Name: Agent Operations Platform Workshop
Uploaded: 2025-11-12T20:31:35.947Z
Duration: 40 min 24 s
Description: Agent Operations Platform Workshop

Transcript for "Agent Operations Platform Workshop": Hello, Hello, folks. folks. Welcome. Welcome. on in. Thanks for joining us today for this Rubik workshop over Rubik Agent Cloud. We're excited to be hosting y'all today, especially with this this new thing that we have going on with our agent cloud. Appreciate you joining us, learning about the product. My name is Chase Macra. I'm on the marketing team. I'm just here to teach you a little bit about Goldcast if you've never used it before. Over on the right side of the screen, you'll see a couple different tabs, chat, docs, and q and a. You know, I'll post some messages and chat as we're going, but the two tabs you should focus on mostly is the docs tab that has a bunch of links to further resources over Rubik agent cloud, oh, and other items about, the product and about us. So feel free to browse those on on your own time. And then the q and a tab is what you'll use to interact with us. Really, to get the most out of this session, ask your questions. We'll answer them as we go. So use that early and often. And those are really the main things that I have to say. So I'm gonna hand it over quickly here to Alpika and Demani to start walking through the product and show you a demo. Thanks again for joining us. Alpika. Awesome. Thank you so much, Chase. And good morning, good afternoon, depending on where you're joining us from. Thank you for taking the time to join in join us for this Rubrik workshop session for Rubrik Agent Cloud. Before we move forward, I just want to flash the safe harbor statement. Now Rubrik Agent Cloud is in a beta stage. So since it's not GA yet, please take a few moment to read through the slide. And in just about few seconds, I'll move forward. I think we are good to go. And a quick introduction. My name is Alpika Singh, and I'm a senior technical marketing manager here at Rubrik. I've been with Rubrik for over four years now, and I primarily focus on everything that is customer facing and a lot of sales enablement for our AI platform, SaaS workloads, databases, as well as unstructured data. I'm here with Tamani. Tamani, please introduce yourself. Morning, everyone. I'm Devani. I manage our field engineering team for our AI products here at, Rubrik. We've been here for about seven and a half years, and this is probably the third or fourth, product that I've supported. So very excited to to bring what we have for you today. Awesome. Thank you so much, Damani. And just quick agenda. Damani and I will walk through what the solution looks like, why we are doing this, and, have a discussion around it. We will have a a q and a section as well. And, to top it off, Damani will walk us through the product to show you how the product actually functions and works. We will leave you with a guided lab so you can try the Rubrik Agent Cloud, at your own convenience and, see how it can benefit your organization as well. With that, I will get started. Oh, yeah. Q and a portion. So, while I'm talking, Dimani can answer questions. And while Dimani is talking, I can answer questions. So please, feel free to add in all your questions even if it's specific to you. Dimani talks to our customers' prospects on a day to day basis, so he has a lot of experience, with this. So please, start asking questions if you already have them. With that, let's get started. Now the agent era has arrived. It's it's no longer the time when we were using chat GPT to figure out some hello world prompts and, you know, how to write small, small apps. You know? Now we are talking about agents that are actually, modernizing your workflow. It's opera making your operations go faster. Right? Now, wherever you see, be it TV, be it, radio, your billboards, everywhere, you have an ad for an AI agent. So it's no longer, just a a thought that it would be here, but agents are here. And the industry experts definitely think so and so do I. If you look at Salesforce, they have Agent Force and it's working really well. In fact, they have stopped hiring new developers, and they want to use agents to actually guide them and develop more code. So AI is here to stay, and it's, it's great times to be in. If you look at the different definition of agentic AI or agents, right, it's it differs from company to company. You will see different, definitions. Just to take a step back, and I just want to have a very basic definition for agentic AI or agents. Now agents are essentially LLMs with access to your tools and your dataset. When I talk about tools, it could be your CRM CRM applications like your like the Salesforce or Dynamics or, you know, your, other database application. And it could live in the cloud, different SaaS applications, or on prem. Now 33% of these enterprise apps, that are designed for humans right now will be agentic AI by 2028. And this is the stats these stats are from Gartner. And, at least 15% of the decisions will be made autonomously by these agents. Now the agents are coming. It it's it's how you adopt, how quickly you can adopt. So the basic questions which you should be thinking about