Bringing Coordinated AI to the Mainframe

Hi, everyone. Welcome back here to Techstrong TV. I'm going to introduce my two guests for you here, and this is kind of a part two of an ongoing discussion.

But if you didn't catch part one, it's okay. It's going to stand on its own as its own segment. But I do encourage you to go to Techstrong TV or the Techstrong TV YouTube channel or wherever you want, whatever screen and app you like to watch videos on.

We probably have an outlet there for Techstrong. And do check out part one. But let me introduce you to my two guests here today.

First of all, I want to introduce you to my friend, Priya Dhody. Priya is the VP Solutions Marketing for BMC AMI Solutions. Priya, great to see you again.

I hope all is well. Yes, indeed. Nice to see you, Alan.

Thank you. And then joining Priya and I is Matt Whitburn. Matt is VP of Product Management and Design for the BMC AMI portfolio.

Matt, good to see you, and thanks for joining. Thanks, Alan. Thanks for having us back again.

My pleasure. So before we jump in, I feel obligated to say, hey, for people who don't know what the AMI portfolio, what AMI Solutions, what's AMI stand for, guys, so that we get that out of the way? Oh, okay.

Well, actually, we pronounce it AMI, and it stands for Automated Mainframe Intelligence. So we've always believed in the automation, the intelligence. We were a little bit ahead of that game.

Well, now here we are with AI, but yeah, that's what it stands for. Absolutely. Okay, so I'll remember that, AMI.

So in part one, Priya and Matt, we discussed a lot of things. We actually got through quite a bit. But in one of the things that we discussed, and I just want to make sure I get this right here, is that we were moving from isolated intelligence, right?

So using sort of generative AI, if you will, in very isolated, very kind of siloed use cases to coordinated intelligence. And in my mind, that's more of using agentics, actually do things across workflows and so forth. So look, the whole world's going through this, not just in the mainframe world, but we're all living through this.

But I struggle sometimes, and I ask guests on here, can you give me real-world scenarios? If you can name names, great, but even if you can't name names. Right?

How is this working inside real enterprises, real business? Yeah, the simple way to think about it is if you're a developer and you have a question about some of your code, you might want to use an explanation feature in BMC AMI Assistant, which includes GenAI, and that's great. But that will help you document the code, understand what the background is, all of that.

But let's say you're not the only person working on that code. Let's say that there are numerous people, testers, developers, some offshore, some onshore, some at consultancies, some other places, and you all need to coordinate on how to bring that change into production. So it becomes, at that point, it's about sharing the knowledge across these different domains and having a consistent knowledge base.

And the consistency comes from having what they call the RAG, the sort of institutional knowledge that's relevant to the company, but also all of the shared intelligence that comes from the processes in order to get things done. So the big difference is it goes from one person doing one task to a shared team of people, and they have to coordinate, but also in an enterprise environment, they have to make it something that can be repeatable and auditable. So they need to have controls and governance in place to manage all of that.

Matt, you're the product guy. Where does that live? The interesting thing is, I think the answer is it's going to live everywhere.

Because when you look at not just our product portfolio, but a lot of what our customers are using, there's a rich investment that they've made today into a variety of different tools, traditional automation, if I think about operations, like in their monitoring environment and things like that as well. A lot of the promise for me is the ability, which I think Priya was talking about, is breaking down some of those silos so that you can really think about the bigger picture and how you can bring together some of these technologies to drive better outcomes. So really when we think about, I think the way that AI is going to be used and the use of agents and in agentic sort of workflows, a lot of it's, I think it'll start with augmenting some of the things that you're doing today to do things in a smarter and more efficient kind of model.

But ultimately, you're going to get to a point where I think there's just different things you're going to be able to harness that you weren't able to do previously. One example I'd give in the operation space, in the mainframe world, is that there's this role people have had for a number of generations called capacity planners. They spend a lot of time looking at how much capacity do they need in their systems, how do they plan for it, how they make that available across the different partitions that they're managing, for example.

You can solve a lot of problems by adding more capacity to some of your environment. You can help improve your SLAs. But you start getting into questions to do with, what's the cost of doing that?

What are the trade-offs that I'm making? And so that ability to sort of blend some of those roles and some of those organizational silos that used to exist in the past, that gives a tremendous amount of potential for me, and for customers really to try and figure out, and us help them with, what are some of those better outcomes they really want to get to by starting to think about things in an agentic workflow? Yeah.

Priya, so as Matt said, you said, I said, we're all experimenting today. And we're trying to move beyond the silos, teamwide kind of stuff. But it's hard to get your head around this because we could talk about coders, we could talk about DevOps people, we could talk about platform engineers, we could talk about SREs, we could talk about security people, data people.

It's one thing to say it lives everywhere, as Matt says. Yeah, I think it's everywhere because we're all going to use it. But how do we coordinate?

I don't know. To me, it just seems like you need a central processing unit or something someplace. Yeah.

100%. The interchange, I don't know what you call it, but- Indeed ... what do you call it and where does it live?

