Purnima Padmanabhan, Broadcom | VMware Explore 2026

[music] Hi, I'm John Furry, host of the cube. This is the cub's VMware coverage of explore. We have two days of cube coverage from the show floor. I'm here with the broadcast executives to unpack the innovation and news at the show. Here back on the cube for the I think fourth or fifth year, Perneima Padmanaban, the general manager of the Tanzoo division. The world is really talking about platforms. We're here to talk about agentic platforms running agents safely and securely. Great to see you. The world is spun to your doorstep. >> This is your time. The market is hungry for platforms and the AI infrastructure is booming. What's the news to show here? Tell us. >> Well, you of course started with the lead. It is all about AI and I mean you you rightly put it. We've been it's almost as if >> this is the moment that the platforms have been waiting for. Yeah. Right? When you think about what is an AI or an agentic app, it's nothing but a microservices-based app. Right? You can it's a very natural an analogy. And Tanzoo platform, especially the Tanzoo platform for agents is just perfect. It is a simple way for you to rapidly build your agents, deploy them, and give them all the right context. It needs the models, the skills, the tools, the uh data and the services but in a very structured and safe way. >> Talk about the customer situation because the world we saw cloud scale now you have AI native startups, you have AI native applications, everyone wants that context, they want the tokens, they want everything to run fast. Yeah. >> The platform plays a significant role in managing the resources. What's the big change from last year to this year? The big change is it has gone beyond uh experimental stages or line items in boardroom decks and uh if you see last year it was lot of ivory tar organizations that were experimenting with AI solo. How AI is broad-based. >> Every engineer in my organization, every PM in my organization has is AI enabled. Right? So the idea that somebody is not AI enabled is no longer the norm. Right? That is the anti-norm. And so with that what has happened is a proliferation of all kinds of creativity around agents around apps and people are building things very fast. >> But as you know from the past we know whenever you think build things fast you are creating an operational challenge unless and until you bring an easy standardized way to get that speed. >> The agents are super popular. We saw coding come into the enterprise. Gen [clears throat] one was, you know, search and marketing materials, chat bots. Then coding changed the game. We see value. Agents now is another next step function of value. But we're seeing all kinds of dangerous situations. Agents running wild. We saw it. Black had a great postmortem from Open AI around what happened with hugging face. Agents went off the rails. So security and safety because these agents can go code on their own. They're talking to other agents. There's definitely value. People don't debate that the question is how do I run these agents? So where does this fit into Tanzoo because this is a software opportunity but also people don't want to let agents run around their infrastructure. Exactly. They're touching resources that they shouldn't be touching. >> Exactly. And this is where the difference also from microservices comes in. When you tell an app to do something that's what the app does. But agents have by definition agency which means you just give the intent and resources and then agent interprets that intent and decides to do something and that is where the whole open AI incident is an example. The agent didn't do anything wrong per se but it was not the intended consequence. And so you have to look at the problem then in a different way. So you have to say okay I want to be able to run build agents fast. I want to be able to build that securely and I want to run them but I want to run them in a sandboxed way. So what we are saying is imagine if you could run an agent such that agent has access to nothing to start with. That is what we the idea of a deny by default agentic runtime and then you give agents only those resources that you think the agent should have access to. So you don't give it random access to network for it to go and break out somewhere. So you give it access to the model that it can use, to the skills it can use, the tools it can use, the uh data and context it can use and I I've got some interesting news there as well as the services it can use. So that is kind of the big announcement here which is the Tanzoo platforms with agentic foundation is now integral part of VMware private AI cloud. It is the agentic orchestration layer which allows you to rapidly build agents, not have to worry about security, push those agents and then we automatically secure it giving it only access to the resources that are permitted. >> So that's a runtime. >> Yes. >> For agents. So I'm a customer. >> Yes. >> I want security. I want safety. I'm I now have that with the runtime on the private AI cloud solution. >> Correct. Stanzu platform integrated into private AI cloud solution not only gives you the access to infrastructure the models but also a runtime where you can build your agents run your agents observe your agents manage your agents very fast >> okay so we know the foundation models they're I guess they're closed people say closed they're kind of open but they're closed >> um the the discussion around open weights and open models open source has been thriving as you know spring you have a lot of experience running that >> open source will be the innovation area but right now there's concern around the security. >> Yeah, >> we saw, like I said, we saw the hugging face impact