is, like, are you ready to deploy, operate, and govern them, so that you can take advantage of this agentic AI era? Now when you look at the agentic maturity, right, to get the maximum ROI out of your agents, there are essentially four stages of maturity. Now the first part is all about the experimentation phase, where you have probably one or two people working on your team part time on, some initial experimentation, but it's primarily in your, sandbox environment, nowhere close to production environment. And they are trying to figure out what kind of agents your organization can build, to make sure that, your, your operational flow, becomes easier and, easy to operate. Right? The next part is the formalization stage, where you have moved moved on from the experiments, and you actually have defined, a couple of agents. Maybe, you know, say ten, twelve, or two dozen agents, that are actively using your, production data as well as tools. So you have expanded the horizon, and you have expanded the capabilities and access of these agents. The third stage is the proliferation stage, where you now you are now scaling from a couple dozen to hundreds of agents. And now you have a platform, or maybe you have multiple platforms to, deploy these agents. And, also, you are able to monitor the actions of these agents and govern them, to make sure that, they are doing what they're supposed to do. Right? And the fourth stage is the autonomous AI, where now you are scaling from hundreds of agents to thousands of fully autonomous agents. Now what are autonomous agents? Essentially, they are not only assessing or, you know, giving you options as to or rather automating your workflow, but they are also making business decisions on your behalf. Now it could be with a human, in the loop, or it could be totally, autonomous as well. And it goes without saying that you already have all the platform, to support this as well. Now we've been talking to a lot of our customers, a lot of our prospects, right, as to see where they are in the agentic AI journey. And, of all the people who took the survey, what we've realized that and it's a happy realization that close to 90% of, those, survey takers are actually, are actually on their way or on their they have already started the agentic AI journey. They are at different stages, but at least they have started, thinking about, how they will be adopting, the agentic AI era. Without a surprise, over 50% of them are still in the experimentation phase, but, you know, they are making good progress. Right? So, and a large chunk of, the folks actually lie in the formalization and the proliferation phase. So they are, way, well advanced into the, agentic maturity phase. Right? Now what the surprising aspect is that the autonomous AI, which is the most mature, phase where you get the maximum ROI for your agents, we didn't find any of the customers or prospects in that stage. Now it's not that they don't have the capability, to move to that stage. You know, they already have platforms in place to deploy the agents. They already have use cases defined. Right? But the problem is, the issue, what they are noticing is the risks that comes with the agentic AI. Right? And let's talk about the risk a little bit. Now, AI agents are superhumans. Like, they are super fast. They they think, they take decisions quickly, and they're super fast. Now when you talk about humans, they are thoughtful, responsible, and they're a little bit slower to not a little bit, a lot slower than agents to do, or to a to to do a task. Right? But a lot of times, humans actually think before they make a decision. Like, there are certain guidelines which, your organizations might have given it to you for a task. Right? I'm sorry. But, in certain scenarios, you have to sit back and think if this is the right approach. Should I talk to someone? Should I change my approach? Like, those things can happen. And if you look at the current infrastructure, your production infrastructure, be it IT or otherwise, they're designed for these humans. And now AI agents are actually assuming the nonhuman identity or the NHI, sorry, and doing the same task. Right? So a, the design becomes an issue where it's it's designed for humans, and that comes with a lot of risks. The first one is the hallucination aspect. If things are going great, you know, your, agents will behave as expected. But it's fairly common for agents to actually hallucinate. And when that happens so sorry. And when that happens, the output that is produced, from these agents are not good. They may not meet your business requirements, and they can cause a lot of harm to your business as well. They it can cause significant, revenue loss as well. The other aspect is, the agents can go off track and go completely off rails and essentially start doing tasks they are not supposed to do. They can take actions when they are not supposed to. Like, they can go ahead and delete a database. They can go ahead and update SSN of a customer even though it was not requested. So a lot of times, these risks just come out of, are there, and they come out of surprise. Right? And then, you may or you you may or may not realize it right right then, and it might