What is it? Well, Matt, I think it's really about orchestration, right? And having that central plane for governance and some of the BMC AMY Assistant and agentic framework you're building out, right?

It is, for sure. The interesting thing when it comes to orchestration, think about it like today with just people have to orchestrate every day, in terms of figuring out what's the responsibilities that they have between their roles, the tasks, what are they responsible for, what are they allowed to do, what are they not allowed to do? All of these things become even more important when you start going into an AI and an agentic world, because the governance of how you control that, because you still need humans.

We still see humans being in the loop for quite some time, basically, in terms of having the oversight for managing these systems and making sure that the actions that are being taken are things that are allowed to take place. So the governance is really key, but orchestration of those activities is really fundamental. The interesting thing I think we're probably going to see is it's probably going to be a bit of a federated model, like in some cases.

And again, if you think about this from just how people would manage their systems today, there's different things that different teams are responsible for, different things they have the authority with. They have people in their leadership organization who will help coordinate those activities together, for their responsibilities, how those SLAs are managed. I think when you consider this in an agentic world, it's probably not going to be that different.

There'll be things that the operations agents, for example, will be able to do on the mainframe. They'll be orchestrated together in terms of their responsibilities, which is where a lot of our tooling can ultimately help you with that. But it's then going to be in the context of, well, what's my enterprise policies?

What's my governance over the data center as a whole? So making sure those kind of things fit together. So that's where orchestration is going to be key, but it's, for me, probably understanding how those things fit together right the way across the enterprise in likely a federated way.

Fair enough. Priya, if it's okay, I'd like to come back to you. There's a growing conversation.

It stopped. Paul, can you scroll up, man? Yeah.

Like this way? Yeah. Okay.

Yeah. Come back. Give me two more questions.

Keep going. Good. Okay.

Count me in, and then you'll edit right in from there. Three, two. Priya, if it's okay, I'd like to come back to you.

There's a growing conversation around standardization of communication models between AI systems and people, AI systems and other AI systems, and almost more importantly, AI systems and enterprise platforms. And look, to me, this is AI getting normal. This is the laws of how we've always done things.

You need to standardize communications. Why does it matter so much right now for organizations who are really in the midst of trying to modernize in place rather than replace... Let me say that again.

Modernize in place rather than replace critical systems? Yeah. Listen, so anytime you introduce change, you introduce risk, and risk is the one thing you do not want in a mainframe environment, because these are typically running core systems, key platforms, mission-critical, revenue-producing transactions, data, et cetera.

So what that means is when you think about how to modernize in place, you have to understand that every customer, every install is unique. There isn't just one stack that customers use, and they have to integrate with all of the different piece parts, whether it's the mainframe stack, the hybrid stack, the emerging AI stack, as it were, across clouds. In terms of standards, though, Alan, which I think is where your question was getting, the way that BMC is thinking about it, and I think we're aligned with the rest of the industry, is we're using some of these emerging interaction protocols like MCPs, like A2A, as the standards that we're working towards.

And you can think about them in a very simplistic way. They're not unlike APIs were 10 years ago. They're ways to communicate between systems that exchange data in a structured way, and they feed the agents that Matt and his team are building.

So these are agents that we're putting in front of every single subsystem you can imagine in the mainframe platform, be it the database, the CICS servers, the security data. All of these things we're building out agents for. So the way that the data will communicate, or the way these agents will communicate, they will have this agent kind of governance layer or agent server layer that will ensure that all of the key policies, roles-based policies, auditability, all of these things happen.

But at its core, integration is 100% required because if we don't have that, then what we're building is just a second stack next to the one that already exists, and I don't think that helps anybody. Right? Yeah.

We can't just replace, we have to kind of build on top of. And I think building on what Priya is saying as well, what we're going to see, we see this today, especially like in the mainframe space, is that people want a mix-and-match model with different vendors helping solve different problems as part of their mainframe environment. So, having that ability to work using standards, especially like in A2I or whatever the next generation of standards will be down the line, is going to be absolutely key because the last thing that our clients want is that inflexibility or even worse, that thought of vendor lock-in.

So, us having that ability of saying, "Look, we can use the standards. " So if you need to integrate with your ticket management system, as an example, it's a good model where whether it's the way that you're managing your code or your operations, that's something that we don't provide for ourself. There's a lot of good choices out there in the market.

Our agents and our technology is going to want the ability to communicate with that, so you can build that bigger picture together. So the standards are going to be really, really important and it's just going to build a lot of confidence with people as our clients start rolling out this technology. I agree.

I've learned more about API since we started adopting agents here than I ever knew. And one of the biggest lessons I learned is not all APIs are created equal . Some of them really stink.

Yeah. They're so limited. Yeah.

You don't find out till you connect to them that you really can't do what you need to do. And that's where, if the APIs were great, we wouldn't need A2I or MPC or these kinds of standards. I'm hoping that these are integrations, these are technologies built for AI natively, so we'll be able to go deeper faster, right?