from the open AI agents going off the rails and going rogue, >> whatever you want to say. How are you guys thinking about this? Because this becomes a real um I think a challenge, but also, you know, no one wants to jump in unless they know it's trusted. This is a big um issue because you got innovation [clears throat] on one hand and you're scared on the other. People don't want to be scared. They want to be confident. >> Yeah, this is a perfect segue. the second for the second announcement which is we are Broadcom is investing heavily and I when I talk about billions of mythos tokens right to make sure that we scan and secure open source for the enterprise we have a very strong open-source dis discipline with our enterprise spring capabilities uh we are the primary sole committers to the spring open source and we we take that job very seriously So in addition to Spring, what we are doing is we are expanding the support for additional libraries in Python, NodeJS and the rest of the Java ecosystem. And add to that the capability that we already have with our Bitnami secure images and our data services. You have a whole set of trusted open-source libraries and images that you can start using. have your agent build code with have your artifact repository um automatically mirror it so that you can be off to the races without worries. >> We hear a lot about mythos. I mean, everyone's talking about it because it's so powerful as identifying vulnerabilities everywhere. What's been your experience and how do the VMware customers, Broadcom customers leverage that? What's the update there? And you mentioned you're investing billions of tokens. Yes. >> What how does Mythos play into this and how does that impact the customer? I'm glad you asked that. Look, we have had Mythos since the beginning. Um, original glass-wing launch if you may. And there are two ways we use mythos. One is we are leveraging mythos to scan our own commercial repositories. And really, it's not just blindly scanning is there's a lot of skills that we have built around it. How to red team and how to blue team against the software. And we are finding issues and we are patching it. And so we're putting out we are aggressively patching our commercial software so that our customers are secure and we're proactively doing that but we are also doing the same thing for open source and that many people may not know. So we are as I said um the the the really focused on spring and the Java ecosystem below this below spring and we have been scanning all of that with mythos. Now let me be clear if there was any doubt we are finding issues we are finding vulnerabilities and these vulnerabilities are not uh just your low-level vulnerabilities these are things that can completely bring down an enterprise and so we take that job very seriously and we've been investing billions in scanning that software but also patching and so in June we did uh uh the patch release in spring which was the largest patch batch release we have done in our 23 year history as a steward of spring and we have also been working with a lot of banks and a lot of government institutions and they came back to us and said beef up your enterprise spring more so what we have added with enterprise spring now is a program where for critical infrastructure companies they get early access to the patches so that they can be patched and ready even before the vulnerabilities get announced to the open. We have added complete support for clean secure builds for everything not only in the spring but also all the Java dependencies of spring. We are automat we we're doing day zero simultaneous patches not just for open source spring but for all the out of support versions the long tale of out of support versions we are providing patches for that and so and we are also best part doing CVE only patches so that developers have no excuse they simply take the patch and they are secure so lot of investment in making sure that we have a spring solution for the enterprise that takes a lot of the worries out of their mind >> so trusted open source is really kind you're you're doing and the trust on the commercial code >> there is trust on the commercial code and there is trust on the open source and what we have done now is we started with spring which we are very strong on with Java and we have done the enterprise spring and now we are expanding that same discipline >> that same trust and be providing libraries clean secure libraries for Python NodeJS and other Java artifacts so the idea idea is if you come to brat Broadcom you get a set of trusted sources for your open- source stack and of course commercial software we are already >> the customers want that I was seeing a lot of agent building I want to come back to the the main announcement um there's huge demand for agents I mean everyone's doing agents you were talking to me before you came on camera that you're running your fleet of engineering on agents you guys are agentified your your operation but customers want to go faster um what specific speifically is going on with that platform and take us through the runtime. >> What's the what how is it set up? What do they get out of the box? It's not really a box anymore, but you know, initially um how does it work? Does it integrate with their data sources? How does the data play? Because the data is the key ingredient. >> I'm glad you asked that. >> So, take us through how it works. >> So, look, first I said the first part is this agent platform which is Tanzu platform is integrated into VM private VMware private AI. The platform is installed as far as a user is concerned. Once the platform engineer has configured it, all they need to do is use one of the built-in harnesses that we have. We have a Python