Acting with Confidence AI as an Agent of Change – BMC Software

Hello and welcome. So glad you've joined us for the third part in our series around adopting ai using AI in the mainframe environment. My name's Mitch Ashley and I lead the software lifecycle engineering practice with the Futurum Group.

Now we're gonna be talking about acting with confidence. We're using AI as a change agent or agent of change, if you will. And I have a great guest with me talking today about this topic, Anthony Dessaro.

Anthony is Senior director of architecture for AI with BMC software, and our series is, uh, presented or sponsored by BMC software. Thank them very much for putting this together, Anthony. Welcome, Mitch.

Thanks for having me. It's been great. The f the first two segments we've gotten to do together has been really fantastic.

So I'd really love to hear more about kind of AI agents or agentic ai and how this differs from the traditional ai, generative AI type solutions we've been talking about. Absolutely, and this is my favorite subject of late, I've been spending a lot of time in the space around AI agents and agentic AI basically comes down to autonomy with, with ai. So in our first two segments, we talked about AI as an advisor and AI as a partner in the second segment.

Now great conversation around that, but there's one fundamental challenge with both of those perspectives on ai. It's a reactive model, Mitch, right? The AI is waiting for you and I to go interact with it, whether that's a chat experience or even it's infused at a product experience.

It's sitting there waiting for us to go and interact with it. This is wonderful technology. This is great technology.

We don't want it sitting idle 24 by seven waiting for us to approach it. We want this technology to work for us 24 by seven. So this is what I like to talk to our customers about, is this is a shift from reactive AI to proactive AI with AI agents where we can have agents that are running, again, 24 by seven, working on a task, working on achieving a goal.

These agents would observe the data in their environment, they reason over this data to make decisions on what they should be doing, and then this is what sets it apart. Now, the agents or the AI acts upon that decision that it makes, it takes an action. And the best part of that is after that sequence, the agent can learn from that experience and get better at its craft or at its work that it's, it's trying to do.

So agents would, could exist in, you know, different organizations and over time those agents would learn and grow to that particular environment. This is a game changer when it comes to the AI space. Now, what does that, what does that look like in a mainframe environment?

Could you have some examples of things you might have an agent, tasks that might perform, things that it might do for you? Oh, yeah, absolutely. So we, you know, the previous segments we talked a lot about, let's go back to the cobol, uh, example that we had.

The COBAL code does, you know, code explain for example, uh, or explain a given situation in your ops space. Well, let's go back and look at the COBOL example. Instead of just explaining the code, maybe now the, you know, the AI agents can get to a point where they could do a code review, they could make recommendations on how to improve the code.

The AI agent could actually fix the code itself, right? So this is where the agents of change come in, where agent, uh, uh, agents of action come in. So it's no longer just giving you information.

It can go out there, look at the code, improve the code for you, do a pull request, and get it back in, in, into the system all on its own without potentially human involvement in that. But we are nowhere near that part of total autonomy with AI agents. Our approach and the approach that we see a lot going on in industry today, Mitch, is keeping that human in the loop.

Why we wanna keep that human in the loop. We want to keep controls and guardrails around the AI and its decision making so that we understand what the AI is gonna do. And then maybe you hit that button in the user experience and you tell the ai, yep, I agree with what you're going to do.

I'm gonna go run some script to fix some database problem. And you could say, yes, go off and do that. I agree with your steps.

So the way you wanna look at this, and the way we're looking at this as AI agents, as actual digital workers and orchestrate these digital workers just like you would your staff within your organization, you're gonna give these AI agents very focused responsibilities and jobs to do, and they will get better at that job over time with that learning capability I mentioned before. So it's that taking that next step. And I, I, I'll give you just a real quick example.

So I created an application about three weeks ago, and I had three AI agents that I created that were with me in my development environment, and we created this whole app over a four day period. And I did not write one line of code, I didn't write any code to, uh, generate, generate test cases, run those, uh, unit tests. I didn't write any code to deploy it to containers and into my containerized environment.

The AI agents did that all for me, and I was just orchestrating them just like I would an engineering team over my career. And it was fabulous experience. It really is pretty amazing to see yes, see AI work and doing so much of that for us.

You know, I think in my own experiences, Anthony, of code bases where code had part of it hasn't been touched for a long time. Folks that may have been done that have long gone and everyone's reticent about even going into it and trying to make any changes, more or less understand it, think about the power, not just of freeing is up to do other tasks, but also going back and doing more with the code that we have, the things that we may have, you know, not wanted to touch for a long time. Yeah.

Uh, I'd love to hear your perspective about, uh, this isn't an all or nothing when we talk about agents or agentic ai, it doesn't turn everything over. You mentioned about, let me review your work, okay? Yes.

