internally. And we've also spoken to a couple 100 enterprise customers over the last few months about what their AgenTeq journey has really looked like. And in today's webinar, I'd like to share more about what we've learned and how that's actually shaping our product roadmap for what's coming ahead. The first thing that I think is important to understand is really the stages of what we call AgenTic maturity based on the curve and the customers that we've spoken to. Now, we really think about there being four key phases in terms of the agent life cycle. That first phase is experimentation. Most customers might still be here where you're looking to build your first one or two agents, typically in a sandbox environment. Nothing really rolled out into any production setup today. The next stage we think about is formalization. At this point, you might've had one or even many successful pilots. You're increasing your investments and expanding the capabilities and access. And critically, you actually have your at least your first, but usually your first handful of agents that are actually deployed inside of your organization. The next phase looks like proliferation. You've gone from maybe your first handful to dozens or even hundreds of agents now in production. You have centralized governance and tooling for rapid deployment. Even if it's just your first version of what that looks like today. And finally, we have autonomous AI. You have up to thousands of agents. At this level, you have orchestration that's running independently and these agents might even be making independent decision making. Sometimes with human in the loop or sometimes even without actually that level of guidance. As we've spoken with again, a couple 100 customers, we actually took notes and surveys about where each customer was along the journey. And one of the things that we found fascinating is that almost all of the customers we spoke to had already started their AgenTik journey. Just about 90% had started building at least in the experimentation phase. And we saw that about half were still in that experimentation phase. But folks are making that move up towards the next set from formalization to polarization with the eye towards like an end goal of being able to do some full automations in the future. Of the 51% of organizations that were still in the experimentation phase, most of them said that over the course of 2026, so in the next twelve months, they're looking to be able to move that move, make that move over to formalization and productionization. But interestingly, the thing that I think is slowing down the adoption curve is not necessarily the ease of building agents. And in fact, there are many tools that help make building agents a lot simpler now. Rather from what I hear from a lot of IT and enterprise leaders is that it's really corralling the risk that agents pose. It used to be that when you thought about risks towards your business, you predominantly thought about disaster risk. You know, for example, a fire and flood. And then you started to change into cybersecurity risks, cyber hacks and others that would actually pose potential large amounts of damage through a surface area that they could start to access using the technology providers your company was using. But now we think that a lot of the future risk is going to become through AI. Which has the potential to be able to do 10 times the damage in one tenth of the time, due to the nature of how quickly these systems are able to actually operate. So while AI poses a large amount of promise for a lot of organizations, one of the things that's really slowed down adoption inside of the enterprise is the particularly unique risk vector that it poses. And we think that agents pose a unique set of risks for a few different reasons. The first are AI agents are superhuman. They operate with a nonhuman identity, are able to operate much more quickly than a traditional human would be able to. But they're still interacting with a lot of those same production work systems inside of an enterprise that were initially architected with a human employee in mind. And so this is proliferating the use of NHIs and tools that may not have been previously as well set up for them. The second real risk that we think about is that agents operating are using LLMs which are fundamentally non deterministic in nature. And so what that means is we're really giving production systems access via agents that may be either A, prone to hallucination or just B executing on a set of actions that they thought was the correct way for them to be able to go ahead and execute on these actions. We see this come up now in industry already with systems like coding agents that may have inadvertently dropped a production database, all the way to hallucinations that have led to leaked credentials and others. And so in practice, we're seeing that this risk and uncertainty is one of the key things that is really slowing down production AI deployments. In speaking with security, IT and engineering leaders inside of organizations, One of the key challenges that they've really been thinking about when it comes towards this risk is how it looks to actually be able to do a centralized governance and visibility across a number of different production applications and agents that they might be building. The most consistent question that I hear is how do I get a single pane of glass? How am I supposed to be able to tell for example, where are all the