be too late before you can take any action. And when we talk about these risks, these are not, just hallucinations. They have actually happened, and they have executed actions as well. Like, if you look at the past couple of years, there have been several instances where your AI coding platform just went rogue and deleted an entire company database, during a code freeze. And, you know, you realized it. You have to go fix it, but the AI agent was not supposed to do that. And and and in other times, there were actually API keys that were exfiltrated, by a rogue agent. Now if you think about cyber attack, you know, it's using rogue people, or rogue identities. It takes time for them to exfiltrate and, do what they want to do. And it and when it comes to agent, it can be done fairly quickly. It can be 10 x faster. Right? So you now also have the data exfiltration risk, that comes with these agents. So when we talk to our IT or security persona, the number one question they have is, like, is there a platform, that can give me a single pane of, glass, access to all my agents, that are running on different platforms, that have and gives me a picture as to what applications they have access to, what they are doing with it, what actions they are taking. Is my data, getting leaked or exfiltrated? Are there rogue agents' environment? Is there a way where I can prevent an AI agent from making an act or taking an action that they're not supposed to? If a an agent have has made a mistake, can I recover from it quickly? So these are some of the common concerns that most of the IT and security teams have now, and that's essentially stopping them from moving from stage two to three to actually the fully autonomous AI stage. And this is where Rubik comes in play, and I'll hand it over to the money for the next steps. That made that. Sounds good. So, yeah. So the the platform that we built in, the Rubrik, agent cloud is designed to help, our customers move quickly from not knowing what agents do and maybe being blocked by, deploying agents in their environment, to giving visibility to what those agents are doing in their environment, what applications they're able to touch, which, identities they're able to use, and unlocking, that capability so that, our customers can confidently move forward with, their AI projects, especially if they're if they've been blocked, from using agents because of of lack of visibility on, on what's happening there. So that's kinda where we see this transformation but from strictly people using, environments to automate to that digital transformation to that, transformation to using AI agents and how we can help in each one of those, stages. Go to the next slide. So yeah. So we've introduced this platform called, Rubrik Agent Cloud. And, the general principle is, because we sit at sort of this intersection between where we see customers' data, we can see the agents' customers are running, and we also see the, customers' identities, we're in a unique position to know, what data agents are accessing, what identities they're accessing, and be able to see that agent workflow and give, monitoring and observability to that. Our platform will also give, an ability to, provide remediation as well as, some preventative guardrails, that we can put on as well, again, because of the visibility that we see. Next slide. So, as I stated, from a monitoring standpoint, we'll we're able to discover all of the agents that, are flowing through your environment. We're able to, intelligently apply, governance policies to those agents so we can see things like, what the risk levels of those agents are, whether or not the agents have been authorized for use, what tools those agents are using. We can provide alerts when we see, things going off the rails. We can also, block, agents from taking actions that they're not authorized to do. And then we also are, providing a remediation, feature where, because we've backed up the data or at least we're informed of what, data the agents have accessed. If we see an agent take a destructive action or, make a bad update, we're able to go back to our backups and recover the data, back from those backups and and rewind, your system back to before the agent, caused an issue. Go ahead and go to the next slide. So the way this all works is by introducing, this AI gateway, that we've created. The AI gateway allows us to see, the the, conversation between agents and their LLMs. So, if you look on the left hand side, traditionally, agents will talk directly to an LLM through an API of some sort. And so we intercept those API calls by having the agents send the data to our gateway first, and then we'll relay that information off to the ALM. But because we've sent it to the gateway, we now get to see that conversation, so the prompts and completions that the, agents are making. And then we can pull some metadata out of that to understand what that agent is doing. For more of the closed system, so things like Copilot, AgentForce, where we don't have the opportunity to see, what the agents are what endpoints the agents are talking to, they do have, APIs that we can use to get the same information. So we'll have, some integrations directly with