Go ahead and perform that task, talk about that. We talk about human in the loop, talk about that process of how we judge, we start to use AI more and more to do agentic processes, but we still have control over what's happening. Yeah, Yeah, yeah.

A absolutely, you know, it's based on the technology itself today and the state that we're in, even though we've made tremendous strides over the past two years, we're still in a very early stage. And, you know, ai, in every agentic system that you use today, you, you see the messages that come out that AI could make mistakes. You, you should double check what AI is telling you, et cetera.

And really that's what what it's all about. So even the example where I, I built this app over this four day period with AI agents. The AI agents didn't do anything without my approval.

So when it, if it, it generated a script or it generated a module or whatever I was directing it to do, I'm still a very seasoned engineer. I was a, you know, I sit there, I study the code that it generated or the action that the things that it wants to do in my environment, and I am given that approval to, to, to move ahead with that action. That's where we are today.

That's where we need to be in the, in the near future. So when we build AI agents and we look at use cases for our gentech system, that's what we're looking at to address is how do we have these agents that are working 24 by seven on some use case, but how do we connect that human back into the loop for that oversight that needs to to happen? You almost wanna look at the AI agents as a junior staff member, right?

They're gonna go off, they're gonna go do things and they're gonna say, okay, I wanna go off and do this. You, you and I today would, would, would, would, would review that, right? That, that, that young develop developer or that young ops person that's new in our space, we would, you know, review what they're proposing to do and that we would give them the green light to go do it would not, or, or we would have an exchange with them on how to make it better and improvement, improve it based on our experiences.

It's the same thing with AI agents today. You give the green light when you want the agent to go off and do that based on your satisfaction with what it wants to do, and then you could also then guide that agent on how to make better decisions over time. It's a real fascinating area that we're getting it into.

I'm telling you, Mitch, the use of agents in an agentic AI system is gonna completely change how we build software systems moving into the future. We are just scratching the surface on this. Usually, uh, you know, from, from A-A-B-M-C-A-E portfolio perspective, we deliver a lot of functions and features through our product experiences across the breadth of our, our portfolio.

Imagine a future where BMC software delivers solutions as a collection of agents that get deployed and then augment our existing solutions or provide additional insights to you, but it gets delivered as agents. These agents then could be reused amongst many different use cases, right? So, you know what, back in the day when we got into this whole mindset of reusable services, right?

And service order into architecture, it's like you wanna create the services to do specific jobs, and then we tie all these services together to build a system. It's the same mindset with these, with, with AI agents. It's like, we're gonna create this pool of AI agents, but then we're gonna start to mix and match these agents to come up with use cases that were unimaginable just a few short years ago.

You know, it, it has drastically changed our business already. We are, we create software totally different than we would've six, 12 months ago. It was really pretty amazing.

You know, you said the word guide guidance, I think is really kind of the human in the loop. It isn't just approval. It is, yes, I'm guiding where, where, where AI is going for me, what the agent is doing.

Now, there's some other things that I think you need to be, have in place, right? When we talk about trust in what the agent is doing, so you wanna know what it did and how it did it and what happened, what are those things that you need to have in place to be able to get that information, to have that oversight and insight to what's happening? Yeah.

Yeah. So, and, and, uh, we talked about, uh, having that transparency or that observability level in your, in your AI system. Well, that pushes right down into the AI agents, right?

An AI agent. First of all, when we create an a AI agent for specific goal or task to achieve, there are things that we do when we build the AI system to make sure we ground that agent to only do that type of work that it was set out to do. And there's different AI techniques that we, we use to do that.

But at, at the end of the day, that that agent itself is totally audit, um, uh, the auditable, traceable, all the observability, it's all there in, in the, in that individual agent that we're, we're creating. So all of our agents that we create part of our agent system inherit this. The other thing that we could do is, and mentioned that we can infuse into B-M-C-A-M-E assistant, uh, additional knowledge from the LLM.

Well, now with AI agents to keep them grounded and guard railed to do the job as they want, we can infuse into the AI system specific pieces of knowledge that is targeted for a particular agent in that system. So the agent could use will, we'll use a language model. The agent will use knowledge base information, whether that's provided by the customer or provided by us, targeted to that particular agent.

Everything to keep that agent grounded and or fenced to do just that task or, or or goal that it was designed to do. And again, totally auditable. Um, so you can get into like the admin consult for BMC Amy E assistant, you'll be able to explore all the agents that we have, and you'd be able to get to all that observability information for those agents in the same, in the same way you would forward an entire agent system.

Yeah, I like to think of it as we don't have to just have free range agents where they're operating on information and data that we provided. There's the concept of artifacts, right? This is, these are sort of the, the rules of the road.