agents that are actually actively running inside of my ecosystem? And what tools and data can they have access to? Before you can build a really good and resilient governance or security framework, you need to be able to start with visibility. What's actually running and what's the blast radius that I might have directly there? From there, the types of questions come into, how do I stop leaks and attacks? How do I measure and improve quality? And then finally, agents present a particularly unique set of challenges, I think for anyone who's been thinking about recovery and resilience. Which is how do you recover from an AI mistake? What do you do when something has gone wrong with a production AI system? For us at Rubrik, this is the exact set of challenges that we think we are uniquely positioned to be able to solve. And one of the reasons we think that this is particularly compelling is that building agents now, I think has never been easier. There's a number of managed tools all the way from open source frameworks like NA to N or LangChain through managed platforms that make it easy to go from initial proof of concepts to actually a demo in days or weeks. But from our firsthand experience, what can actually end up taking a lot longer is being able to go through the entire governance process of making sure you have design docs, AI committee approvals, are able to get into your production ready agent state. And that you've submitted this into a framework that actually works well for your entire organization. Now, our experience, a lot of these systems are set up for really good reasons. But do slow down the overall cadence of AI agentic adoption, as well as introduce a lot more overhead for the centralized committees that need to be able to run these processes using paper and manual processes today. This is the exact challenge that we think we're uniquely well set up to be able to solve with our eyes on the Rubrik Agent Cloud. Rubrik has a unique position to play in this market because of the core roots that we've had in data, identity, and now the Predabase acquisition, AI platforms. The Rubrik Agent Cloud really combines these three key assets, an underlying understanding of what data an organization has inside of it, as well as the metadata for how that maps towards applications and schema. The identity layer that we've integrated into the product in order to be able to help with resilience that understands access and access management. And then the Predabase layer, which brings in enterprise LLM infrastructure that can deploy inside of a customer's own cloud and allows you to be able to access information that's critical for running agents in production. And we've taken these three key ingredients. And today are introducing what we call the Rubrik Agent Cloud as our product, which really comprises of three key pillars. The first pillar is monitoring observability. As we mentioned before, you can't actually govern or secure what you can't see. And so monitoring and observability starts by giving you full visibility into the agents that are running and the applications and data that they can access. The second pillar is governance. Governance can mean different things for different people. But for us, the high level framework we think about is the ability to define policies and then enforce policies or notice detection and violations of policies in real time from agents that are running. This takes governance outside of the simple people on paper process route and actually into something that is embedded into the core technology workflows you're using today. And then finally, the last pillar of our Rubrication Cloud is remediation. We've always had a mentality that something can go wrong and eventually inside of a business will go wrong. And you need to be resilient when something goes wrong. We've taken that in a step further in our remediation pillar with a new capability that we launched called agent rewind. That allows you to undo a destructive action that an agent has taken on a production system by being able to restore that from a previous healthy state in snapshot and backup. Our view is that the Rubrik Agent Cloud provides the secure and governed way to be able to operate agents at scale inside of an enterprise organization. And is going to accelerate the adoption of AI by removing some of the fundamental roadblocks that stop enterprises from being able to adopt this at scale. The Rubrik Agent Cloud is building on some of those same key substrates of data and identity that the Rubrik Security Cloud built, but is unleashing a whole new set of capabilities for us to make the modern enterprise AI ready. And so to go deeper into how this works and also to share some of our own firsthand experiences with it, I wanna hand it over to Adam, engineering leader at Rubrik and working on the Rubrik Agent Cloud directly with us. Go ahead, Adam. Thanks, Dev. First, I wanna start by talking about customer zero, our own internal Rubrik AI governance committee. We at Rubrik were trying to deploy hundreds of use cases for our agents across our org. These tools included desktop assistance, low code and no code agent building platforms, and bespoke complex agents we're building ourselves. As Dev mentioned, building agents, prototyping agents is easy part. Getting them in production is where you really hit the challenge. So we had complex requirements for many stakeholders across the org, from InfoSec, legal, privacy and IT, and a complex people