based harness as well as a Spring AI based harness and simply give their intent. They've got an agent. They push the agent connected to models, connected to services and they're off to the races. Now, one thing more that an agent needs is context, right? And till now, usually it's a separate team that is curating the data. Then you have to go and negotiate what access how do you give access a lot of stuff going on >> how do you figure out the identity how do you figure out our back what should you give the agent what should you not give the agent and so on and so now we have baked as part of this announcement baked into the platform what we call AI ready data foundations and what it allows you to do is simply point your data sources curate the data you need very rapidly so speed is important you don't wait for a large warehouse project to happen. You simply bring what you need. >> We take care of both structured and unstructured data. Unstructured data is important, right? That's where agents have >> magic, lot of context, right? Like contracts and things like that. So, you bring all of that and we have built in a data pipeline that automatically does chunking for you, does vectorization for you, pulls the context and normalizes the data for you. And then you can curate that data set as a data product that an agent can connect to. It's arbackbacked. You only provide the data that it needs. You don't have to give it access to all the original data sources and you can move fast. So the value really is about unlocking innovation fast. Doing it low cost. You don't have to transport the data to a cloud to start getting intelligence out of it. It is right there where data is sitting. Right? So it is it is cost-ffective and it is secure because it's governed and controlled. >> It's like gas in the car. you get the right formulas. It's interesting. Agents that have governance, security, and data prepped, >> yes. >> And nailed down always seem to work better. What's the importance of all that together? Because you're protecting against essentially rogue access on one hand. At the same time, you're almost giving the brain to the agent. >> Exactly. Is the brain. >> Exactly. And this is an interesting one, right? As I said, the right way to secure an agent is to put it in a black box and give it nothing. But then you won't get any intelligence also. So then you start saying okay that's how I've secured an agent but then I need to give it things that give it intelligence but I want to give it on my terms in a curated uh way with identity with arbback with credential management and so that is where we connect and connection to models and connection to tools we had already talked about it but now being able to connect it to curated data sets all on a single platform means I can move faster. It's all about speed because at this point like you said earlier, you know, we've got the world is changing in front of us, right? And it is all about unlocking that intelligence >> and if I don't give the agent the things it needs, you don't unlock it. >> Well, it's a double-edged sword. The scale involved we're talking about with all the infrastructure advancements, agents can scale in a good way >> or in a bad way. Yes. Pick your path. Yes. And that's what you're kind of getting at here. Okay. So, now the next question on this is customers. Yeah. Yeah, >> the number one thing we hear on the cube in this past, I'd say year, maybe 18 months has been >> things are stuck in pilots. I got to get stuff into production. >> Okay, I'll do some deterministic workloads. Those are easy to throw agents at because it's well known end to end, but they want to expand the aperture of use cases. >> Yes. >> And they want to get them into production. Take me through your thoughts on how you see that playing out based on the new runtime and the trusted source. >> I I I I think that's an excellent question and it is not unknown to anybody. The most at scale use case that has been unlocked is coding and we see that >> continue to be even in customers where coding is not their primary job. If I'm a banking agent coding too >> agents are coding too like >> that's what I'm saying. Yeah. So agents for coding what I'm saying is the number one use case is to use agents for coding. That is and that is across industries right? Any industry that I talk to if they are building a custom application they are using agents to build that right? So co coding agents is something that is unlocked and as I was telling you earlier my entire I'm in the business of building software and my entire software life cycle is today running on a network of agents a swarm of agents that actually runs on my own platform that goes all the way from agents for requirement building agents for validation agents for building code which is only a small part but also agents for testing and documentation. So that entire life cycle is automated and look the the benefits are in incredible two to 3x improvement in productivity 70% reduction in find time to find issues right so the the the the type of outcomes that I get are amazing so first I see is coding coding is the use case everyone is doing now the second use case beyond that initial chat use case like you see is workflow automation so people are taking portions of the workflow that are business specific and saying can I start making it more relevant and agentic if I'm an insurance company and I want claims processing being able to have an agent assist a claims agent for example a real human agent being assisted by an AI agent but the agent can quickly correlate across sources using the data products I saw I told you about it can quickly come up with some inferences and help that claims agent so this is a very curated business process