Copilot and things like Agent Force where, we can get the data directly from them and then, again, infer the same information through the eight way. And then last but not least, around endpoints, where we will look at having an MDM script essentially that will run on an endpoint and tell us what agents are running there. So if you have things like cursor or cloud, if those, agents are using, a endpoint for NLM directly, we can redirect the traffic through our gateway. Otherwise, we can scrape the logs that they're using to get, again, that same level of visibility. This takes us through the three types of agents we've seen in the industry so far and being able to collect that data into the Rubrikade Cloud and provide, that functionality that we're looking for of, the visibility, guardrails, and remediation. Next. Okay. So, let me go ahead and share my screen, and we'll get going with our demo. So alright. So, once you've, logged into the Security Cloud and gone to our new agent Operations Room, we'll come in and show, this dashboard view. So here in the dashboard, we'll be able to see things like the number of agents, that are in the environment that we've discovered. Again, we'll automatically discover agents just based on them flowing data through the gateway. We'll assign agents risk. So, we'll look at, whether or not the agent has access to, destructive tools. So if you have, tools that can read, write, try not read, but write, update, or delete data, then, we'll assign a higher risk to that agent. If you have tools that are just reading data, those would be low risk agents. Sorry, Damani. You might be sharing the wrong screen. is it that you guys see at this point? Slides. Okay. Now we see the right screen. Sorry. I didn't click a button there. Alright. So let me I'll start over from the beginning. So once you've logged into, the Ruby Security Cloud, we'll go into this agent Operations Room. And here you'll have, a list of the agents, that we've discovered. So we'll show you total number of agents discovered as well as, 10 list of what agents are there. Again, we'll assign risk to those agents. So we automatically, based on the agents flowing through, have, registered those agents and then understanding what tools they have access to, we'll assign that risk. The risk is, based on what kind of tool the that that the agent has access to. So if that tool is able to write, delete, or update data, it's a high risk agent. If it's just reading data, it will consider it to be a low risk agent. We'll, give you an idea of the number of tools that we've discovered and then a list of any violations against policies that we've seen. So take us into our policy policy engine where you can, add the initially, out of the box, we'll have some predefined policies, but eventually, we'll give a capability to, write your own policies. So, you'll be able to do things like say, hey. I only wanna have read only tools in my environment. I don't want agents that are, I don't wanna let agents run that are, using, read, write, or destructive tools. Maybe I wanna authorize specific tools or specific agents for use, so I'll, enable just those. Or maybe I only wanna authorize agents to access certain applications. So we'll have rules like that. Go here to the inventory view. Here we'll give you a bit more detail of each agent. You'll be able to search for agents by name. We'll talk about what applications that agent has access to, what platform the agent was created on, maybe who owns the agent. If we click into this agent, we'll give, this detailed map you we've discovered that the agent is doing. So the first thing we'll take a look at is the identities that the agent is using. So if you have identity user identities that are being used through the agent or if the agent isn't able to act autonomously, maybe it's using a service principle, we'll detail out what, what identities that agent is is, being used by. This also will inform us as to what permissions that agent may have access to and give us an idea of, what the capabilities of that agent are, or what it could potentially do. We'll also see what, applications that agent is actively, accessing and then what tools are being used against those applications. So here we have an example of violation of one of our policies based on a tool call that's being used against Microsoft SharePoint. So if you look at this particular tool call, we'll see that it's, trying to delete a record from an accounts table, and that violates our read only, tool rule. And we can see that the it's a delete file tool. So In this case, we haven't chosen to block this action and instead it was allowed to go through. We'll give you a detailed view about what it is the agent was doing. Here, what account was removed and what the actual action was that had taken place. We'll provide a remediation capability. So if I click on remediate, I can create a ticket in your ticketing system. So, the the owner of this agent needs to go back and update it so that it's not making these destructive action tools, we can do that. We'll also send the full details. So if they, there needs to be some more investigation as to what this agent was doing or touching, you can do that through