and paper process for getting approval. Both a heavy weight on our stakeholders, as well as the developers themselves. And even once we're through this approval process, we had approval, but no real audit ability going forward. So again, the core concerns for everyone was visibility, governance and remediation. Let me talk about the life of an agent or an agent request. The flow starts with a user issuing a query. For example, what's the status of my order? Then an agent will send that prompt to the LLM along with a list of tools available for the agent to use. For example, Check Order Status, Return Item, or Charge Account. The element will select the tool call and the agent will invoke it on its path. Finally, the response comes back from the tool call and the response is active user. The key questions here are a couple. First, what users are interacting with my agents? Second, what agents exist in my organization? Third, what apps and data do my agents have access to? And finally, what specific actions are being taken by agents? If you take all this together, you wonder, are my agents compliant with our policies? So how does the Brick Agent Cloud work? Our AI gateway discovers and tracks all agent prompts, tool sets, and actions against your system and provides you the ability to monitor. We build first an agent inventory, the list of all agents acting within the system. Second, a tool inventory, the collection of all the tools that all the agents have access to. An agent map, which is able to display both the agents, the tools, the users, the actions. And then a policy engine that allows you to put in place automated policies for governing these agent actions. And finally, a remediation system. When an agent goes wrong, how can I help rewind the destructive action that agent took? Here we're trying to unleash agents, not risk with the Rubrik Agent Cloud. Again, the first agent operations platform that allows you to monitor, connect popular agent builders, instantly and generate signal of a visual map of all agents and actions. Govern, the ability to allow you to set policies which contain guardrails, monitor behavior and enforce in real time. And finally, Rewind, the ability to roll back from destructive actions of an agent they can quickly bring your business back online. Please sign up for early access. And with that, let me hand it off to Jackie, who's gonna give us a demo. Awesome. Thanks, Adam. So let's give you a quick demo of how the Rubrik Agent Cloud, Agent Operations Platforms work. This is your centralized governance platform for monitoring all the agents across your organization. So for the first pillar, you can connect agents via our model gateway. We'll go here to gateway management. This lets you work with any custom agent you've built using popular large language model providers, as well as privately hosted closed or maybe open source models that you have on Bedrock or Azure, like Azure OpenAI. So if you're using platforms like EndN, Lanechain, this is a great method for integrating those agents into the platform. We also support integrations with some popular agent builders, such as Microsoft Copilot Studio, Bedrock Agent Core, and some other ones as well. So after you connect your platforms, we'll automatically discover all the agents running across your organization and provide that centralized view of all your active agents, high risk agents, inactive agents, all in one place. You can also see the apps that they're accessing as well as who's the owner for each of those agents. So let's click into one particular agent, the M365 Data Protector. So here you can see over the last seven days what identities have been using this agent, what applications this agent is using, as well as what tools are associated with that application, and also any actions that have happened in the last seven days. So in particular, there's a delete file action that occurred in the last seven days. We also have that activity log, which is your audit trail of all the actions taken by this agent. You can see here for this particular agent. So for the next pillar, Govern, let's take a look at our policies. The agent operations platform lets you configure predefined or custom policies based on your organization's unique needs. So essentially in reference to what Adam was saying, if you could apply the policies that your AI governance committee are actually setting forth and see if your agents are actually following them and also have a way to enforce them. So you can apply fine grained controls on agents, tools, applications, identities, data, outputs, and more. So whether it's something simple like agents only have read only tools, showing these read only tools or actions, or something more complex like I don't want Salesforce agents reading data from SharePoint. So let's go ahead and dig into the violations tab. Policies trigger violations so that can be viewed per agent or across all your agents. So here I can see which violations are triggered by actions, as in something actually happened, or I can also see if a violation was triggered by the platform detecting that the agent could access something it shouldn't, even if it hasn't yet. So for example, in this second violation, which is a configuration violation, I can see that agent has access to an unauthorized Salesforce tool, which we're seeing in the system prompt or the agent's configuration. So let's go ahead and dig into one of these violations, which gets us to the last