Yeah. They are. They are, right, because Alan, the difference is an API is a data-sharing vehicle, but an MCP server is actually going to take an action on your behalf in another product.

So in the DevOps environment, you could begin a piece of code in, let's say, Claude or something like that, and then commit it into the mainframe environment by connecting into AMi DevX. That's just one example of AMi DevX code pipeline. So I think that's the big difference is that it, to your point, it is natively built for this environment, for the agents.

It's deeper and wider. Yeah. Yeah.

And in the language that these are all written in, right? Right. Yeah.

And I'm glad you brought it up, Alan, because it is something that we have kept a very close eye on as we've gone about MCP enabling, like a lot of our product portfolio. " But it's given us chance really to say, "Well, are they the most consumable? " And then also as well looking at the way that we actually bring them to market, like for us, is through something like our MCP gateway that we deliver in our AMi platform.

" And it's like, well, you might be participating, but that doesn't mean you're doing it in a good way. So us making sure we're being diligent on the way that we're doing that, and then providing it through our MCP gateway, again, it goes back to having the governance and the control and making sure it's fit for enterprise IT. Yep.

So guys, you both mentioned, and more than mentioned, emphasized governance, explainability, human oversight. I think for good reason. We're all a little bit nervous.

We're all a little bit kind of doubting Thomases, in terms of how this works. But where does rubber meet the road, so to speak, on that? And I've been in security 25, 30 years.

There's the constant tension between risk management versus let the business go as fast as it can, right? And so this is not new. Similar kind of thing.

How do we balance, especially when it comes to mainframes? As Priya, you said, this is mission critical. We can't afford downtime.

We can't afford mistakes. We can't afford security issues. How do organizations go about balancing this?

Well, the term human in the loop comes to mind, and that is core to the way we're designing these systems. But even within human in the loop, there's a sense like, oh, trust but verify. I don't think it's that.

I think it's actually test and verify. So if you think about ops, for example, you're going to want to look at your monitors, understand the kinds of predictions that they're making using classic AI, ML, and you might start to think about, hey, look at some of these recommended actions that my GenAI, agentic AI is recommending to me. Let me see if those are correct.

So we're in the test and verify phase right now, particularly for operations, which is so, if you think about mission critical, what's their number one KPI? MTTR, uptime, SLA uptime. They cannot afford to fail.

So they're a step before trust but verify or human in the loop, where the human just kind of, hey, I took a look and looked good, check. They're really testing it to gain confidence before they can really move to that next step. Fair.

Matt, I'm going to throw one over at you here. So even outside of the mainframe, in general, organizations are struggling with fragmented workflows. We've got dev, as I mentioned before.

You got dev, you got ops, you got DevOps, you got database administration, security, platform. Can AI agents help bridge these silos? Right?

Because another thing I've learned is I founded a bunch of three, four different companies. If you don't give people a place where they can succeed, don't blame them for not succeeding, right? It's on you.

Are we giving these AI agents an impossible task? I wouldn't say it's an impossible task. It's going to be a progression that people will go on over time.

Kind of in the way that Priya was suggesting about the way people are going to get started, how they're going to build their confidence, and then ultimately let them do more and more powerful things. But that idea of actually breaking down some of the fragmented workflows is one of the things I think we're most excited about. If you look even, I would say, in the mainframe space, over the last 10 years even as an example, you used to go to talk to a lot of clients and you'd ask, "Who's your CICS prog?

" You have distinct roles for doing that. A lot of those roles have started sort of blurring together over time, and people needed to start understanding more and more across domains. A lot of the potential, I think, that we see here is just giving people a lot more of the intelligence and the power to actually understand what's happening across those different domains they might not have had previously.

So to give you a real example of something that's going to be coming very soon to our portfolio, we have very rich set of tools today for database administrators, so they can help manage DB2 and IMS on the mainframe. And typically, the DBA roles have been very separate from people doing operations, like right now. But what we've got through the ability of bringing together some of these technologies and to bring it forward in an intuitive way, is that in some of those data tooling technologies that people use right now, we're going to be able to start using the MCP capabilities that sit in our operations product to bring in sort of real-time information about actually what's happening with those production systems, actually, as they're running right now.

So enriching a lot of the information that was available, but not at their fingertips, really, when they're actually doing some of their tasks about managing their databases. So that's kind of just a simple example, but you can also think of it from the context of, hey, the people who are writing code, maybe they need to know more about the operational characteristics of their dev test systems, as an example. Goes back to the, I wonder how much capacity is allocated to it.

So some of those sorts of things, it just lets people start thinking about the art of the possible of what they can do by bringing these things together. And as you start doing that, then you start getting into a sense of like, well, hey, maybe I'd be better off actually having an agent running in the background, managing and thinking about some of these things for me. So some of those fragmented workflows and how we break them down is actually one of the things that gets me most excited about how agentic AI can be used in the mainframe.

I love it. Guys, we only have 15 minutes. We're probably at 20, 25 already.

I apologize, but we're going to have to do a part three.