for that industry And we are starting to see a lot more applications of that. >> Yeah. And and that's a trusted environment too when you got the the workloads that are that are like the claims example. Let let's talk about the ecosystem because now agents are crossing boundaries. >> Yeah. >> APIs were great in the cloud era because you APIs are great and the security there. We talked to Amish about that here on the cube. Um but when you start going across the ecosystem, there's trust. >> You have to be trusted. How do you view that connecting the connective tissue between ecosystem partners or MCP servers? This becomes again another kind of microservices-l like thing. >> Exactly. And and again uh the problem is not very um different from what we tried to solve in the past. So I I do think the MCP approach that anthropic put is pretty pretty interesting because what it is is like MCP is the new API, right? And every service is like by the way when you build an a service or a build application on the platform you can automatically expose an MCP for it in that is another feature of the platform and so in a way you're exposing a curated API right that an agent can use or that a model can use. So we see MCPS really evolving as that curated API and data products applications all are all publishing those MCPS and they become the way in which you talk to each other but as you talk through to each other you have to maintain identity so you need a platform that traverses through the identity you need to maintain the security you need to maintain the credentials so again you need that orchestration layer and that is why the relevance of PR platform becomes again Yeah, I mean all the all the music that we were hearing in cloud native orchestration, telemetry, observability, control planes, all kind of playing into AI. I'm really fascinated by this AI ready foundation. Um, zooming out, what problem does it actually solve? So, if you don't have an AI ready foundation, what crops up? What are some of the things people would see and what specific problems do you see? So as as we said um today on Tanzoo platform you simply can have an AI foundation and run agents and I think you're talking about the AI ready data uh AI ready data is really about context right and so let us say the AI ready data foundation wasn't there what would you have to do right that is what what what's the problem I'm solving right >> so you're running agents now you have to give agents the context so you could say hey agent has access to my G drive agent has access to my shareepoint. It has access to my uh S3 drive with open table format f files. Do you really want to do that? Do you want agents to have direct access to all these sources or do you want to be able to curate specifically things from and and and information from those sources and then provide it to an agent? I think that is what they are. The main problem is security risk. >> Security risk and tampering. >> Security risk and tampering. Making sure there is governance and arbback. Right. That is also part of security. Governance >> and then ease of use. Right. >> Mhm. >> Do you want and and and do you want an agent to go and traverse through all those resources to find an answer or do you want to have something pre-curated? And by the way, something that uh I was talking to my team earlier, this also reduces the amount of tokens you burn, right? If you have done some curation beforehand, you're not asking the agent to sift through everything. >> So efficiency. >> Efficiency. >> High efficiency. >> High efficiency. Exactly. >> Okay. So when you're learning, so your fleet of agents are swarming around coding away, doing the work. Um a lot of CIOS and CEOs on the business side are saying, "Hey, we see revenue. They want to move faster." What's the main message to that audience? Because it's clear the mandate's been let's put intelligence into our business. Yes. Let's create a business brain and make it a strategic moat for our competitive advantage. Business models are changing because of it. Okay. Now it gets handed to the technical team. What's the message? Because that's the top focus we're seeing. We've seen the tech teams and the platform teams rise up to solve the problem which is move faster to have that competitive edge. >> What's the message to the CIO? I think the message to the CIO is uh this is the chance to shine, right? Because this is a revenue opportunity for business. I think not only is it a revenue opportunity, it's also a competitive advantage. And if you don't do it, it's a competitive disadvantage because everybody else is going to be doing it. So there is no choice to for me was there a choice not to do agents for my coding? I don't think so because somebody else would be accelerating their coding processes and innovating on their features much faster than me if I didn't do that. So for me it's a business imperative and what I what I would say just even if I look at Broadcom our CIO Allan has just risen to the challenge right rapidly providing um access to models figuring out how to sandbox it figuring out how to standardize things figuring out how to secure and interestingly I partner with Alan >> also on providing trusted sources for open source also on providing a platform where agents can run so lean on I would say the ecosystem system to get things in place fast. You don't have to invent everything. Get >> get going. That's we have >> and get going on the business imperative. So, is it more important for you to innovate on things that are not relevant to your business? Just take that off your off your table, off your plate and innovate on your business imperatives. >> We got two great news announcements and and following Tanzoo and Spring. We love the open