pillar, remediate. So here I can see the M365 data protector agent deleted some records erroneously in SharePoint while attempting to scan for a credential and violated the read only tools policy. I can also see the Asia blogs here, to get an understanding of what led to the deletion. And lastly, one of the most powerful features, if I go here to remediate, I can actually rewind the action that this agent took by restoring the data that was deleted. Our platform uses an AI agent itself to extract information from the tool call and intelligently figure out what exactly was deleted. Since I also want to prevent this agent from being able to delete in the future this agent as well as future agents from accidentally deleting SharePoint data, I can take advantage of another remediation method which is tool clocking. So I'll go here to the tools inventory and here I can see a list of the tools that I've discovered and things I actually see in approval status. And for the delete file tool, I'm going to go ahead and mark this tool as unauthorized. And since I have that policy which blocks unauthorized tool access, now future agents can't use this delete file tool on SharePoint. So that's a quick look at the products in action. Are you tired of long review processes? Do you want to get agents into production faster and more confidently? The Rubrik Agent Cloud could be your way to unlock more agents at your organization. Sign up for early access today through the link provided and we're looking forward to working with you. Awesome, thank you presenters for walking us through sort of the state of the agent workflow and how teams can start to prepare themselves and address some of the common issues that are slowing down AI adoption and acceleration. I wanna jump into some of the Q and A, into some of the questions that we're getting from the audience. So maybe here's one, first one for Dev. Do you actually need to be a customer of Rubrik Security Cloud in order to take advantage of the Rubrik Agent Cloud capabilities? Yep, thanks for the question. One of the unique things with Rubrik Agent Cloud is you can actually get started without being a current Rubrik customer today. So we don't have a dependency on backup in order to be able to actually run our key pillars on monitoring and observability, as well as even governance. The way you can think about the Rubric Agent Cloud, is it's something you can onboard to with really no additional external dependencies or prior contract commitments and start to use. We're signing up our first set of beta and design partners on it now, and we're looking forward to onboarding you for early access. The unique thing is that while you don't need to be a Rubrik backup or RSC customer, if you do subscribe to the services, we can augment your experience. So using some of the key signals that we have from within RSC, if you're an RSC customer, we can add additional signal that we can trigger in for our governance layer. And it is also the other part capability just for the agent rewind feature. And so Rubrik does need to protect the data that we would be looking to allow you to be able to do agent rewind on in order for that particular feature within the remediate pillar to work. However, for the rest of the platform and in particular, to get started with monitoring observability, as well as governance, you can start as a net new customer. Awesome, great to hear that people can start cataloging all their agents, observing them and applying policies right away with just the Agent Cloud platform. That's great. Another question came in, I'll give this one over to you, Adam, about MCP. So organizations are pretty eager to adopt MCP, but adoption has been slowed for numerous reasons. What are some of the key challenges organizations face from your experience in adopting MCP for their agent workflows? Yeah, great question. Thanks, Michael. I mean, NCP is obviously a very powerful technology. It's a really easy way to connect agents to your to your systems, your production systems, and and let them take action. I'd say there's sort of two big challenges. One is that, you know, again, because of how easy it is to connect to your systems using MCP, anybody can do it. And it's know, that's one of the core things that motivated us to build agent cloud was this sort of ungoverned tools access. And I think, you know, in this new agent world where it's so easy to get started building, downloading, low code, no code, connecting with MCP, we see a a real big need to sort of understand, you know, what is actually out there being used for MCP and what kind of things can those tools do. Second is MCP is a, you know, fairly new protocol. I think it's, like, just about a year old. And so a lot of the standards around authorization, authentication still evolving. So again, I think, you know, MTP is a great enabler for tools, but I think a lot of the sort of security and governance is is still still being developed. And so I think that's the big challenge for enterprises today. How to actually use it safely? Makes sense. So having that layer in between of observability and policy management, it seems just as critical. Jackie, maybe this one's for you, since you walked through some of the capabilities in detail. There's some questions about policies, policy management, specifically how flexible and customizable are the policies? I saw that you created one, but maybe you could talk about the ability to be a bit more granular and can people actually