your ticketing system. And then, if this, agent was being backed up or the data was being backed up by Rubrik, we can go ahead and initiate this rewind action. So here in the rewind step, we're able to take you directly to the backup of the system that the agent had affected, just before, the point in time that the agent ran. So whatever backup ran before that agent operation took place will allow you to restore directly from that. So this is pretty much, it. View of specifically all violations that are happening, you can go into a violation, and again, takes you, back to that remediation page that we talked about earlier. We also have our activity log. So this is an immutable, log or log of what the agents have been doing. So a lot of our customers will use this to, with our auditors to say, hey. We know exactly what our agents were doing. We can give you a complete audit trail or even from a debugging or troubleshooting perspective if they wanna see, in the real world what that agent has been doing. So that's all I had from, that perspective or from a demo perspective. Happy to take any q and a, questions you guys might have, live on the call here. It looks like, few things have been happening in the chat as well. Yeah. We've been good on questions. Folks, if you have questions, please do ask them. Alright. So, as we're waiting for this to come in, as Alka or Alka has, posted, so, again, the platform is designed around monitoring, governing, and remediation of agent actions. We are, in early beta at this point, and we do anticipate having a production product, by the January. If you do wanna sign up for our early access program, I think believe there's some details here on the the slide and maybe in the docs as well. Happy to have you participate and, and get feedback and, show you what we've built so far. So I'm just starting to look through the q and a here. I'm not sure if you want me to just read through these or answer them as we go along here. You can start answering them. Like, I'm just Okay. So one question we have is, one learn about agent tools through MCP. It seems that this discovery of tools is being abstracted. It's not shown. Also, eval tracing of what an agent is doing is provided. Is this a Rubik in house eval library? Yeah. So, essentially, I think this question is asking about, whether or not we're able to discover NCP tools. And then, the fact that that's sort of abstracted because we're just looking at the all in traffic. And the other question is sort of around, if we're doing any looking at the traces, that that come through. So, yes, we're looking at both. Currently, the proxy we have is is just looking at, the LN traffic. We do intend to also create a proxy for MCP as well to help give more visibility into the tools. But just through the all in traffic, we are able to see a lot of details about what tools are being used in it for, what those tools are doing. And then, where there's tracing involved, we are using that tracing information to help, also inform us as to what's going on. How we're doing that is through an LMMR backend, that we're using to to look through the the information and, provide those, those decisions. So it is a custom or a fine tuned, model that we're using to to, infer what those agents are doing. Matt Tyra asks, or is there a bulk action for restore of, this is product continuation of another question of all changes made across multiple systems by an agent? So I I think you what you're asking about is around, if we see an agent do multiple touch multiple systems, can we restore all of those systems in a bulk restore step, or do you have to go one by one through each system and do a restore? Today, it is one by one. Although doing a bulk restore is on a road map, but the initial product, it'll be individual restore of any, systems that an agent has touched. First part of your question, Matt, was if there are multiple restores needed as a rewind, do all of those happen as separate operations? So, yes, if if, again, if that agent has talked to, more than one system that needs to be restored, you'll have to go through, each system and restore them independently. But we would show each of the systems that, were affected by that agent rewind or that agent. Salvador asks, how Rubrik agent cloud connects secure protocols to different agent sources. What is needed for the same service account, tokens, etcetera? What other product licenses are needed for this to work as expected? So from a licensing perspective, we'll have, one, license to enable, agent cloud, and it covers everything that I I should. So the rewind capability, the remediation capability, the ability to block calls, discovery, it's all covered under one license. We're anticipating that'll be a a flat fee, initially, and then we'll probably figure out a usage model, later on. The first part of your question as far as the protocol that we use, what we're doing is we're looking at the actual OpenAI or anthropic, AI API calls. So, because we're, at the API layer, we don't have to worry about, like, decrypting SSL packets or anything like that. We actually see the prompt and the the, response or completion come in completion come in. So no service accounts are