Transcription
We're all using AI in our day-to-day lives, whether we do it intentionally or unintentionally. It goes way beyond creating a lot of the fun images that we've seen, where we can take our photos and do fun things with them, like show us as different characters. We can use AI really to make us more productive at home or even on our jobs.
A lot of things that we seen and used AI for, I know personally I've gone and looked at trying to plan a nice vacation for myself, maybe to some tropical location. Using an AI chat tool to say, hey, plan me a 12-day itinerary in this location. We've also used it for work, where I had to go and maybe start a presentation, collect my thoughts and at least give me that starting point to advance what I'm trying to do and accomplish that task. But AI has really one thing in common is trying to make us more productive and answer a lot of the questions and get to us to value much more quickly.
As many of you may or may not be aware, mainframe is absolutely mission critical to what we do in our everyday lives. Every time we go and buy a cup of coffee or go to a store and make a purchase, there's a very good chance that that transaction is back-ended by a mainframe. However, in going out and talking to a lot of our clients, we do face a lot of challenges running operations for the platform. First, we're asked to be doing more with less in terms of skills and staff that we have on our teams. Next, we want to be able to treat the mainframe as it is any other part of our infrastructure, whether it be part of the hybrid cloud. Or something that's running on-premises. And last but not least, we want to bring a new generation of mainframe professionals and accelerate the learning that they have.
We've seen a lot of cases where AI doesn't necessarily help us in our day-to-day lives with the results that we get. And mainframe is really no different. I've seen cases where we've had clients go out and ask questions about software that they're running and talking to our support teams. We've had cases opened up where we've used general GPT tools that have provided responses back to us that weren't necessarily accurate in helping to solve the issue that the client had. It gave them some information, but it wasn't always necessarily the right and most accurate information.
I've seen this in my case where I've gone out and issued a prompt out to a GPT tool that is based on a large language model. That large language model has a lot of detailed information, but that large language model isn't always necessarily keyed towards the mainframe. So the response that I get back looks like a mainframe answer, but may not be very specific to the use case that I was trying to solve. For example, I asked it a question about a very specific message for CICS. It came back and told me that there wasn't actually an error or a problem that occurred, but it wasn't the right one based on other documentation that we had. So in this case, it gave me a mainframe-centric answer, but wasn't always really what I was looking to get out of that information.
Because AI solutions don't necessarily give us the right answer right out of the box, we do wanna find ways that technically we can get answers that are more well-grounded in what we're trying to do. And this is really where retrieval-augmented generation comes in. What we can do with the large language model Is call out to a rag? Which does stand for retrieval augmented generation. And what the RAG does is help ground the large language model and maybe more relevant or more up-to-date information. The RAG is really built by bringing in and ingesting a lot of documentation that may be critical to the mainframe. It could be best practice documentation, it could be papers that have been published, anything that we want to do to ground the answers that we have from the large-language model. So that next one we go out and ask the same kind of user prompt, we might get back the more appropriate answer that we have for CICS and exactly what that error message meant, because we do have information that was detailed and critical to what that response was.
But not only can we ground the answers in pre-built information into the retrieval augmented generation, we can also have clients that can go out and add their own information on their own best practices. This is how they can personalize it to their environments and their teams. So now that we've solved the generative piece of it by augmenting the results out of the large language model with retrieval augmented generation based on best practice documentation and even client enabled documentation, we can go one step further and start adding agents and agentic AI into the picture here.
So next we can start to use agents that are on the system. To automate many tasks that we might have within our environment. So these might go out and do other system resources. We may have agents that call out to hybrid cloud services. They don't always have to be running on the mainframe. We can integrate a lot of different things to automate a lot the tasks that were doing. Maybe we want to open up a ticket within our service desk. Maybe we wanna get the status from some of our core monitors. There's a lot of different things that we can do within the agentic framework. We're just starting to really scratch the surface of what we might want to do. From an operation standpoint, maybe we want to go out and do a health check as part of our environment. Maybe we want go out and find different ways that we could optimize and ensure that our workloads are running as efficiently as possible.
Agentic AI and generative AI really come hand in hand. So that now when the user goes and enters a prompt, not only do we get the information that is well-rounded based on the rag, but we can also maybe get live updates of information coming from the various agents that are running on the system.
As we've shown, we can use AI in a variety of different things across our lives, whether it be for work, for pleasure, or managing operations on something like the mainframe. It's meant to increase productivity and make us better and faster at what we're trying to do. It helps us automate repetitive tasks. And if we look at it from an operations approach for the mainframe, we wanna make sure that we get accurate, trusted results back from an AI tool that we're using. We can do that through use of RAG or retrieval, augment and generation coupled with agentic AI and using agents to automate a lot of manual tasks for us. This will all make us better at managing operations for the Mainframe. Thank you.