write their own policies to enforce? Yeah, that's a good question. So we are initially starting with more rules based policies, but definitely see that organizations have a lot of unique policies that they're looking to do as well, especially when you go to speak with them, so we're also exploring doing policy definition using natural language as well as using customizable small language models to do things like PII detection or data exfiltration. Awesome. So I continue capabilities around doing real granular policy management. Sounds like it's being actively built, which is exciting. Maybe this one question can be for you, Dev. Does Rubrik Agent Cloud just work on agents that are deployed or can it pre deployment? Yeah, that's a great question. I think when people see the governance functionality that we have done it out here, we often see that this makes a lot of sense kind of at runtime, but some customers also wanna be able to do this at the post development stage. So you can think about that as the point where you've created that demo, haven't quite started to run it in production. And you're just really looking to be able to test and validate your policy guardrails are working as intended before you roll it out. We can actually operate in the setting before you were deployed into production. And the way you can do that is you can just proxy your tests or staging traffic directly through either our AI gateway or using any of the prebuilt kind of integrations that we have. And so as you start to use this in sort of a test or sandbox setting, that also will populate within the context of Rubric agent cloud. And it's a really good way to not only make sure that you have the right sanity checks in place, but also that the governance framework that you thought about pre production is actually working as intended. We're looking to roll out more features to be able to support this kind of testing early soon. Awesome. Yeah, maybe on that note, are there other areas that you're pretty excited about for the future of sort of agent management They'd be the working on or you see as like, you know, future opportunities for people to consider. Yeah. You know, the key challenges that I feel like I hear from organizations working with agents today is like, number one, how do I get visibility across my entire agent ecosystem and surface area? I think we're doing the right set of job things to be able to talk about today. Number two then, looks like, how do I start to be able to do even more advanced agent workflows to be able to help understand the performance of agents and improve those over time. At ProtoBase, we've had a lot of background and experience in helping organizations improve models and agents from a quality standpoint. And I think as we extend our governance area, that's gonna be an opportunity for us to be able to reintegrate some part of that technology stack as well. To help organizations improve the agents' performances. And then finally, I think we're gonna be able to go even deeper into how we do some of the remediation capabilities as well. So you saw how AI agent rewind works through the demo that Jackie gave. I think we're going to have a set of capabilities that are gonna allow us to do even more deep and granular things through that. Nice. Yeah. It seems like governance is the hair on fire problem today that we need to solve, right? So we get agents out there, but love hearing about them. The continued development on things like improving agent performance, something that we have a strong background in as well. Another question that came in for you, Jackie, you had kind of briefly talked about integrations with some of the popular agent builders. What are the ones that we support today? And yeah, and maybe talk a little bit about kind of how we think about the level of support across these different tools. Yeah, good question. So there sort two ways to connect your agents. So there's the the builder tools such as Microsoft Copies Studio and Bedrock Agent Core that we are launching in this first wave, and then we're also adding additional platforms as well such as Salesforce Agent Force. And then of course if you're building custom agents using like a Google ADK or an NADAM language name, you can always use the model gateway method. Basically any agent builder where you, instead of pointing directly to OpenAI and you can actually swap out the large language model that you're using, totally works with the Rubric Engine Cloud. Cool. So if you connect to the API, essentially you can start cataloging all agents across these popular tools. That's great. And so there was a follow-up question, or just separate question, just probably follow-up. Agent Cloud then creates a single inventory across all these tools. I know some of these tools have their sort of like unique, you can look at just the agents with them, but this then aggregates them, is that correct? Yeah, totally. I think what we've heard when speaking to a lot of organizations is people are typically using more than one builder tool. I think especially at these large organizations, it's definitely two or three or more different build tools that they're using. So as you said, some of these platforms have ways for you to monitor and govern maybe not govern, but ways to monitor within just that platform. But what we're seeing for a lot of these larger organizations is maybe that works for kind of a developer who's just working on, let's say, just agent force agents or