source. We love the trusted sources. The runtime is super compelling. You mentioned sandboxing. Chris Wolf teased this out and put some context around it. Explain how's that work because this seems like a template for execution. Yes. When wants to start loading in the agent fleets, people are putting their toe in the water. Some people are jumping in the water full full tilt. >> What is the playbook? How does this sandboxing work? Is it how important is it? Is it a progression? Take us through Yeah. >> the sandboxing of these agents. >> Sure. The and sandbox is at multiple levels of the infrastructure. So at the core infrastructure level having that container sandboxed into a VM that agent container that is one part of the sandbox. The second part of the sandbox is when any container is instantiated with an agent we have this concept of deny by default and that is a sandbox. So basically an agent when it is instantiated doesn't have access to anything. It is completely sandboxed and then you start providing it the tools and the intelligence with a very credentialed process which is called the bind process and the bind process gives it access only to the things that you bind to. >> So it's like a prep area. It's like getting ready for the agent to kind of launch. It's almost like a launching staging ground for the agent. >> It it is actually a runtime. It's a runtime with all the it it is completely a sandboxed runtime with access only to the things it needs and so that when you are running the agent the agent can't suddenly say oh I actually decided I need something else but if it tries to access it won't have access so you are preventing these unintended consequences that come from an openly running agent or just throwing an agent on a Kubernetes infrastructure >> pima congratulations on some great news here explore Um, I want to ask you a personal question to to close out. We've had many conversations. We've been covering the CNCF since the beginning of the foundation in KubeCon. Actually, everyone knows the cube is there. >> The cloud native world from a personal standpoint. You've been really working hard. Tanzoo. The timing is great. >> The way the world's going, we're seeing a lot of cloudnative foundation that's powering a lot of the AI innovation. Um, it's kind of the same game but different because it's got a different twist. Where's your head at on this? because this is a win for the cloud scale and open-source game. Explain your thoughts on you from a personal perspective been driving this change and building out and operating cloudnative >> systems. Now we're got the AI on top almost. It's uh like you almost opened it right John which is we were waiting almost this is like you're waiting for this opportunity built a lot of >> um core fundamental principles how to create standardized environments how to create a a locked box you know a locked approach to execution how do you create observability how do you create telemetry so we've solved all the hard problems >> that are needed to run an agentic or an AI app it's almost like the last this five years were a practice run to get here and then I'm just excited about the opportunities. >> It really is a nice fit. It's a really good fit and remember the conversations we had just go back say eight years ago. Um we got to meet the developers where they are in the CI/CD pipeline and then shift left came. Yes. >> Then the you know Kubernetes rose up and all these orchestration capabilities. >> Agents are kind of the same thing. Exactly. It's it's it's the CI/CD pip. They're coding as you mentioned is the biggest demand. Yes. So a lot of the same princip first principles of cloud are now moving into AI but it's different. Uh in short how would you explain the difference between the AI native world we're in from the cloud world because a lot of people are migrating into AI from the cloudnative ecosystem I mean they're synergistic some are more infrastructure but now you have cloud n I mean AI native developers who are managing fleets of agents. >> I see them as very different. So for example I'm building software using agents but the agents itself are building cloudnative software right so in this particular case they are building cloudnative software and I might have agents also within my software right so the the nice thing is it's not a a a break here >> right the principles of building clean modules building modules that do the functionality providing clean interfaces being able to talk about it Being able to scale independently, being able to use resources efficiently, being able to co coll co coll co coll co coll co coll co-locate things all those concepts that we are we have right evolved and standardized perfectly apply to agents >> and private AI cloud is private cloud with AI basically that's the big news here >> that is the big news here which is we have t rather than talk about piece parts we have taken all the components from broadcom >> our core VMware cloud foundation with AI factory to serve models we have tied it with our security and networking capab capabilities and now with the Tanzoo platform for agents that all comes together under the VMware private >> and you can manage the constraints on the supply chain as the AI infrastructure continues to build that's more capability on the performance side >> that is correct we can we can uh once this AI factory is there we can manage how much resources it's using how do additional resources get incorporated it becomes much easier for you >> thanks for coming on the cube really appreciate congratulations on the news thanks for coming and sharing >> thank you >> I'm John F with the cube Of course, we have our two days of live coverage coming up. Check it out at the cube.net. Thanks for watching.