needed necessarily. When you do create the model router or the gateway, you would put your, model and its keys into the gateway so that they can access OpenAI or Anthropic or whatever, LOM that you're using. And then, the gateway gives you a key or set of keys that you can use with your agents so that they can access the gateway. Alright. Looks like the rest of these were already answered. If there's any other questions you'd like me to take live, I'm happy to do that. That's one question that just came in. How does Rubrik Agent Cloud handle multi agent or orchestration and memory? Systems across sessions. Yeah. So, the the, today, what we're looking at is more we we have a way to identify what who each agent is and then what action that agent took sort of in series or or, as a a string of actions. We don't quite have a a concept of a session yet. That is something else that we're looking at so that we can say, hey. This all of these actions belong to a particular session. Where we're kinda starting out with right now is just identifying who the agents are, what actions they're actually taking, and then if any of those actions, need blocking or not. And eventually, we do wanna get to a point where we can say, hey. Here's a session and kinda link all of the actions that took place within a session as a a single body. So to answer your question, we haven't we're not quite there yet of under understanding, you know, agents talking to agents for one thing and then a session across those multiple agents and all of the tool calls that each one made as a single body of work. So that's next our one of the next things on our list of things to do. Awesome. Thank you. I think those are the only questions we are seeing right now. Thank you, Damani, and we already did the live demo. So I just wanted to thank you all for coming for this session and joining us and, you know, asking great questions. So thank you for taking the time as well. Chase, any final words from you? We do have the the story lane. No. We do have the story lane. Did you wanna walk through that? No. I think we're good. Since Damani did the live one, we are good to, good to skip this one. okay. Well, Oh, okay. Actually, yeah, to give people? a a sense of what it what it'll be. Absolutely. So, what we will leave you with is, our guided demo. Let me quickly open that up. Sorry. I should have been ready to this for this. Okay. So this will be a step by step walk through as to what, you will do if you when you onboard Rubrik Agent Cloud in your environment. So, you know, the first step is obviously onboarding. There were a lot of questions around how do you handle multiple platforms, how do you onboard. So, you do have to log in to your Rubik's Security Cloud instance, and then you can get to, the agent operations, room, and then you can start configuring your gateway. So like, Damani mentioned, there are two methods. Either you can use the model gateway, or, you can also use the native integration option, for Microsoft, Bedrock, AgentCore, etcetera, for connecting your platforms. Right? The, once you onboard it, what you will see next is the set of agents, which which the man in the money already walked through, so I'm gonna skip that. And to use the demo, you can instead of clicking on next, you can actually use your, right or left arrow key as well to go back and forth. Once you're towards the end of the demo, demo of the section, you can go ahead and go to the next section. So, every topic that, Damani covered, which was, you know, right from onboarding to monitoring, how do you govern, how do you actually do the remediation, you can see it action, see it in action, using this guided demo as well. So we are going to leave you with this demo. And, you know, if you have questions, reach out to our sales team, to see a live demo or have a deeper dive conversation with them. I'm all done, Chase. So what's you? Yep. Thanks again, folks. Appreciate you joining. Like, as as we said, the link to the that demo is just now put in the chat by me. It's been living the docs tab as well. And if you haven't clicked over there, there's a couple other links too. But, you know, up above us is that get early access link if you wanted to, you know, check it out in, you know, live. But, you know, the the story lane's there for you to, like, get a a a sense of what it's like right now. And, we appreciate you being here for this webinar, and we'll stick around for, you know, a few more minutes in case any other questions come through. And if, you know, any issues clicking through the the story lane link there. Thanks again. Thank you so much. Oh, I see a question that has come through. Is this including the regular license, or additional licenses required? So, yeah, this is not just Rubik's Security Cloud license. If you have Rubik's Security Cloud, you will need additional license, to get access to Rubik's Agent Cloud. And and, you know, please do click on get early access. Try it out in your environment to see how it can help you. I think we are good to go, Chase. Yeah. No more questions. so. Awesome. Thank you again, folks. Enjoy this the, storyline, and we'll see you again on the next one. Have a great day. you all. Take care. Bye.