something like that. But especially in a larger organization when you have IT and security and they also want that view to agents, logging into all these different platforms and kind of working at the level of the AI developer starts to not really make sense. So what we really heard is this pain around being able to see ASUS across all the platforms in one place and be able to apply the same sort of policies and restrictions across all the platforms rather than having to configure and set it up for each one. Makes sense. Yeah, and it's nice too, to have all your users sort of like looking at one single pane of glass or single dashboard of all the agents across these tools together. Adam, another one came in thinking about, you know, flagging of agents that are risky. How do we determine if an agent is higher or low risk? What are some of the logic behind that? Yeah, good question. I think Jackie might've highlighted a little bit of this during demo, but there's a few different dimensions that we think about risk on. So one might be access risk. So what applications and tools I guess or sorry. Access risk. What permissions does this agent have access to? What systems can it touch? Then we talk about action or capability risk, which is what tools does this agent have and what types of specific actions it take? For example, can it read? Can it write? Can it delete? So that's sort of tools risk or action risk. Then there's activity risk, which is not just like what tools does the agent have access to, but what is it actually doing? Because we can see the history of actions. We can start to think about, not just again, it can access, what tools it has access to, and then what actions it's actually taking. And then finally, we think about sort of behavioral risk or, has the agent misbehaved in the past or is the agent misbehaving as it relates to maybe it's system prompt, let's say, or its intended purpose. So again, a lot of different ways that one can think about risk when it comes to agents. And we're trying to build a flexible policy engine that can allow you to implement the type of risk that you're most concerned about. Awesome. And another question to me, Dev and Jack, I know you're working with a lot of early customers and talking a lot of organizations. What are some of the common use cases that you're seeing for AgenTeq workflows today? The really nice thing about AgenTeq workflows is I think that they tend to be something that every organization has. Oftentimes we hear like hundreds or more use cases inside that they're looking to be able to tackle. So we've worked with a number of organizations that do use case scans or have a hackathon project across the company where they're building out agents. And what they come up with, even internally at Rubrik First Party, is the over a 100 plus use cases that they're looking to be able to tackle. You can think about use cases, anything from like back office automation to making some support functions operate a little bit better. The key challenge actually, I think is not the ability to dream through the use cases or even to build out those first demos. But what we hear is just the practical friction of needing to review a 100 plus designs for being able to solve out those use cases. What we see is that tends to be one of the biggest inhibitors. And so organizations often have to go ahead and pick one use case that they're gonna start off with. There are a few and use that to be able to like test out the framework or policy. You know, Rubrik, we ran through this as well as we had our engineering team that was keen to be able to build more of these agentic use cases, but we needed to make sure to be able to do so continuously and responsibly. And I think what we asked for was rather than having to have every use case individually reviewed each time in-depth data access tools logs, you know, what we asked is like, let's just get a let's just ask for and create a framework. As long as we're operating within that framework and as long as the framework has the appropriate teeth into the actual technology that we're developing, then we feel like we actually have the license to be able to innovate, you know, as we'd like. And that's a large part of the motivation that's led to our internal piloting through Rubric Agent Cloud as well. Awesome. All right. Well, I think that wraps up the questions. If we did miss one, we'll definitely get to it. I wanna thank our presenters again, Dev, Jackie, Adam for walking us through that today. Great conversation about the state of agents and agent adoption, common pitfalls that we see, the hundreds of customers we've talked to as they onboard and trying to deploy agents, and then how to start to remove those blockers, move faster and put agents into production confidently with a framework like Agent Cloud that allows you to observe them holistically as a team, create the policies that you gives you peace of mind and then have a mechanism to remediate them should something goes off the rails because you should always expect your agents, right? At some point might do something that you didn't anticipate. So love seeing the full solution. Appreciate the good dialogue. And I again wanna invite everybody who joined us today. Thank you for joining us. If you're excited about Agent Cloud as we are, please visit rubric.com to sign up for early access. We're onboarding customers right now since we just launched it and excited to get you started. Thanks again, everybody.