Here's how we operate or what things we do when we use certain terms and our prompts. There's also, that becomes the con canonical knowledge that you're using to set those guardrails and processes. Talk about how you take, you know, we're learning how to use AI too while we're implementing agents and new technology, and that can become part of this long-term expertise and knowledge that we're creating through using ai.

100% I agree with, with that segment. Yes, absolutely. It's so transformative in what's gonna happen with AI agents.

It's the level of trust will come with AI agents over time, and we are, no, I just want say this, that we're nowhere near AI agents replacing a human being that it's really kinda disturbing to read that and when, when, when, when folks are talking about that we're, we're, we're so far away from that. When it come, when it comes to ai, the way we should be looking at AI agents is digital workers that make you and I better at our job day in and day out and make our solutions better day in and day out, deliver higher value to our customers day in and day out. That's the way we should be looking at AI agents as we move forward on this journey.

Very good. Thoughts on next steps for folks that wanna move into this more agent, agent type process, I'm sure it's not jumping into the deep end of the pool, right? There's a, there's a stepwise, uh, approach that you can take to take in on certain kinds of work, what kinds of risks you wanna try take on, et cetera.

Yeah. So of you, when you start looking at agents that you wanna build out for your agentic system, you know, start with agents in that advisory role, right? One of the things, and, and why I'm heading this way, one of the things that we learned is when you treat your AI system as a monolith, where it's trying to handle all kinds of requests coming across, so you, you, you, your, your, your portfolio, just like each one of every one of us, if we're overtaxed doing many, many different things, we're not really good at any one particular thing.

But with AI agents where you can give them specific language models to use specific knowledge, to use specific guidance to use on what their job and role is, all of a sudden the responses you're getting from the AI system are really accurate and really good. So even if you're getting into the realm of, let's just start with the advisor part of it, start breaking down your areas of advisement into agents, different agents, different advisor agents, and you'll see how much better those responses are from the AI system. When you can focus those a, those, those, uh, agents around its advisement responsibility, start there.

That, that's a really, really, really good starting point when you wanna start moving into the actionability part of that. And it doesn't have to be actionability as far as, you know, dealing with something tangible that puts your production environment at risk. When we talk about an AI agent taking an action, it could be simply observing a situation and sending Mitch sending you an email and just bringing something to your attention that's actionability, but it's just sending an email or sending a message to a, a team's channel or whatever it may be.

You could have those, those levels of actionability. It doesn't have to be things like, you know, IPLI think for or something, right? Start killing processes.

No, please don't Starting. Um, and things of that nature. No, we're not, we're not talking about that.

So advisement, observing a situation and maybe just sending emails and notifications to start the notifications that will start building again, the trust with agents and start building you down, it will start moving you down the road towards an tic AI system. And as the technology that we're using evolves over time, our agents will evolve with that technology. Our agents will learn and grow within the environment they're running.

Our agents will learn and grow with each other. 'cause we've been, right now, up to this point, we've been talking about individual agents doing a particular task. But in a true and agent system, this is where you could actually orchestrate more than one agent to work together as a workforce or a digital team to solve complex multi-step IT problems.

As an example, you would have multiple agents working together to solve problems. And then when you follow that with using open industry standards like the MCP, the model context protocol to connect agents to real time systems or agent to agent protocol to get agents to communicate with each other. We get to a point where my organization, I'm creating AI agents, you're creating AI agents, all of a sudden we get to a point where we can collaborate on use cases together.

You provide certain amount of agents, I provide a certain amount of agents, they work together and collaborate to solve problems. So I see us at BMC working very closely with our customers in this journey because our customers too will be rolling out agentic AI systems in their environment. And I can see us working with our customers and collaborating on use cases with our collective agents together.

So it's gonna open up lots of opportunity, open up lots of doors for options, uh, as we go forward. Very exciting. This is, this is the game changer.

This is the transformative part of AI in the enterprise. What a great time to be in technology. It really is.

It is exciting and fun And all on the frame, right? All on the mainframe, you know, it is. I I love the way that this series came together, Anthony, and I appreciate your, your insights and also folks like Eric odell and the team at BMC software that helped put this together, thinking about this process, this journey you go through of using it as an advisor, learning what you need to learn to move on to the next step, to use it as a partner, AI as a partner, and then into more the operator, the agent.

And I think, uh, that this is a really good series. I think that will help people along as they get to that place in their journey. Like, okay, I get what they were talking about.

I see now how I can do this. Maybe a little different than I was thinking before I, I used it. So, uh, thanks again.

We appreciate all of your knowledge and your insights, and we hope everybody has gotten a lot out of this series. Thank you for joining us. Everyone.

Thank you to BMC software for having the foresight to put this together and help all of us on this journey that we're on together. And that's, that's the important part. We're in this together.

So we'll see you next time, hopefully on another series, on another conversation about AI adoption and use in the mainframe environment. Take care, everybody.